Home

REV Performance Vehicle Instrumentation

1. i VISOS 34 o EL panne ES 5 LE kat DLC 1341 LK 45 DI F ing 3d NIVIVEDO 1 ZacONWONOX E 13d 00400XL Gala LOVE vd tawzvd 479801 KOMA vOdli2SOL oawovd STY LX nvax 34004 59 YagA z L1X3 22 11X3 H3 61 11X3 81 113 Da 91 11X3 6 Z1X3 02 21X3 IZZ zz Zz1x3 z z1x3 y IT Sc c1x3 92 2LX3 103 6103 01 11X3 bb bLXS zip 1 11X3 na 1 11X3 67 A 8 Computer Listings A APPENDIX A 8 Computer Listings AVR CAN Datasheet AT90CANI28 Datasheet Lotus 2001 Service Manuals UQM Powerphase CAN Communication Summary UQM Powerphase 75 Manual Breakout Board Eagle Files Amplifier Eagle Files Source Code for AVR CAN SPY Source Code for Lotus REV Source Code for Engine Sounds Source Code and other files for Neural Network Estimation 68
2. Qt a cross platform application and ui framework Nokia 2011 Online Available http qt nokia com products 16 J Pearce Electric vehicle telemetry School of Electrical and Electronic Engineering 2010 17 Main roads wa on twitter Main Roads Western Australia 2011 Online Available http twitter com perth traffic 18 R Hanna Incidence of pedestrian and bicyclist crashes by hybrid electric passenger vehi cles U S Department of Transport Tech Rep 2009 19 L Garay Vega J Pollard C Guthy and A Hastings Auditory detectability of hybrid elec tric vehicles by pedestrians who are blind 2011 20 R Wall Emerson K Naghshineh J Hapeman and W Wiener A pilot study of pedestri ans with visual impairments detecting traffic gaps and surges containing hybrid vehicles Transportation Research Part F Traffic Psychology and Behaviour 2010 21 H r 734 pedestrian safety enhancement act of 2009 Open Congress 2009 Online Available http www opencongress org bill 111 h734 show 22 Follow up report President obama signs pedestrian safety enhancement act into law CNET 2009 Online Available http reviews cnet com 8301 13746 7 20027830 48 html 23 Blind advocates disappointed in nissan e v sounds for pedestrians New York Times 2010 Online Available http wheels blogs nytimes com 2010 06 17 blind advocates disappointed in niss
3. Online Available http standards sae org j1939 21 201012 35 Introduction to sae j1939 Kvaser Advanced CAN Solutions 2011 Online Available http www kvaser com en about can higher layer protocols 36 html 36 Controller area network Bosch 2011 Online Available http www semiconductors bosch de en ipmodules can can asp 37 Devicenet Open DeviceNet Vendors Association 2011 Online Avail able http www odva org Home ODVATECHNOLOGIES DeviceNet tabid 66 Ing en US language en US Default aspx 38 Canopen CAN in Automation 2011 Online Available http www can cia org index php id canopen 56 REFERENCES REFERENCES 39 Nmea 2000 1 edition 2 0 National Marine Electronics Association 2011 Online Available http www nmea org content nmea standards nmea 2000 ed 20 asp 40 Leaf canbus decoding MyNissanLeaf com 2011 Online Available http www mynissanleaf com viewtopic php f 448 t 4131 41 Can protocol specification CAN in Automation 2011 Online Available http www can cia de index php id 164 42 Lotus Elise Service Manual 2001 Model Year Onward Lotus Cars LTD 2001 43 D Kingdom 2009 rev management and on board embedded systems School of Electrical and Electronic Engineering 2009 44 C Watts Electrical designs for the renewable energy vehicles within the rev project School of Electrical and Electronic Engineerin
4. There is one element of EVs that proves to be both a blessing and a curse and that is the amount of noise that they produce Electric and hybrid vehicles tend to produce far less noise than their ICE counterparts Whilst most proponents of EVs tend to use the reduced noise pollution as an argument for widespread adoption of EVs the reality is that this can be a negative element of EVs Recent studies done with hybrid vehicles have indicated that there is a significant risk factor associated with low vehicle noise A study by the National Highway Traffic Safety Administration within the U S Department of Transportation indicated that hybrid electric vehicles were two times more likely to hit cyclists and pedestrians at low speeds than conventional ICE vehicles 18 Similar studies have done to evaluated the risk to blind pedestrians and have found similarly that 14 3 ENGINE SOUNDS Engine Noise Distance 4 5m 80 75 70 e Sound E REV Lotus Intensity 65 dBA M Rev Getz 60 4 ag AM ii Standard Getz S BMW 55 4 10 20 30 40 50 60 70 Speed km h Figure 13 Comparison of Electric and Internal Combustion Engine Vehicle Noise at Different Speeds EVs pose a greater safety risk as they are difficult to audibly detect 19 20 This is a large enough concern that US congress made moves to legislate the minimum amount of noise vehicles are expected to make in 2009 21 Earlier in 2011
5. Furthermore this data is then passed through the neural network and the outputs are compared to the output of the training data From this comparisons can be drawn on how well the trained networks match the output data Two basic networks were tested with hidden layer composed with a 5 15 hidden layers These were tested on a standard feed forward network network structure in Figure 24 and an Elman type recurrent network the results of which are illustrated in Figure 28 In training the networks batch learning was applied with a training time of 30 epochs before termination and every set of data in the training set was utilised Initial weights are randomly generated between 2 and 2 Feed Forward Hidden 5 Feed Forward Hidden 10 Feed Forward Hidden 15 1 7 o io IO on nora m uco mo o EI a 8 9 2 10 0 0 a E 4 a a L D 2 w 0 0 a 2 a a remis Trend Trermis Elman Hidden 5 Elman Hidden 10 Elman Hidden 15 none um o EJ ura pa mur mo o mm oco mo nm 3 S m 9 TO o n n o Figure 28 ANN Testing with Full Training Set Variable Hidden Nodes and Network Structure 46 5 2 Design amp Evaluation 5 BATTERY ESTIMATION From the results we can determine that the network structure and number of hidden nodes has little effect on the overall fitting A second set of tests was done this time only training the neural network on data points
6. maximum and average cell voltage in the last BMS data cycle Included is also an extra feature that constructs a data file that can be used for training offline neural networks a feature discussed later in this paper 2 2 3 Enhanced Logging Functionality Whilst the current logging functionality produces incredibly detailed files for the purposes of eval uating the car it does not log data from the BMS or GPS in the format that it is passed to the core program Having such functionality would be useful as the car PC software could be more rigorously tested by feeding it real world data As such two new logging functions have been added log raw GPS and log raw BMS which create files based on incoming GPS BMS data 10 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN Status Bar Environmental Theme Transparent Frames Stylised Buttons Figure 8 GUI changes 2011 This new feature allows a debug version of the software running on a separate PC to pass in data obtained in the field which in effect emulates the drive the data was obtained from This is useful for modules such as Engine Sounds which produce data based on BMS data as tuning the system to play the correct sounds for different actions is impossible to do in the online system 2 2 4 GPS BMS Core Modifications The GPS and BMS classes initially exhibited poor performance in situations where their serial connections would fail This was due to a desig
7. CAN Breakout Board AVR 2 x 34 Pin IDC Breakout Board AVR Amplifier AVR CAN PC AVR CAN Tachometer 12v Supply Lines Stripped to Stripped DE 9 to DE 9 Stripped to Stripped Stripped to Stripped Table VI Connector specifications in system 30 4 2 Design 4 CAN BUS SPY Initialise Parameters Receive CAN Signal Translate CAN signalto RPM Adjust PWM Output on pin Interrupt Routine Timer up Send RPM and other data over serialline Figure 19 CAN SPY Program Flow 4 2 2 Software The program design is quite simple and is done entirely in C It must be compiled with Atmel Studio Version 4 as the newer version 3 does not currently support all Atmel chips of which the AT90CANI28 is one A suitable CAN Library was found although it was used for the AT DKV90CANI development board from Atmel and not Olimex As such there are subtle differ ences in the library although most code unessential to the boards functions have been removed From here the program design is quite simple and illustrated in Figure 19 The CAN line on the Lotus is to be run at 500 kbit s This will result in a message received in 0 1 milliseconds 2 microseconds bit 50 bits per CAN frame By far the bottleneck of the system is the serial line 10 kbit s results in a time of 100 microseconds to transmit one bit As such tuning the timing of messages so that the tachometer can reasonably respond to changes whils
8. Layer Weight Adjustment 5 Wiki Wir Aw Hidden Output Layer Weight Adjustment 6 Aui C1 Uis 7 Ak AW ke Ap where ais the learning rate Figure 26 Neural Network Back Propagation Equations for a Three Layer Network Structure The most difficult aspect of ANNs is training It is difficult to ensure that the training data supplied to an ANN is going to be appropriate to approximate a particular function The problems of overtraining and undertraining are relevant to this discussion As we are only looking to estimate a curve and not solve a characterisation problem overtraining of the ANN is unlikely to affect our estimation Undertraining is a huge concern The largest set of any training data produced by the 41 5 1 Background 5 BATTERY ESTIMATION REV Lotus is going to be the battery placed under a 30 ampere load or less From the battery dynamic figure presented earlier this is not a load that is particularly good for estimation As such it is important to ensure that training is weighted towards situations where the battery is under an increased load lest it tend to favour data produced under low load conditions In addition it should be obvious that the closer the relation of the input data to the output data the more accurate the neural network will be It is important to note that although the ANN can approximates non linear function it does not infer that the internal structure and weights have an
9. REV Instrumentation amp Document Structure 1 INTRODUCTION 1888 First electric speedometer invented by Croatian scientist Josip Belusic 6 1901 Speedometer first featured in Oldsmobile 7 1925 Combination Speedometer Odometer invented by Arthur Warner 8 1940 Padded Dashboards advocated by car safety pioneer Claire L Straith 9 1959 Volvo becomes first manufacturer to include front lap shoulder seat belts as standard 7 1969 Volkswagon includes first on board computer system for fuel injection 10 1975 Real time fuel injection systems become common 1980 ALDL created by General Motors as first diagnostic system for testing Engines on assembly line 11 1985 First commercially available Sat Nav produced by Steve Lebbezoo 1988 SAE recommend a standardised connector for diagnostics 12 1990 GPS based satellite navigation begins to appear in market 1991 California Air Resources Board requires new vehicles sold in California to require ODB I 12 1994 ODB II specification required by CARB 12 1995 Airbags become a common feature 1996 ODB II required now required for all new cars sold in the United States 12 2008 New cars sold in United States required to implement ISO 15765 4 CAN standard 12 Table I Brief History of Automotive Advances 1 Improve increase and communicate the information collected from the car in real time in an effort to inform the driver of various conditions to ai
10. This means that smaller chips such as the ATiny series of Atmel controllers could be used to drastically reduce the 32 4 3 Evaluation 4 CAN BUS SPY Chip Purpose Price 1 Price 25 Notes AT90CANI28 uC CAN Controller 10 52 9 38 Requires Transceiver Has VO amp PWM AT90CAN32 uC CAN Controller 9 41 5 90 Same as above SJA1000T CAN Controller 7 68 6 20 Requires uC Separate transceiver optional TJA1054 CAN Transceiver 3 40 2 73 Bare bones transceiver w Fault Tolerant Mode MCP2551 CAN Transceiver 1 50 1 15 Bare bones transceiver w High Speed Mode Table VII Costs of Various Chips Source Mouser footprint of designs and the number of unused pins Unlike the AT9OCAN128 it does not require an external transceiver although it can make use of one if the engineer so wishes Finally a discussion on alternate design would not be complete without discussing the role of the CAN transceiver The role of transceiver is to convert the digital UO of the CAN controller to that of the system used by the bus as defined in the CAN standard Whilst some chips like SJA1000T include in built transceiver functionality it is more common to use a separate transceiver There are several transceivers available but the most common are the TJA1053 TJA1054 PCA82C252 and MCP2551 The choice of transceiver is usually not greatly important provided they conform to the standard and specification th
11. appropriate situ ations Currently most noise generation systems will switch off after they exceed certain speeds but this is the extent of adaptive noise making Directional sound equipment is being employed by ECTunes 27 whereby audio sounds are focused in the travelling direction of the vehicle so that those outside of the cars path are not disturbed the noise of the vehicle Similarly General Motors is looking into technology that can sense whether pedestrians are in the vicinity of the vehicle as opposed to other vehicles and therefore emit appropriate warning sounds The REV project s first attempt at an engine sound reproduction program began in 2009 and was produced by Chris Hellsten 28 This research primarily looked at generating intermediate sounds for a specific RPM from a set of engine sound samples as well as software methods of emulating engine noise The completion of this paper resulted in the Ferrari on a Stick program which formed the basis of all engine sound emulation work by the REV project Subsequently the system was further refined by Karri Harper Meredith in 2010 29 Whilst Hellsten was primarily concerned with the design of the software Harper Meredith s focus was primarily on implementation into the car As such extra code was added into the Ferrari on a Stick platform that interfaced it with the Labjack I O board and re badged as the Engine Audio Replication System E A R S However this version of th
12. gasoline powered cars that are almost as quiet as electric and hybrid vehicles yet are exempt from restrictions posed by the act Those that raise this argument state 15 3 1 Background 3 ENGINE SOUNDS that pedestrian safety is not a hybrid electric vehicle problem but a quiet car problem The following section presents background on solutions that have been undertaken in effort to manage the silent nature of EVs before building on and implementing the solutions developed in previous years 3 1 Background Since EVs have been criticised as a possible safety risk at low speeds a few different noise making solutions have been proposed and even implemented as is the case with a few commer cially available EVs Solutions tend to mainly take the form of two variants The first is to make the car emit certain warning sounds in particular high risk situations akin to the truck reversing sounds that most would be familiar with The second is to emulate the sounds associated with an ICE vehicle The latter is quite often the most popular solution as it also tends to feed the public s association of high performance with loud engine noise Emulation of vehicle noise particularly that of which is associated with engine revolutions is not a particularly new concept The video game industry has been working on technology to reproduce vehicle noise for the last two decades to produce realistic racing simulators This commonly involves th
13. older buffers results in a large echo effect and in use sounds like a jar of bees Similarly speed of input data to Engine Sounds is important if intermediate continuous data is not supplied to Engine Sounds an RPM jump from say 2000 to 2500 RPM will result in both 2000 and 2500 being played simultaneously As the fundamental frequencies in these are far enough apart it can result in a musical effect and sounds similar to a musical chord As a result of these two effects experimentation revealed that a smaller number of buffers of around 2 3 with a large volume cut on older buffers gave the best sound quality In addition to pop effects caused by changing sound buffers they can also be induced by dra matically changing the volume In order to reduce this effect volume changes are implemented by decaying the sound down to the target volume level Simply put the volume is modified from the source volume to the target volume in degrees of 10 until the current volume reaches the target volume This adequately reduces popping effects from volume changes Jitter is an interesting effect that only occurs in one particular scenario and is slightly related to the pop and response effects It is reasonably obvious when a track switch occurs if you are listen ing for it any other time it is difficult to distinguish However if the RPM is fluctuating between the track switching break point it becomes extremely obvious as the frequency shift is n
14. point x of the search space represents the state of some system and the function E x represents the internal energy of the system From this the goal is to reduce the function E x to it s minimum value As an example this value would be the mean square error or some other measure of fitness of the battery estimation function when applied to some testing set of data New function estimators are generated that neighbour the current function in the search space and if fitter than the current selection they replace it This can continue until it reaches some pre set error tolerance or allocated computation time is exhausted GAs take a different approach to finding the optimal solution in a search space A set of strings are used to encode a random set of solutions In every generation a random set of individuals are selected evaluated according to some marker of fitness and modified either combined or randomly mutated to form a new population This new population then undergoes the process again and it continues until a candidate meets the fitness criteria or computation time has expired Either of these methods can form part of a valuable contribution to battery state estimation Neural networks in particular can be sensitive to their starting weights Use of an SA or a GA approach could aide in producing a more accurate estimator by finding the optimal starting weights and or hidden layer nodes 64 65 5 1 6 Fuzzy Logic Fuzzy lo
15. taken form in the implementation of Engine Sounds a new method to determine battery capacity research and development of CAN bus equipment and updates the Lotus PC UI systems Engine Sounds was finally implemented into the vehicle after a year and a half of development of the code The final system emulates the noise of an internal combustion engine and will prove useful in future studies of pedestrian safety and driver acceptance The Engine Sounds system has been improved in order to reject as many unwanted side effects of audio discontinuity as possible and provide an accurate reproduction of ICE noise The UI system underwent many changes to improve the backend and frontend of the system Core changes to the telemetry BMS and GPS backend systems have improved stability of the system and enable each component to be able to recover in case of error The look and feel of the UI has been improved and should increase the appeal of using the system Logging functions have been expanded to enable the system to be emulated off line Finally the Twitter system was introduced to enable the the car to be able to receive traffic data from Main Roads CAN bus systems were reviewed in depth in order to create a better platform for attaching sensors in order to send and receive data better in the car This does the groundwork necessary for distributed I O in the car and also provides information necessary to build a CAN based motor controller The
16. that had a current member that exceeded 30 and then 40 amperes In doing so this removes low load current values that do not drastically affect the internal resistance of the battery regardless of the battery s discharge level However this does mean that estimations from the neural network do not occur as frequently A positive of this approach is that this significantly reduces the amount of training the neural network is required to do reducing the training set from the full amount of around 55 000 data points to around 3000 As a result the network can undergo far more training and therefore can create a better fit with less noise These results are displayed in graphical form in Figure 29 From these we can determine that using a reduced training set for estimation results in a better network albeit one that can only be used for larger drive currents This is still no ready for implementation into the vehicle and will require the use of a Kalman filter to improve the estimation i 30 i 40 T Percentage of Charge Remairirg Percentage of Change Remairirg Percentage of Charge Remainirg Figure 29 Neural Network Output when Ignoring Current Values Below a Threshold Value 5 2 2 Kalman Filter As the neural network performs poorly on it s own another method is required to ensure a more accurate estimation The method chosen to do this is the implementation of Kalman filter In doing this it is
17. 24 89 N 8151 68 W 1 05 1 5 280 2 M 34 M 1 2 75 Purpose GPS System Fix Data GPGGA Identifier 170834 Time Stamp UTC hhmmss 4124 89 Latitude N North South 8151 68 Longitude W East West 1 GPS Quality Index 0 Invalid 1 2 GPD Fix 05 Number of Satellites 1 5 Horizontal Dilution of Precision 280 2 Antenna Altitude Above Sea Level M Units of Antenna Altitude 34 Geoidal Separation M Units of Geoidal Separation 1 Age of Differential GPS data seconds 2 Station Reference ID 70 Checksum 62 A 3 Telemetry Protocol Source John Pearce A 3 Telemetry Protocol Source John Pearce Detailed here is the Telemetry Protocol as designed by John Pearce Minimum length of message is 33 bytes Field Bytes Comments Protocol ID 1 R for REV Vehicles Packet Length 2 Length of whole message IMEI 7 3G GPRS Device IMEI Latitude 4 Longitude 4 GPS Time 4 Julian Time Speed km h Heading 1 Heading divided by 2 Altitude 1 Metres divided by 20 Reason Code 2 Always 0 time interval for Lotus Status Code 1 Always 0 for Lotus DI Count Number of Digital Inputs Current 0 for Lotus DI 1 DI Count 8 1 8 Digital Inputs Per Byte ADC Count 2 Number of Analogue Inputs ADC 2 ADC Count Each Analogue Input is Scaled to 2 Bytes Battery Level 1 Checksum 2 Modbus CRC 16 63 A APPENDIX AA CAN SPY Breakout
18. 30 40 50 60 70 80 90 100 Time minutes Figure 34 ANN amp Kalman Filter System Convergence Tests 50 5 2 Design amp Evaluation 5 BATTERY ESTIMATION b sampling time hr 100 Capacity Amp Hr Td time delay equal to sampling time vi If drive current gt 40 J Output NN Output NN_noise Output Model Output Model noise State of Charge Estimate Figure 35 Block Diagram of Full ANN amp Kalman Filter System 51 5 3 Summary 6 CONCLUSIONS 5 3 Summary The section detailed various methods used recently in battery estimation problems and then pro ceeded to implement a neural network Kalman filter and coulomb counting combined approach Various designs were evaluated including modifying the the neural network structure and reduc ing the set of data used for neural network estimation The Kalman filter was modified by trialling different values for estimate measurement noise in order to gain the most accurate estimate for SoC The result is a SoC estimator than can produce estimates within 10 of the actual value This estimator still needs to be evaluated in the vehicle and ideally needs additional training data to span the entire capacity of the battery 6 Conclusions This concludes the work done to further develop the systems for the Lotus in 2011 Various systems have been implemented and updated in order to improve the Lotus and prepare it for subsequent studies to be undertaken Improvements have
19. 8 chip and more specifically the AVR CAN development board from Olimex It should be noted that whilst the one off costs of a development board are acceptable use in a production environment is prohibitively expensive In the case of building the CANSPY it was an extremely fast way to develop the func tionality required by the project While the use of this particular development board is acceptable as a once off proof of concept it is not a sustainable design decision in the long term and more purpose built solutions should be used in future Such a solution could be to purchase ATOOCAN128 chips and produce purpose built boards This is a good idea provided that they utilise all the chips features well as at around 11 per chip it is quite expensive Downgraded versions of this chip exist most notably the AT9OCAN32 whose only difference is a reduction from 128 kB of flash memory to 32 KB at a reduced price There are of course other chips in use The SJA1000T from NXT Semiconducter originally Philips is a stand alone CAN controller This chip is found in many commercial CAN products including the PCICAN from Kvaser It is probably the most popular CAN controller on the market currently and has a low cost of approximately 7 per chip This chip unlike the AT90CANI28 does not include multiple I O channels and PWM so it must be adapted for use with another micro controller Though this adds to the cost it is a more versatile solution
20. Board A APPENDIX A 4 CAN SPY Breakout Board _0000000008000G GOO A 5 Tachometer Amplifier Board A APPENDIX A 5 Tachometer Amplifier Board 65 A APPENDIX A 6 Motor Controller Breakout Board Source Watts A 6 Motor Controller Breakout Board Source Watts 6 6 l f o i a X2 OV NM D D H amp 4 gt INPUTS DESIGN BY REV TEAM 66 A APPENDIX Olimex A 7 AVR CAN Schematic Source AVR CAN Schematic Source Olimex A 7 SCHEMATIC JaTumd WF ASPAWOO xeumo MMMWJ duu pi xaWn1o oroz 9 1HOIHA4O 9 Y Aen NVO HAV TL ing 9 99A 095 G P B zu AS AS 095 t 1v1S Moi 961 068 99 ZELT AS AS L HA OS e M OO 62109 mr P 100 xu LZA 9 Z1X3 e A Z HD ONE tha e zLDOG AN EIEE ka T aan 9 L1X3 SL 2 FULL Z EZ zx LL GH Y99 WiNpS Z ZAdZNOO ws YMd noza Niza 1nois NIS MIL Hou ba MEL Joel FELL KELL ZELL ve za zia3 teria bea 0 Z1x3 6z z1x3 82 713 42 71 I Z ATADO E e z vz St 21x3 9121 za erza be Da DEL 62 113 82 11X3 42 413 SLL SLL pZ ON ek SZINVOO6LY 5 3 43HAv REA y Z Sr s4demusoOv ES v d ouvoov EZ Wi tatoo E Us z4dzo0w E 138190 Ee oaao K BEE 3 SOdliWOXINVOXL Fc rad doi OMT WILD OdEINWLOXL Es DUR j
21. Fes R gt 100 Dre Curent gt 40 Amps T 1 T T T T T T T 105 uF 4 0 1 4 E i 1 085 E E Sos 4 E Es E E Ur 4 i 06 L L 1 1 L L L 1 L 0 10 2 E 4 EJ Ed L amp Ej 100 Figure 33 ANN amp Kalman Filter with Variable R Values 49 5 2 Design amp Evaluation 5 BATTERY ESTIMATION As shown in Figure 33 generally better estimators are formed when the values for the mea surement noise R are several magnitudes higher than that of the value being estimated Whilst SoC is treated as a value between 0 and 1 keeping the measurement noise matrix R at between 10 100 seems to result in the best estimate function Keeping the value for R high means that the system must attain several neural network estimates before it begins to affect the value In doing so the response of the estimator is reasonably smooth and outputs a reasonably close estimate of the actual SoC with roughly 10 error at worst Further testing is done to ensure that if the initial estimate is far below that of the actual SoC that the system will converge Figure 34 displays that this is indeed the case This result is appropriate for implementation into the vehicle A diagram of the final system is present in Figure 35 ANN amp Kalman Filter Convergence Test fom 0 T Percentage of Charge Remaining Time minutes ANN amp Kalman Filter Conergence Test fom 0 5 T T T T T T T T T e co ec Percentage of Charge Remaining 0 10 20
22. IHE UNIVERSITY OF WESTERN AUSTRALIA FINAL YEAR PROJECT 20142809 REV Performance Vehicle Instrumentation Author Supervisor Matthew TYLER Prof Thomas BRAUNL Thesis submitted as part of the B E degree in the School of Electrical Electronic and Computer Engineering University of Western Australia Abstract The REV Project is a multidisciplinary effort to design build and evaluate electric vehicles with the goal of demonstrating the viability of renewable energy vehicles for personal trans port The following project outlines changes and additions to the REV Lotus instrumentation systems for the 2011 period Changes have been made to the core GPS amp BMS programs allowing for more robust operation The telemetry module has been rewritten to cooperate with the now running electric vehicle trial A more informative battery monitoring panel has been added in line with a need for more detailed battery analysis The Engine Audio Replication System EARS has been reviewed and implemented into the user interface A new panel has been implemented using the twitter API for relaying traffic data from Main Roads A new CAN SPY device is introduced which lays groundwork for implementing CAN based sensors and actuators offering a more flexible and responsive platform for sending and receiving information across the vehicle This simultaneously offers an expansive and high speed platform for future work This first device is used to
23. PC The in car PC consists of a Dual Core Atom 1 6 GHz Intel processor with 2 GB of RAM running the Microsoft Windows XP operating system This is mounted inside a VoomPC 2 enclosure specified for work in automotive environments and powered by an automotive DC DC Car PC power supply Furthermore it is connected to a 7 resistive touch screen The BMS and GPS are then attached to the PC by way of serial to USB converters The GPS is QStarz GPS 818X GPS receiver and outputs data based on the NMEA protocol The BMS is a custom design by Ivan Neubronner and allows various details about the batteries to be sent to the in car PC For information about the message protocols used in these systems please see Appendices A1 and A2 2 20 System Updates 2011 The system as described by Walter follows in Figure 6 Additions to the REV System result in Figure 7 Following now is a discussion of changes made to the system excluding Engine Sounds and CAN bus which will be covered separately 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN User Interface GPX File CSV File Log File Be Figure 6 REV System 2010 Source Walter 2 2 1 Ul Design Changes The UI has been given a few small cosmetic changes in order to increase it s usability The addition of the status bar is one such element It displays relevant conditions such as BMS and GPS connectivity to be reported on every page of the Graphical User Interface GUI This i
24. President Obama signed the Pedestrian Safety Enhancement Act of 2010 requiring EVs to emit a certain amount of noise 22 Tests done by the REV team indicate that EVs do indeed produce less noise than their petrol counterparts Performed at the foreshore with a sound meter it has been confirmed that on average an EV produces 3 dB less noise than ICE vehicles travelling at the same speed which is roughly half the sound intensity This tends to hold true for speeds under 30km hr after which noise caused due to wind resistance and tyre road friction tend to take over As an aside any team member who has driven the Lotus from the undercover car park to the REV laboratory can attest to the fact that pedestrians tend to not notice the vehicle moving towards them if approaching from behind Although the problem with EV noise only exists for low speeds typically under 30km hr it is important to remember that this is still a speed that can cause injury Of course replicating engine noise is not without it s critics The U S Federation for the Blind whilst praising Nissan for it s pedestrian warning system has made it known they are displeased that driver can turn the system off 23 Similarly anti noise pollution proponents have argued that mandatory noise requirements will increase noise pollution in the environment 24 Lastly there has been criticism at the singling out of EVs in the Pedestrian Safety Enhancement Act of 2010 There exists several
25. a lot of time to calculate Ideally inputs to neural networks should be normalised A common goal is to ensure inputs are kept between 1 to 1 This ensures that when summed and input to an activation function that reasonable output is produced If the input values are too high it will generally cause the activation function to lock the neuron to a value that is close to either zero or one Essentially the neuron becomes always on or always off As such normalising inputs is extremely important ANNs can learn by three methods supervised unsupervised and reinforcement learning In the former input is fed to the neural network and the output compared to some expected data Unsupervised learning is more complex and relies on the machine to learn to categorise it s outputs by itself Reinforcement learning is more applicable in AI systems where input data is produced by the agent acting on it s environment For our purposes supervised learning is more appropriate 40 5 1 Background 5 BATTERY ESTIMATION Figure 25 Forward Propagation Mechanism at the Node as we are attempting to shape the network to approximate a certain response from a set of input data The learning mechanism is described mathematically in Figure 26 ex dy yy Output Error 2 F k Ee Output Layer Error Gradient 3 Yk E s Hidden Layer Error Gradient 4 UIT idden Layer Error Gradien J dy J Wiji Wij Awi Input Hidden
26. also expected that much of the noise will be removed from the filter Figures 30 and 31 47 5 2 Design amp Evaluation 5 BATTERY ESTIMATION illustrate the mathematical models used in the development of the filter NONE ir At E oe 8 Ly Tk 60 8 Pr P Q 9 Figure 30 Kalman Filter Prior Equations for State of Charge p K 10 t PR G En E Ky Ne 11 where Ny is the neural network estimate at time instant k 12 P I Ky P 13 Figure 31 Kalman Filter Posterior Equations for State of Charge Testing of the filter is initially done using a neural network trained with the full set of training data This results in the graph displayed in Figure 32 Comparing this side by side with the sole neural network method we can see that we have successfully removed most of the noise from the estimator The onus is then on matching the data correctly to actual SoC of the battery system Improving this estimation is dependent on tuning Q and R parameters such that the system associates low load conditions with a greater degree of uncertainty than high load conditions In effect ideally the Kalman filter should discard estimations if the load drive current is below a certain minimum However it is important to remember that neural network estimation is poor if used over the entire training set Instead then the neural network that is partially trained using only high load current sets is utilised When a low d
27. an e v sounds for pedestrians 24 Electric vehicle noise pollution NoiseOff The Coalition Against Noise Pollution 2011 Online Available http noiseoff org evs php 55 REFERENCES REFERENCES 25 T Tabata H Konet and T Kanuma Development of nissan approaching vehicle sound for pedestrians How to solve the trade off between quietness and pedestrian safty of the electric vehicles 2011 26 Halosonic noise management solutions Harman Automotive 2011 Online Available http www halosonic co uk 27 Ectunes adds sound to silent evs but only where and when you need it Engadget 2010 Online Available http www engadget com 2010 09 09 ectunes adds sound to silent evs but only where and when you ne 28 C Hellsten Ferrari on a stick a system for emulating engine sounds School of Electrical and Electronic Engineering 29 K Harper Meredith Ears and rev School of Electrical and Electronic Engineering 2010 30 F Ho Drive system design for lotus elise electric car School of Electrical and Electronic Engineering 2009 31 Australian design rules Department of Infrastructure and Transport 2011 Online Available http www infrastructure gov au roads motor design index aspx 32 UQM Motor Systems User Manual UQM Technologies Inc 2011 33 CAN Communications Summary UQM Technologies Inc 2011 34 J1939 data link layer SAE 2011
28. ance is also made The second step is the update state The measurement residual is found by subtracting the predicted next value from an actual measurement of the system output From here the covariance optimal Kalman gain and improved measurements are obtained There are two important values to be considered in the Kalman filter aside from the actual model and obtaining the relevant measurement These are the values for process covariance noise com monly denoted as Q and the measurement covariance noise commonly denoted as R A larger 42 5 1 Background 5 BATTERY ESTIMATION A i Measurement Update correction Time Update prediction 1 Compute the Kalman G ain 1 Project the state ahead E P HT HP HT A Rr AY Bu ia 2 Update the estimate via z 2 Project the error covariance ahead i Z ES K Ly Hi P AP _ A7 0 3 Update the error covariance P I K H P Initial estimates The outputs at k will be the input atk 0 for k 1 Figure 27 The Kalman Filter Source Bilgin s Blog value for either of these indicates less accuracy in the respective system As such using a correct ratio can weight the system towards trusting measurements more or less As it is often not known what the exact values for covariance noise are they commonly derived from experimentation and tuning The Kalman filter forms the final piece of the battery estimation algorithm and will be used alongside a neural networ
29. at Lotus started to drive its instrument clusters with CAN signals and did not meet the CAN standard until 2008 Amusingly this has led to a large market of post 2008 instrument clusters being retrofitted for 2005 era Lotus Because of this and by looking through the Lotus 2001 Service Manual 42 it can be deduced that the tachometer is likely to be a driven by a 12 volt signal emanating from a hall effect sensor on the crankshaft camshaft The Getz runs a similar setup that was investigated by Daniel Kingdom in 2009 43 As such we can use Table V in order to design a CAN to Analog Tachometer converter that Kingdom formulated after probing the Getz This is roughly equivalent to the following equation J hm ua 1 28 4 2 Design 4 CAN BUS SPY PWM Frequency Hz RPM On Cluster 0 0 32 1000 65 2000 102 3000 134 4000 167 5000 200 6000 232 7000 268 8000 Table V Frequency vs Tachometer Value 4 20 Design 4 2 1 Hardware The CAN signal is provided from the UQM Powerphase motor controller over the CAN L and CAN H lines 32 In 2009 Cameron Watts 44 designed a breakout board from the Amphenol 19 connector on the controller in order to easily access the motor control lines Fortunately the foresight was had to break out all the signals provided from the controller allowing easy access to these CAN lines From here on a compatible micro controller is required to receive the CAN signals and relay them
30. at the designer is employing As most high layer protocols implement the basic CAN standard just about any generic CAN transceiver will perform Some do have extra features such as low power and sleep modes which should be considered Table VII displays various costs of different chips as listed on Mouser 4 3 Evaluation Due to damage done to the REV Racer and it s subsequent absence for repairs testing and implementation could not be done directly on the car Instead individual elements were tested and confirmed to function as per Table VIII If all tests pass it is likely implementation into the Lotus will succeed It must be noted that the Lotus motor controller must be connected to a PC and be configured to send CAN signals using the supplied software that came with the UOM Powerphase 75 The final system is to be housed in a project box and installed in the Lotus 33 4 4 Summary 5 BATTERY ESTIMATION Serial Test with loop back to confirm that UART Library is oper ating correctly CAN Connected to CAN compatible vehicle output checked with serial to PC connection PWM Signal Program designed to output OHz to 350Hz increasing by 50Hz ever 3 seconds until upper limit Confirm with probe Tachometer Test frequencies based on Table V Table VIII CAN SPY Testing List 4 4 Summary This section outlined the basic workings of the CAN protocol providing a reference for future and a base for CAN based devel
31. ations ur EINEN RS Rama mes 2 2 5 Telemetry Core Modifications a Li oe ptr a A de e 2 2 6 Main Roads Twitter Panel 5 5429 and Ge aed Boe ed 273 VUL p CCELI 3 Engine Sounds SEE Le cru RC EP RC UNE 3 2 Design amp Implementation 22232 un de RE RR ee ee es 3 3 Evaluation amp System Tuning 22 2323 3 a a A m HII Ea C RET ETC T E cg HRE Raa AE Bag At Neg Ra Neg se Dore 4 CAN Bus SPY Zl Background o doa E no aes d n eina e IRL E Td CAN lt 71 d eroe Oe RO e axe Bee dor Sox e eec peers AN2 Tachometer ua ie S ae sla te S de d e EE A AE EE E A IN TA Gg le EE GENEE EENEG 4 2 3 Alternate Hardware Design 4 3 STT de ES eurer a ex e Ee ec qe A Scere ae eee 4A SUMMA A b won ien USER 0 SE SC Se EAE Ge E 10 10 11 11 13 14 14 16 17 20 22 5 Battery Estimation 34 Delt Background 32 29 uos aco A vt atm s uL COL PLUR ARES 34 5 1 1 Coulomb Counting 4 4699 x o oo once s 37 5 1 2 Electrochemical Modelling 38 5 1 3 Artificial Neural Networks a aree aes EOS 39 5 1 4 Kalman Filter 4 5 uou Re Reo ee d s 42 5 1 5 Genetic Algorithms amp Simulated Annealing 43 beet FUZZY Log acie Bak a tena ck IC E LR 44 5 2 Design amp Evaluation os e a not E ME 45 5 2 1 Artificial Neural Network 46 22 24 Ed eio ere a A E TL 47 Ke SUMMATY uu mure ect e ege De ot te cd de ave cae Deo Denk e deo eee ede S 52 6 Conclu
32. codenamed Rev Eco was the first conversion attempted by the REV team The conversion started and completed in 2008 and underwent significant alteration to install an electric drive motor battery packs and suspension modification to handle the extra weight of the batteries As the first conversion to be attempted by the groups a lot of the monitoring and Battery Management Systems BMS are drop in commercial models as opposed to later conversions which use entirely student designed systems The goal of the Getz conversion was to prove that electric vehicles are a sustainable and valid option for typical commuting distances Figure 1 REV Eco The Converted Hyundai Getz Source http therevproject com The logging and user interface UI systems on the Getz are handled entirely by an Eyebot M6 running a version of Linux designed for the Gumstix chip that powers the Eyebot This system has undergone several modifications over the last three years of the project The most recent update 1 1 1 The REV Project 1 INTRODUCTION Figure 2 REV Racer The Converted Lotus Elise Source http therevproject com was performed this year by Beau Trepp and resulted in a complete rewrite of the system to fix many significant stability and speed issues that existed in the original design The Lotus Elise conversion began in 2009 as the second conversion to be completed by the REV Team This two seater performance vehicle is a much more
33. d back propagating neural networks Selection of network structure is important Within Back Propagating BP networks there are two main structures feed forward and recurrent networks Feed forward networks are as the name suggest and simply feed data through to the output through a number of layers Recurrent networks include recurrent layers which are usually a connection to and from the hidden layer Such a connection allows the network to contain memory of previous inputs Adding a recurrent layer does add some extra computational cost to the system Some typical neural network structures are presented in Figure 24 The next element of network structure is the selection of an appropriate number of input nodes hidden layers hidden nodes and output nodes The selection of input nodes and output nodes is defined by the problem It has been proven that only one hidden layer is necessary to estimate non linear functions The only major design choice left is the number of nodes to use in the hidden layer There are many rules of thumb that have been created although many in the community disagree as whether they have any merit 61 Wanas recommends that the number of hidden nodes be selected based on the logarithm of the number of training samples 62 The only consistent opinion is that only through experimentation can one derive a reasonable number of nodes to use It is important to note that increasing the number of nodes increases t
34. d in driver decision making 2 Collect and collate data from sensory input to be used in evaluating the performance of the vehicle Any system added into the REV Lotus instrumentation systems should aid in completing either one or both of these goals The following sections of this document outline and evaluate systems developed for the Lotus and provide background for design decisions Whilst a lot of the systems are interrelated the document has been structured to review each system in it s own section as to ensure that there is significant detail and focus on each individual component and to guide future work in each area The document is ideally read in the order that sections appear as occasionally information will appear later that is dependent on a prior section As a secondary yet equally important goal it is the hope of the author that this document provide a reasonable of amount of information and background for future students that may be 2 OVERALL SYSTEM DESIGN required to work with the Lotus 2 Overall System Design 2 1 Background 2 1 1 Software Distance Since Charge H HH 73 Km Charge Remaining gt peed krr h Ac 30 Rena ning Distance rm 84 Time 14374 Figure 5 REV Lotus UI 2009 Source Varma 2010 Source Walter and 2011 The first attempt at unifying all aspects of the system was completed by Daksh Varma in 2009 13 This system was composed of Visual C backe
35. e Applications SAE J1939 Trucks Bus Fleet Management amp Passenger Vehicles system contain a priority Those transmitting a lower priory will detect if a higher priority is emit ting a message and will back off until the higher priority message has finished receiving O bits are dominant on the bus so if a node emits 1 but reads a 0 it will know that a higher priority message is being transmitted and go into the waiting state The object layer handles message filtering and message handling Under this layer messages a particular can node is not interested in are filtered out to reduce processing time Finally the messages are constructed in the object layer to be sent by the transfer layer The data link layer specifies the main workings for the CAN standard It provides the following attributes of the CAN bus e Multi master capability Any CAN node may send a message if the bus is idle e Broadcast communication All messages transmitted are received in all nodes All receiving nodes decide if they like to accept this message This guarantees data consistency as all nodes in the system use the same information e Sophisticated error detecting mechanisms and re transmission of faulty messages This guar antees network wide data consistency e Non destructive bus arbitration If two or more CAN nodes request simultaneously a mes sage transmission the protocol guarantees that the message with the highest prio
36. e batt volts is below the setpoint and when the fuse is open circuit ie 8A and above REV X9 EYEBOT JR CONTROLLER GENERATED Command Status Status O 1 1 90 Jcontrlleris ONLINE NONE E EE D Amps A o0 100 0 MEGATIVE CURRE NONE RUNNING 100 10 Amps A 13 10 0 POSITIVE CURREN NONE HARGING REGEN BRAKING 10 1 p D Lo o o 2 Ll Volts B 1 30 o PACKVOLTS NONE BATTERY PACK VOLTAGE 300 1 Volts B 2 1250 0 AUXVOLTS NONE 12 V BATTERY 12 5 100 Status c 0 o o0 CHARGING ON NONE Manual mode Charging Status C 1 o 0 CHARGING C1 NONE Manual mode Charging Condition 1 Status c 2 o 0 CHARGING C2 NONE Manual mode Charging Condition 2 eS ALE AAA A eee Min volts volts Status Reset 60 A 2 GPS Protocol Source NMEA A APPENDIX A 2 GPS Protocol Source NMEA NMEA 0183 is the GPS protocol most widely used and consists of a set of messages GPRMC 220516 A 5133 82 N 00042 24 W 173 8 231 8 130694 004 2 W 70 Purpose Recommended minimum specific GPS Transit data GPMRC Identifier 220516 Time Stamp UTC hhmmss A Validity A ok V invalid 5133 82 Current Latitude N North South 00042 24 Current Longitude W East West 173 8 Speed in Knots 231 8 True Course 130694 Date Stamp ddmmyy 004 2 Variation W East West 70 Checksum 61 A 2 GPS Protocol Source NMEA A APPENDIX GPGGA 170834 41
37. e computationally expensive for a computer or a micro controller to perform In the case of the Lotus which uses an 83 cell battery back this suddenly takes a lot of processing time It is also likely to be inaccurate as the Lotus only attains new data for a particular cell roughly every 15 to 20 seconds As such it is important 36 5 1 Background 5 BATTERY ESTIMATION that any estimation methods extends itself to being easily generalised to the entire pack SoC In the following subsections are discussions of various battery estimation methods Another good review of battery estimation methods is present in a series of papers by Gregory L Plett although he does not detail models based on learning algorithms 49 50 51 5 1 1 Coulomb Counting Coulomb counting is by far the most presently used technique is estimating capacity of battery cells that do not have a strong relationship between capacity and terminal voltage Battery ca pacities are typically rated in amp hours so the problem is well conditioned for calculating the depletion of the battery by monitoring the current being drawn from it However it does suffer from several before mentioned caveats Particularly it s dependence on knowledge of the previous state Additional drawbacks include inaccuracies in current sensor readings However these issues can be accounted for with more sophisticated implementations The most significant drawback with coulomb counting is it
38. e ee 27 LG SE AN Extended Frame 6 ae sl e sy e y ea e 0300 AMS ne a SX 27 17 OBDII Pinout Diagram Source http www pinoutnet 28 18 CAN SPY System DeSIgli oso e e dex e dex ee ee 30 19 CAN SEX Program PLOW e a Ta e e Sle RT 903 EI BOO Be aA H 31 20 Accurate Feedback CAN Message Source UQM CAN Communication Summary 32 2 Thundersky LiFePO Cell Source Thundersky Pty Ltd 35 22 LiFePO Discharge Characteristics Source Thundersky Pty Ltd 35 23 TBS eXpert Pro A Coulomb Counting State of Charge Indicator Source TBS Blectronics tiva edd ee eee awe Re quater s es Ta 38 24 Feed Forward and Elman Recurrent Neural Network Structure 40 25 Forward Propagation Mechanism at the Node 41 26 Neural Network Back Propagation Equations for a Three Layer Network Structure 41 27 The Kalman Filter Source Bilgin s Blog 43 28 ANN Testing with Full Training Set Variable Hidden Nodes and Network Structure 46 29 Neural Network Output when Ignoring Current Values Below a Threshold Value 47 30 Kalman Filter Prior Equations for State of Charge 48 31 Kalman Filter Posterior Equations for State of Charge 48 32 ANN amp Kalman Filter Testing with Full Training Set 49 LIST OF FIGURES LIST OF FIGURES 33 ANN amp Kalman Filter with Variable R Values 49 34 ANN amp Kalman Fil
39. e software is tightly bound to the Labjack system and is not ideal for implementation into the Lotus and is based on obtaining data from sensors that simply don t exist on the Lotus Further work is mentioned on an embedded version of the software but it does not appear that a physical implementation eventuated 3 2 Design amp Implementation There were two completed versions of engine sounds supplied to be implemented in the Lotus The first completed by Chris Hellsten and improved upon by Karri Harper Meredith was written in C in 2010 The second is a C implementation by Chris Hellsten that was written after he completed his thesis Both of these programs were originally developed upon the assumption that there would be access to either a foot pedal or tachometer and that the speed of the car would be known However this was not the case initially and it was eventually decided that the current would be a suitable method of approximating an appropriate RPM value Thus either program would need to be modified Each version has several advantages and disadvantages although ultimately the Cft version was selected The original C code has the advantage of being directly portable to the C system As such linking to the functions in E A R S is a trivial task It also includes an emulated gearbox system 17 3 2 Design amp Implementation 3 ENGINE SOUNDS which the C code did originally have Unfortunately it does have several d
40. e starting to be employed in EV s In terms of emissions plug in EVs do not produce any when in use Their footprint is instead calculated by the energy source used to recharge the vehicle s batteries The REV vehicles are powered by a solar grid and therefore are zero emission vehicles However vehicles that are charged off a typical mains grid in Australia would be supplied from burning coal This would likely still be cleaner then burning of petroleum fuel sources EVs are not new EVs were reasonably common during the early period of the 20th century particularly in cities where range limitations had few consequences Part of their appeal was the fact they did not require any manual effort to start at the time and thus they were commonly stigmatized as women s cars partially due to aggressive marketing as being more suitable for female drivers However with the invention of the starter motor by Charlie Kettering itself an adoption of technology used to automatically open cash registers made by NCR gasoline vehicles became far easier to start the construction of large highways and the discovery of plentiful oil in Texas Oklahoma and California caused EV technology to stall for almost a century EVs enjoyed a small resurgence in the early 1990 s as auto manufacturers invested in cleaner technology This was partly in response to the California Air Resources Board which pushed for lower emission vehicles This resulted in the controver
41. e use of dynometers and purpose built microphones for capturing engine vehicle noise In game systems commonly use a set of samples and then frequency shifting to create appropriate engine noise The most recent addition to the world of engine noise systems is that of Nissan s Vehicle Sounds for Pedestrians V S P 25 This takes the form of warning sounds as opposed to that of repli cating engine noise In this system the car produces a sweeping sine wave 600 Hz to 2 5 kHz The rationale behind this is that most age groups will be able to hear a noise in this range and be able to respond to the vehicle appropriately This system can be shut off by the driver or shuts off automatically once speed increases beyond 20km hr V S P is currently employed on the Nissan LEAF The HaloSonic system is a noise making solution produced by a joint venture between Lotus Engineering and Harman automotive 26 Unlike Nissan s solution this is a representative of the engine noise replication camp This system is currently configured to produce the sounds associated with V6 and V12 engines Whilst this system has been around since 2009 it 1s still in development and has only demonstrated in a Toyota Prius and the Lotus Evora 414E concept car It is currently unknown whether the HaloSonic system will be bought to market 16 3 2 Design amp Implementation 3 ENGINE SOUNDS There is currently a large focus on producing systems that only create noise in
42. ed the Electric Car Sony Pictures Classics 2006 6 D Zubrinic History of croatian science 1995 Online Available http www croatianhistory net etf et22a1 html 7 Milestones in automotive history AAT Car Auto Diagnosis amp Auto Repair Help 2011 Online Available http www aalcar com library timeline htm 8 History of warner electric Warner Electric 2011 Online Available http www warnernet com history asp 9 America on the move National Museum of American History 2011 Online Available http americanhistory si edu onthemove themes story_86_10 html 10 Cracking ferrari s engima code Ferarri Online Article appeared in Sports Car Market January 2011 2011 Online Available http www ferraris online com pages article php reqart SCM_201101_SS 54 REFERENCES REFERENCES 11 T Boynton General motors computerized vehicle control systems A short history 2011 Online Available http tomboynton com GMnetworks pdf 12 Eobd a detailed history Omitech 2011 Online Available http www omitec com en support technology briefs detailed history of eobd 13 D Varma Renewable energy vehicle instrumentation Graphical user interface and black box School of Electrical and Electronic Engineering 2009 14 T Walter Development of a user interface for electric cars School of Electrical and Elec tronic Engineering 2010 15
43. ether the car is exceeding a certain minimum and maximum RPM threshold This appears to be partly based on transmission information obtained by Frans Ho 30 This does create several new issues The speed of the car is not always available due to the intermittent nature of GPS and there is no throttle sensor on the car A cursory glance at the gearbox code reveals that it s RPM tracks are essentially based on speed so the value for current is instead used to indicate throttle in this scenario thereby avoiding the need for current RPM conversion The speed is a much larger problem as if the GPS cuts out it will result in the car sounding like it is stuck in one gear The solution for this is not perfect at the moment At present we can obtain the motor RPM from the CAN bus and as gear ratios are known we could use these to get an estimate of speed to operate independently of the GPS However there is no sensor 20 3 3 Evaluation amp System Tuning 3 ENGINE SOUNDS currently installed in the car to obtain what gear the vehicle is in nor is it particularly necessary as the car is driven solely in first gear around 90 of the time So a reasonable estimate for speed can be made solely from the RPM value of the motor delivered by the CAN bus line and as such can act as a fail safe in the event the GPS is off line As sensory data is not available currently for a perfect implementation of the Geared Engine Sounds two switch able modes a
44. field and has remained almost unchanged for most of the last half century Car dashboards were originally modelled after aircraft panels and provided information regarding speed temperature battery and fluid levels in the car The first implementations of such sensors were mechanical in nature but with the advent of modern electronics have mostly been upgraded to electrical sensors As technology advances complex communication systems and digital display panels have become more ubiquitous Below in Table Iis a brief time line of advances in vehicle instrumentation CAN communication networks have been arguably the largest advance in vehicle instrumen tation Introduced in 1986 the CAN system was proposed as a solution to the ever increasing amount of electronics being produced in the modern car The system reduces the wiring of the ve hicle making it easier to add new electronic systems to the car In it s inception the CAN bus was mainly used as an output of the on board diagnostic OBD connector and provided information on faults from the Engine Control Unit ECU However in modern vehicles it is used for any number of functions including driving instrument panels and controlling motors It is not uncommon for vehicles to have multiple CAN buses controlling different elements of the vehicle 1 4 REV Instrumentation amp Document Structure There are essentially two main goals of instrumentation with regards to the REV Project 1 4
45. first CAN system in the car was designed utilising an automotive motor controller 52 7 FUTURE WORK and is able to pass data from the CAN bus to the PC in addition to driving a standard analogue tachometer A review of battery estimation methods was undertaken culminating in the implementation of a new estimation system for state of charge The shortcomings of various battery techniques were discussed as well as the issues with implementing a state of charge estimator in the lotus An ANN coulomb count Kalman filter approach was designed and evaluated resulting in an estimator that can estimate state of charge within 10 of the actual value Several improvements were made to the Lotus in 2011 to differing elements of the car This creates a solid foundation for branching out into separate sub systems of the car and provides much potential for focused development of different areas It is the hope of the author that a solid guide to the REV Lotus systems has been provided both for the present and for the future 7 Future Work There are several future developments that could be undertaken to improve the current system e Development of a map server with routing engine for supplying navigation data to the REV Vehicles e Improve the graphics used in the Lotus user interface and implement new classes for con structing user interface elements e Development of touch screen hardware with embedded CAN controller e Developme
46. g 2009 45 AVR CAN Development Board Users Manual Olimex Ltd 2011 46 A Padhi K Nanjundaswamy and J Goodenough Phospho olivines as positive electrode materials for rechargeable lithium batteries Journal of the Electrochemical Society vol 144 p 1188 1997 47 Tesla model s The battery pack CNET 2010 Online Available http reviews cnet com 8301 13746 7 20018836 48 html 48 Panasonic launches type 18650 li ion batteries with a peak capacity of 3 1 as Panasonic 2010 Online Available https industrial panasonic com eu news nr2010051E002 nr201005IE002 Press Release Li Ion_NNP_E pdf 49 G Plett Extended kalman filtering for battery management systems of lipb based hev bat tery packs Part 1 background Journal of Power sources vol 134 no 2 pp 252 261 2004 50 Extended kalman filtering for battery management systems of lipb based hev battery packs Part 2 modeling and identification Journal of power sources vol 134 no 2 pp 262 276 2004 51 Extended kalman filtering for battery management systems of lipb based hev battery packs part 3 state and parameter estimation Journal of Power Sources vol 134 no 2 pp 277 292 2004 57 REFERENCES REFERENCES 52 53 54 55 56 57 58 keed 59 60 61 62 63 K Ng C Moo Y Chen and Y Hsieh Enhanced coulomb counting method for e
47. gic is the final method to be discussed Fuzzy logic is a form of many valued logic and can be used to define logic values between 0 and 1 The fuzzy system has been proved to be a universal approximator and is capable of fitting any non linear function The membership functions of the set need to be generated by an expert or some other system This could be derived from battery characteristic models or from an unsupervised neural network In operation there is a mapping between the inputs to some fuzzy set which is later transformed to give a crisp point that is more certain of the measurement being made Popular models for fuzzy logic are often defined from impedance measurements and frequency response of the target cell Whilst research on using fuzzy systems alone in state of charge deter mination is available more recent research tends to focus on adapting it with other methods such 44 5 2 Design amp Evaluation 5 BATTERY ESTIMATION as neural networks to form neural fuzzy networks More recent research focuses on using fuzzy logic for BMS functions in addition SoC estimation in order to maintain health of the battery pack 66 67 5 2 Design amp Evaluation Many different networks were constructed trained and evaluated in creating the final estimator Initially an entirely neural network approach was intended based on previous work that was avail able in the literature However these papers had access to more t
48. he time it takes to forward and back propagate exponentially so if execution time is a major concern the hidden node count should be kept as low as feasibly possible The choice of activation function is also important in defining an neural network It is required that the activation function used is differentiable Furthermore it is required that the activation function at the hidden layer is non linear as this is what enables the neural network to act as a 39 5 1 Background 5 BATTERY ESTIMATION Hidden Layer Hidden Layer Input Layer Output Layer Output Layer Context Layer a Feed Forward Neural Network b Elman Recurrent Neural Network Context layer retains memory of hidden layer nodes Figure 24 Feed Forward and Elman Recurrent Neural Network Structure non linear estimator The four most popular functions are Sigmoid Gaussian and Radial Bias functions Functions selected at the input and output layers are usually chosen to mimic the data that should be input and output To use the battery estimation problem as an example the sigmoid function would be a good choice because output is clamped between 0 and 1 i e output is re stricted between O and 100 Usually this does not have too much effect on the overall function approximation The activation function at the hidden layer is far more important Sigmoid and tansig functions are popular as they are easily differentiable by computers and thus do not take
49. he transmission line The below Table IV illustrates the recommended transmission length lines max limit for a particular transmission speed Figure 15 is of a typical high speed CAN network physical layer There are two main disadvantages when dealing with the CAN bus The first is that of bus utilisation It is possible for higher priority messages to hog the CAN bus This can can occur if a high priority node is constantly transmitting This can be the case for motor control systems To get around this problem most vehicles will have more than one CAN bus with one usually dedicated for ECU signals as is the case with the Nissan LEAF 40 The second is that of a limit on the number of CAN nodes due to the address limit in the specification Currently most CAN networks are limited to 110 to 255 nodes This can be gotten around by interconnecting CAN networks however in practice it is rare that this situation occurs in automotive applications CAN Message frames are specified in CAN 2 0 Part A and CAN 2 0 Part B 41 The former describes the base CAN format whilst the latter describes both base and extended formats In order for CAN compatibility a message must match a description in either part A or part B Figure 16 describes the format of an extended frame CAN Message In operation typically only the Identifier A Identifier B and Data are specified AII other fields are constructed by software and hardware routines These include the S
50. hey are reasonably low cost non toxic and exhibit high thermal stability A flat voltage charge ratio essentially means that these cells can deliver close to constant power for reasonably low C rates which is ideal for EVs A graph displaying discharge capacity vs terminal voltage for differing C rates is displayed in Figure 22 There are some drawbacks with LiFePO over conventional lithium cells They have a lower energy density 34 5 1 Background 5 BATTERY ESTIMATION Figure 21 Thundersky LiFePO Cell Source Thundersky Pty Ltd and discharge rates are also reduced In both cases this is usually solved by using a larger battery but research is being undertaken to improve both aspects These advantages are also slightly off set by LiFePO s longer life span in contrast to conventional lithium ion cells LiFePO cells are currently the most popular cells used in hobby electric vehicle conversions and in a few commer cial vehicles Whilst Tesla employed LiFePO cells in the Roadster it is speculated that the new model S uses a LiNiO battery pack jointly developed between Tesla and Panasonic that offers an improved capacity based on Panasonic s 18650 cell technology 47 48 Voltage V um PLY PSE R JR 09 E ME 19 TE B 85 LYF battery s discharge curve under normal temperature 20 40 60 80 100 120 Discharge Capacity Figure 22 LiFePO Discharge Characteristics Source Thundersky Pty Ltd 35 5 1 Backgrou
51. ing neural networks and ekf Industrial Electronics IEEE Transactions on vol 57 no 12 pp 4178 4187 2010 K Hornik Maxwell and H White Multilayer feedforward networks are universal approxi mators Neural networks vol 2 no 5 pp 359 366 1989 How many hidden units should i use comp ai neural nets 2011 Online Available http www faqs org faqs ai faq neural nets part3 section 10 html N Wanas G Auda M Kamel and F Karray On the optimal number of hidden nodes in a neural network in Electrical and Computer Engineering 1998 IEEE Canadian Conference on vol 2 IEEE 1998 pp 918 921 R Kalman er al A new approach to linear filtering and prediction problems Journal of basic Engineering vol 82 no 1 pp 35 45 1960 58 A APPENDIX 64 Y Zhou J Sun and X Wang Power battery charging state of charge prediction based on genetic neural network in Information Engineering and Computer Science ICIECS 2010 2nd International Conference on IEEE pp 14 65 Y Lee W Wang and T Kuo Soft computing for battery state of charge bsoc estimation in battery string systems Industrial Electronics IEEE Transactions on vol 55 no 1 pp 229 239 2008 66 A Salkind C Fennie P Singh T Atwater and D Reisner Determination of state of charge and state of health of batteries by fuzzy logic methodology Journal of Power Sources vol 80
52. ions and emulate an ICE vehicle as closely as possible To realise the consequences of this is to know that EVs often use fewer gears than ICE vehicles and as such RPM is often much higher However running the car at the reported RPM value of the electric motor will get annoying as for a speed of around 50 km hr would result in a sound of around 5000 RPM This is much larger than the cruising RPM of most ICE vehicles but wouldn t be altogether unusual in an EV Current to RPM translation is reasonably simple However to understand it requires an under standing of how much current the car passes in certain scenarios Firstly although the motor is technically designed to pass 400 amps the most it is configured to draw currently is 200 amps This is done by putting your foot to the floor During a regular suburban drive not exceeding 60km hr and accelerating reasonably the driver will be unlikely to exceed 70 amps for heavy ac celeration 30 amps is passed when driving at around 50km hr As such we could assume that at 30 amps we would playback at 1800 RPM and linearly increase sounds to 7000 RPM at 200 amps This works fairly well but is strange in practice because gear shifts are not emulated which was the next task Fortunately a gearbox emulation design was present in the C version that was easily portable to the C version This code adjusts the RPM value that engine sounds should play based on the speed of the vehicle throttle and wh
53. k and coulomb counting to form an estimate for SoC This is not unusual as the Kalman filter in one form or another is implemented in many battery estimation algorithms Although not used in the final method a review of battery estimation techniques would not be completed without briefly discussing genetic algorithms simulated annealing and fuzzy logic 5 1 5 Genetic Algorithms amp Simulated Annealing Genetic Algorithms GA and Simulated Annealing SA are two methods that are employed to find optimum solutions in a search space though both approach this in a different manner GA operates similarly to evolution and normally utilises strings which act as an analogue to DNA that represent attributes of the estimation function and uses mutations crossover and inheritance to formulate optimal solutions SA is based on the metallurgic concept of annealing whereby metals are heated to cause atoms to be come unstuck and are then cooled so they settle in an optimum position The main point of difference between the two methods is that at any one time SA only has 43 5 1 Background 5 BATTERY ESTIMATION memory of discarded solutions and the most optimum solution where as GAs maintain a pool of solutions from which to spawn improved estimators Both suffer from similar problems the most important of which is the threat of the optimising on local minima and getting stuck although there are methods to reduce this In SA each
54. n error in the original code that assumed that a connection could never be closed by their connected module This disconnect can happen due to driver errors but is most commonly exhibited in situations where the hardware undergoes a reset or power cycle This is a particularly common occurrence with the BMS as it will often reset itself after a successful charge cycle and occasionally when the vehicle is plugged or unplugged from an outlet As such the programs were redesigned to attempt to reconnect upon a disconnection The old logic compared with the new logic is displayed in Figure 10 2 2 5 Telemetry Core Modifications The original telemetry module was included in the 2010 version of the software but has under gone some simplification and bug fixing to get it in a working state 11 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN Estimator Value Figure 9 Screenshot of battery cells page Old Structure Initiate Class Conned to BMS Connected ii Start Update Timer Poll Serial P ort Parse New Messages Update Data Structure des New Structure Initiate Class Start Update Timer Connect to BMS Conneded No Poll Serial Port Parse New Messages Update Data Structure Figure 10 Comparison of BMS GPS logic 12 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN Originally John Pearce based the protocol on that used by the AT telemetry units 16 A fe
55. nd 5 BATTERY ESTIMATION In estimating the capacity of a battery cell there are a few main elements to be considered The first is of State of Charge SoC Anyone who has used a car with a fuel gauge will be familiar with this measurement which is essentially an estimate of the remaining that can be used to drive the vehicle with respect to a reference point of 100 capacity The second is State of Health SoH SoH is typically calculated as a ratio of total capacity of the pack currently compared to what it was when the pack was new Battery packs tend to lose some charge as they age so this is essentially what SoH is measuring This section s main focus is on determining SoC and so SoH is mostly ignored In using a battery pack as opposed to just a single cell we suddenly introduce a third element of capacity pack balance Balance is essentially a measure of how close all terminal voltages in the pack are to the mean voltage In a perfectly balanced pack not only is terminal voltage equal across every cell but so is the remaining capacity of each battery Pack balancing is the responsibility of the BMS and is by far it s most important aspect It does not matter if 99 of the cells in a pack are currently at 100 of their rated capacity if even one cell is at 0 the vehicle cannot be driven or else that cell will be damaged and require a replacement In summary this means that a battery pack s usable capacity is limited by the capaci
56. nd and an Adobe Flash front end Data was passed between the two programs by writing to XML files This was largely a rather inflexible platform and a new system began development just six months later This system implemented the core functionalities that would be present in each subsequent system BMS communication GPS tracking maps a music player and some logging functionality 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN The second attempt at a unified system began in mid year 2010 by Thomas Walter 14 The QT C framework 15 was selected this time around and has proven to be a reliable choice QT allows for fast development of user interface elements and their associated back ends and abstracts away from thread locking and other concurrency problems The signals and slots system allows for easy communication between system elements The end product is a developer friendly easy to use system that exhibits good extensibility Currently further developments are being made to the original code written by Walter rather than rewriting for a different platform In future it may be necessary to move off this platform as the instability of it s parent company Nokia has left it s future in doubt The availability of purpose built tablet based operating systems such as Android and the upcoming Windows 8 could be viable alternatives in the future 2 1 2 Hardware The Lotus consists of centralised system around the in car
57. ne currently used in the REV Getz to estimate SoC and is reasonably popular with EV enthusiasts in Australia Coulomb counting will form part of the solution for estimating the remaining state of charge in the REV Lotus 34 5 1 Background 5 BATTERY ESTIMATION Figure 23 TBS eXpert Pro A Coulomb Counting State of Charge Indicator Source TBS Elec tronics 5 1 2 Electrochemical Modelling Electrochemical modelling is another method of estimation This technique attempts to model a cell at a molecular level and determine the SoC therein Due to the nature of chemical reactions it employs significantly more complex mathematics than coulomb counting 54 Such models offer great accuracy in estimating the remaining capacity of a cell although their complexity is often what makes them ill conditioned for direct implementation into vehicles 50 The many parameters used in a cell model must be obtainable from the vehicle This often includes variables that are not usually obtained from most battery management systems such as individual cell temperature The biggest drawback of such a system is the extra number of sensory data that would be required increasing cost and complexity of systems in a vehicle As such it is unlikely that a commercial vehicle would implement such a model especially as other methods of SoC determination are cheaper to implement Most importantly to build a proper model of a battery there are certain parameter
58. no 1 2 pp 293 300 1999 67 K Chau K Wu and C Chan A new battery capacity indicator for lithium ion battery powered electric vehicles using adaptive neuro fuzzy inference system Energy conversion and management vol 45 no 11 12 pp 1681 1692 2004 68 Pybrain The python machine learning library PyBrain 2011 Online Available http wwww pybrain org 69 Numpy NumPy 2011 Online Available http numpy scipy org A Appendix 59 A l BMS Protocol Source Ivan Neubronner A APPENDIX A 1 BMS Protocol Source Ivan Neubronner REV 012 NEU BMM01 COMMAND SET 1st byte 2nd byte 3 amp 4 byte 5th byte line feed amp return 2 bytes 1 word 1 byte E ee AER HAL ee eae EE Vots v 1 0 0 p BateyBMMNo1 V 133029 Batt volts 3 30V 29V v 15 o 0 BateyBmmnoto V1933532 Batt19 volts 335V 32V Total volts Identit A A A A AAA AA A SADA vits H 38 0 0 Al Battery BMM s H 38 0 100 100 BMM s set to 3 8 volts LT Min volts NEREE CINE NOMEN NEUEN AAA CEA EA Min max M 3 0 0 JjBateyBMMNo3 M 3 30 35 3rdbyte MIN 4th byte MAX log logger laSa Rela tn ovo ie 30V 35V Logger is reset to 50 00 after read Clear po j o o M 0 0 0 Loggeris reset to 50 00 without read NOTE When the Max Volts setpoint is passed the 5th byte in the V command displays Shunting Current in Amps 1 10 0 1A Fault condition is when th
59. nt of a CAN based motor controller system that can respond to user requests on a touch screen e Development of CAN based I O board for interfacing vehicle sensors e Development of an Android based operating system specifically for car touch screen use e Development of an embedded engine sounds solution using a chip with on board DSP and swappable SD cards containing car audio information e Evaluation of ICE audio emulation with respect to pedestrian safety e Further study and evaluation of battery estimation models 53 REFERENCES e Development of hardware state of charge estimation solution for integration into Battery Management Systems e Development of a State of Health indicator for Battery Management Systems 8 References References 1 Can bus practical bus length Softing 2011 Online Avail able http www softing com home en industrial automation products can bus more can bus bit timing practical bus length php navanchor 3010538 2 The rev project The REV Project University of Western Australia 2011 Online Available http therevproject com 3 Wa electric vehicle trial WA Electric Vehicle Trial 2011 Online Available http www waevtrial com au 4 L Schipper H Fabian and J Leather Transport and carbon dioxide emissions Forecasts options analysis and evaluation Asian Development Bank 2009 5 E E Firm and C C L of Congress Who Kill
60. obtain feedback from the motor controller The problems with estimating battery capacity for the REV Lotus are introduced Current methods of estimation both in production and in research are discussed A new method is tested offering performance with an average error of approximately 1046 through use of a combined neural network coulomb count and Kalman filter approach Acknowledgements The author would like to acknowledge several people to whom this project would not have been possible without their input and support I would like to thank Thomas Br unl for providing this project and his leadership of the REV Project Furthermore I would like to thank Ian Hooper and Ivan Neubronner for all their expertise and assistance throughout the year I would also like to extend thanks to Beau Trepp our spirited early morning discussions helped generate new ideas and forced me re evaluate several approaches I would finally like to thank friends family and work colleagues all of whom have accommo dated with my erratic schedule whilst this dissertation was completed Glossary of Terms API ANN BMS CAN CARB ECU EKF FUDS GA GPS GUI ICE IO KF OBD PWM REV RPM SA SAE SoC SoH Application Programming Interface Artificial Neural Network Battery Management System Controller Area Network California Air Resources Board Engine Control Unit Extended Kalman Filter Federal Urban Drive Scheme Genetic Algorithms Global Positi
61. oject 1 INTRODUCTION Figure 3 2010 REV Formulae SAE Electric Source http therevproject com Figure 4 2011 EV Works Ford Focus for Electric Vehicle Trial Source http therevproject com 1 2 Electric Vehicles 1 INTRODUCTION 1 2 Electric Vehicles Electric vehicles EVs are slowly coming to prominence as an alternative to conventional ICE vehicles Ever depleting reserves of petroleum is going to force an uptake of a different propul sion system within the next century In addition pressure from the effects of global warming to reduce emissions means that cleaner technology fuels will need to be found for the transport in dustry which currently produces 2396 4 of CO2 emissions worldwide EVs are positioned well to respond to these factors due to their efficient drive systems and ability to use multiple power Sources Most modern EVs use some form of battery storage to drive an electric motor In the past lead acid batteries were used but new designs have focused mainly on Lithium Phosphate cells LiPO4 Lithium cells have a flat voltage response over most of the capacity and have a reasonable capacity compared to their lead acid counterparts These cells are those used in plug in EVs although hydrogen fuel sources could potentially be used to power EVs in the future which would still use similar drive mechanics as current EVs DC motors are most commonly used although more expensive powerful AC motors ar
62. oning System Graphical User Interface Internal Combustion Engine Input Output Kalman Filter On Board Diagnostics Pulse Width Modulation Renewable Energy Vehicle Revolutions Per Minute Simulated Annealing Society of Automotive Engineers State of Charge State of Health LIST OF FIGURES LIST OF FIGURES List of Figures 1 REV Eco The Converted Hyundai Getz Source http therevproject com 1 2 REV Racer The Converted Lotus Elise Source http therevproject com 2 3 2010 REV Formulae SAE Electric Source http therevproject com 3 4 2011 EV Works Ford Focus for Electric Vehicle Trial Source http therevproject com 5 REV Lotus UI 2009 Source Varma 2010 Source Walter and 2011 7 6 REV System 2010 Source Walter 2 cR da a 9 7 REV System 201 x uut eege et EE a GSS GO Sie cibus 7g 10 8 GUI changes 20T T ic ote orat erit AC oct A rita Sette BT e tos e NC e es 11 9 Screenshot of battery cells page o oy REX EUR 12 10 Comparison of BMS GPS logic 3 25 sco ex exo RE er See aa 12 11 Twitter Program Structure 0 SEL SY e sc Y e TRANCA A SEXO 13 12 Lotus Twitter Feed Display 4 9 RES mum RE R9 RE 14 13 Comparison of Electric and Internal Combustion Engine Vehicle Noise at Different SEET a tuor dr latins tae le 549 y SES We SE ARA S NEM BAM S NEN NNI 15 14 Engine Sounds Program Logic 19 15 High Speed CAN Network oce ope ax ave eye e Bee e See a
63. op library although it adds a significant overhead to each function call As this is to run in a real time with many calls made each second this was not going to be an appropriate option As such the decision was made to run each the Lotus UI system and Engine Sounds as separate processes and allow them to communicate over TCP To implement the system so that is controllable through GUI the program is executed as QPro cess object which creates a running process on the machine and allows it to run in the back ground Engine Sounds acts a TCP server on the local host address of the machine using port 5000 From here linking the programs is enabled by connecting to the server and transferring commands across These commands determine whether Engine Sounds should start stop change volume or play a different RPM sound file These commands are detailed in Table II Commands are both transmitted and read as ASCII The final system operates according to Figure 14 There are currently two types of cars that the system can currently emulate a Ferrari and an ICE Lotus These are user selectable on the system Screen 18 3 2 Design amp Implementation 3 ENGINE SOUNDS Command Argument Action STAR None Initialise Engine Sound buffers NM Volume Percentage of system volume 0 to 100 Change volume to HHHH RPM 0 9999 RPM Load and play sound according to HT QUIT None Close engine sounds Table II Engine S
64. opments Several different CAN controllers have been reviewed and their strengths discussed A CAN based approach has been developed for attaining information from a CAN enabled motor controller and has been used to drive an analogue signal In addition a systems has been described for passing CAN message to a PC 5 Battery Estimation 5 1 Background Battery estimation is a large issue currently facing EVs An important feature of any vehicle is being able to accurately estimate the amount of fuel remaining so as to make decisions regarding use of the vehicle It also a particular important parameter with respect to evaluating the range of the vehicle and it s behaviour at differing battery levels The battery cells typically used in EVs are Lithium Iron Phosphate LiFePO cells which have an extremely flat voltage to charge ratio so common terminal voltage estimation methods do not work Thus research and development of more accurate methods is required Suffice to say a vehicle without the ability to determine it s fuel level would not be successful in the market It is important to note though that fuel gauges in cars are notoriously unreliable and tend to have a large margin of error As such any estimator that could achieve 10 of the actual fuel level would be appropriate for implementation The battery cells employed in the Lotus are LiFePO cells This chemical make up was discov ered at the University of Texas in 1997 46 T
65. ot perfect and there is a slight variation in pitch between the up shifted old track and the down shifted new track This effect is most present in current RPM translation of the system as by it s definition the geared version of the program tends to favour certain RPM values and as such prevents fluctua tion Stability can be attained by several methods moving averages and kalman filtering although reduces response time and or adding code to detect when this occurs Detecting this situation is the ideal method as it does not negatively impact on the response time in any meaningful manner and is simple to implement All that is required is to reject any changes to the value of RPM that increase it below or beyond a midpoint threshold if the last two values for RPM are within a certain percentage of each other In this case the tolerance was set to 5 and this reduced fluctuations to an acceptable level 3 4 Summary The Engine Sounds system successfully builds upon the work undertaken in previous years and introduces a working system to the current Lotus UI system The Cf code has been improved to 22 4 CAN BUS SPY reduce audio disruptions from track switching discontinuities that occur during system use The addition of emulated gear system has allowed the system to better approximate the sound output of a conventional petrol vehicle 4 CAN Bus SPY The instrument cluster in the Lotus consists of a speedometer tachometer and va
66. ounds TCP Commands Start Execute QProcess em d Connedto Engine Sounds Conneded to server SendRPMVolum _____J___4 Information X T T Start Active Libraries Start TCP Server Waitfor client Has client connected e dE Waitfor client Received start command Terminate Engine Sounds Lotus Main System Initialise buffers gt System Cal t TCP Communication Engine Sounds Process Waitfor nextcommand Receive volume command Changevoume Receive RPM command Change RPM sound Receive stop command Disconnect client Terminate Engine Sounds Figure 14 Engine Sounds Program Logic 19 3 3 Evaluation amp System Tuning 3 ENGINE SOUNDS 3 3 Evaluation amp System Tuning Following now is a discussion of tuning the system to get a reasonable performance In defining what a reasonable performance is we set up a brief list of criteria Firstly it should respond quickly to user input If the driver puts the their foot down on their accelerator it should be expected to be produced an increasing RPM sound in well under one second Secondly we ideally want a low amount of jitter This means that if the accelerator is held constant sound tracks should not flip back and forth creating wave discontinuities Thirdly track swapping should be as inaudible as possible Finally the sounds played should be appropriate for the specific driving condit
67. powerful vehicle than the Getz with the goal of proving that electric vehicles can meet the high performance demands imposed on sports cars As a later conversion it includes a BMS system that was designed at UWA unlike the Getz It also uses a much more powerful three phase AC motor instead of a series wound DC motor The Lotus includes a fully featured in car PC operating a 1 6 GHz atom processor running Microsoft Windows XP As such it does not have as many limitations as the Gumstix powered Eyebot M6 Functionality is driven by a program designed in the QT C framework The final vehicle to be converted is the REV Formula SAE vehicle This donor vehicle for this conversion project was the UWA Motorsport race car from 2001 The end goal of this conversion is to produce a new SAE Formulae vehicle purpose built for the electric section of the Formulae SAE competition Currently this project is on going and slated to compete in Melbourne in December 2011 and then in Germany mid 2012 The REV team has overseen the conversions of the Ford Focus fleet for the WA Electric Vehicle Trial 3 In addition to this capacity telemetric and logging software built by members of the REV team is currently being used to evaluate electric vehicle transport in the metropolitan area As these vehicles are driven far more often then the current fleet of REV built cars the team hopes to collect far more useful data for performance evaluation 1 1 The REV Pr
68. raining data and tended to only operate on single cells and not battery pack Their tests were also far simpler usually involving standard battery discharges tests such as the Federal Urban Drive Scheme FUDS and Manhatten power cycles which generally place higher loads on batteries than those in practice Limitations on data obtained from Lotus BMS in comparison to a bench test focusing on one cell also leads to discrepancies as cell balancing issues obscure the true capacity of the pack In the end a multi step approach was required to obtain a reasonable estimate for battery capacity Before continuing on with the design a few notes on the training data must be made The data consists of a single set of approximately 55000 data points This was obtained from a test drive of the Lotus performed on 5 08 2011 for approximately 110 minutes on a day with a max temperature of 23 degrees Celsius During this drive the battery pack could only discharge 3096 of it s maximum value as two cells were well below the pack average and further driving would have caused damage to the cells Whilst it was planned to obtain further data damage to the Lotus incurred just days later had meant access to the Lotus was suspended and as such no more training data could be obtained There is a second training set that was taken on 29 07 2011 but as there was a scaling issue with the amperage of the coulomb count the data had to be discarded Table IX illustra
69. rawbacks that meant it was passed up on The code is tightly bound to Labjack meaning that a rewrite of the code would be necessary Sound quality is much poorer and it produces very audible popping noises from sound wave discontinuity as different sound files are shuffled in and out It also lacks any form of volume control and as most of the sound processing functions are not from any well known libraries would be be difficult to implement Adding to all of this is that the code was very poorly documented with barely any comments and no accompanying explanation for various functions in the program The CF version was far more usable It is cleanly commented and logically structured Whilst it lacks the gear transmission emulation functions these are easily ported from the C version of the source code By far the greatest advantage is in its use of Microsoft s DirectSound libraries and as such audio quality is much improved Fortunately there are DirectSound ports on Linux and mono enables C code to be runnable on Linux so the choice of C does not impact our choice of platform too greatly The major issue with the selection of C is it s inability to be linked from C Whilst both languages are similar QT C is a type of unmanaged C code and as a result the compiler tends to mangle the names of linked functions This of course is not compatible with the managed nature of C This can be gotten around through the use of the COM inter
70. re implemented one that bases the value for RPM with a current translation and one that uses speed and current as RPM and throttle respectively As the Lotus is refined with more sensors the program should be refined to have automatic and manual trans mission modes and operate in only the geared version of the code The author predicts that the current RPM version will be phased out although it is currently the most reliable conversion method Responding swiftly to user input is possibly the most difficult element and one to which a per fect solution has not been implemented as of current Quite often attaining a swift response is antagonistic with our second goal which is to attain a low amount of jitter Any sort of processing that correct for jitter such as weighted average means and kalman filters quite often add some processing cost which reduces response times However the biggest factor by far is the speed of which we obtain our inputs Chiefly the suggested mechanism is to use current to determine RPM This is not a bad idea as current is related to the speed and throttle of the vehicle Where this cre ates issues is the current speed at which current values are obtained which is specified in the BMS at only around every 500 ms In addition because the Lotus PC system runs on timers and not threads just because a timer goes off does not mean that the information is processed immediately so there is an additional overhead Although
71. requires knowledge of the initial state and must monitor the current at all times The simple requirement has one drawback that is not often considered having the electrics on all the time creates an added risk where if the vehicle is not correctly put on charge the potential arises for the battery pack to be drained out overnight much faster than it would be if it were disconnected In addition there is always some power leakage occurring and thus the naive solution of saving the last SoC of the battery before switching all power off is not likely to be accurate Another factor is the concept of measurement drift where inaccuracies in readings taken over a period of time can add up and contribute to significant error Coulomb counting is particularly prone to this as it ideally requires attaining values for instantaneous current as often as feasibly possible For this reason coulomb counting is often paired with a way to reject accumulated errors The two most used methods are recalibrating the measurement to some set point usually empty or full capacity 52 and or the use of a Kalman filtering technique 53 In spite of the drawbacks of coulomb counting it remains a popular solution Utilising it with a recalibration is simple to implement and has lead to it being utilised in a few commercial solutions The most familiar implementation to members of the REV Team is the eXpert Pro from TBS electronics Figure 23 This module is the o
72. rious other assorted indicators In 2009 when the Lotus was shipped the speedometer was replaced with one indicating km hr as the previous one indicated miles per hour As a requirement of the Australian Design Rules 31 this was replaced In addition during the conversion the Engine Control Unit ECU providing all signals to the instrument had to be removed and rewired in accordance with its new drive system Whilst the speedometer was reconnected in line with Australian requirements a tachometer is not such a requirement and as the ECU providing this signal was removed it was not attached Fortunately the UOM Powerphase 75 installed in the Lotus has the ability to output information about the engine over a connection from the motor controller 32 This includes engine speed torque voltage current and temperature It can do this through two methods one using a serial connection and the other a CAN bus connection As UOM does not provide information about the protocol used to send data over serial this could not be used It does however provide the protocol for CAN bus connections 33 based on the SAE J1939 21 standard 34 35 As such it was decided to implement a system using the CAN bus In addition to taking data from the CAN bus to create an RPM signal it would be useful to be able to also relay this information to the in car PC As such we define the following set of requirements Take input data from the CAN Bus 2 O
73. rity gets bus access immediately The application layer is the final element of the CAN system The original CAN standard does not include the tasks of the application layer and so is left up to vendor implementation Such tasks include flow control device addressing and transportation of data blocks larger than one message These implementations outside the standard are often referred to as higher layer protocols Some of the more popular protocols include DeviceNet 37 CANopen 38 NMEA 2000 39 and J1939 Table III shows a list of such protocols and their common uses in various industries 25 4 1 Background 4 CAN BUS SPY Table IV Bit rates for specific bus lengths Source 1 Bit Rate kbit s Bus Length m Bit Time us 1000 30 1 800 50 1 25 500 100 2 250 250 4 125 125 8 62 5 1000 20 20 2500 50 10 5000 100 CAN is designed to operate in two different modes high speed and fault tolerant Fault tolerant mode is typically defined at speed 40 100k transmission rates whilst high speed mode operates at above 100k up to IMbit s High speed CAN is the most popular standard for the physical layer and is utilised in the DeviceNet and openCAN standards High speed is terminated with 120 ohms on each end of the network Low speed differs in that each device has it s own termination and can continue operating in case of a wiring failure and is less common Typically the speed of the CAN bus is related to the length of t
74. rive current is applied to the estimator the system instead uses the coulomb count as it s measurement update effectively skipping the measurement step In doing so the system essentially acts as a coulomb counting system that utilises a neural network to recalibrate the SoC when the load on the battery pack exceeds a pre set limit Correct tuning of the Q and R parameters is necessary to form an accurate estimate As coulomb counting is used in our model for SoC selection of a suitable value is reasonably easy A load of the average drive current per time sampling period results in a reduction of around le 4 Fluctuations in the current sensor generally occur at le 6 so we can safely assume this is approximately an amount of error suitable for the value of Q Measurement noise R is harder to determine and is best found through experimentation 48 5 2 Design amp Evaluation 5 BATTERY ESTIMATION ANN amp Kalman Filter Full Training Set m c m c wm 0 mm o o TO 0 20 3 40 50 a 70 a EJ 100 Timefminutes Figure 32 ANN amp Kalman Filter Testing with Full Training Set ANN amp Kaiman Fiter Low R Drive Curent gt 40 Amps ANN amp Kalman F iter High R Drive Curent gt 40 Amos E E gos 4 E Sr S 4 P H E Sos 1 El 3 E Bos 4 E 5 Zon 7 100 E 100 Tre mnes ANN amp Kalman Fiter Higher R Drive Current gt 40 Amps ANN amp Kalman
75. rms and not directly specified The most common implementation of the physical layer is through a two pair twisted wire terminated with two 120 ohm resistors as shown in Figure 15 This gives the CAN bus very high noise immunity through a low level of differential voltage The twisted pair arrangement also helps to ensure that RF interference is kept to a minimum Connector types and pin outs continue to lack formal treatment and auto manufacturers tend to have their own specific connectors The most common connectors tend to be in the form of DE 9 and ODB ii It is often the case that second pair of wires runs parallel to the CAN L and CAN H wires carrying a voltage supply rail and grounding rail in order to provide power to CAN nodes The transfer layer represents the bulk of the CAN standard and as such is too detailed to cover completely here It passes valid messages to and from the object layer and is responsible for bit timing synchronization message farming arbitration acknowledgement error detection error signalling and fault confinement Perhaps the most important aspect is message arbitration Mes sage arbitration occurs in priority based arbitration system All messages transmitted by a CAN 24 4 1 Background 4 CAN BUS SPY Table III Popular CAN high level protocols and their common usage Protocol Use DeviceNet Factory Automation CANopen Medical Equipment Vehicles Automation Rail Electronics etc NMEA 2000 Marin
76. s important and it reduces the need to flick to the system log to check common conditions The next change is re skinning of the GUI to make it more appealing A green background fitting the environmental theme of the vehicle was selected and a set of buttons replaces the grey tabs at the bottom of the screen This has generally been well received by members of the team Lastly a transparent frame class has been added to the code The object orientated nature of the QT GUI forces all UI elements to inherit off a base class frame These however can be promoted to the transparent frame class resulting in curve edged transparent frame seen in Figure 8 2 2 System Updates 2011 2 OVERALL SYSTEM DESIGN Engine Sounds Main Roads System Log Separate System gt DataFlow One Way DataFlow amp ControSigna HTTPRequest TCP Transmission C SSES C SES Figure 7 REV System 2011 2 2 2 Battery Status Panel A limitation of previous design was the lack of information present with regards to the status of the cells in the battery pack Previously battery status was indicated by various colours but this is difficult for a human operator to interpret with reasonable accuracy As such an extra panel was added that details the last recorded value for a particular cell from the BMS with readings updated approximately every 20 seconds Included in this are the values for the minimum
77. s of the batter ies make up that must be known Between different manufacturers various factors can differ be tween chemically similar batteries such as anode cathode surface area Such differences can lead to differing performance Unfortunately such detailed information is not available as most man ufacturers are hesitant to make public their designs The only option then is to reverse engineer and probe the battery for it s various attributes but this is obviously expensive and most of all dangerous as most batteries are toxic 38 5 1 Background 5 BATTERY ESTIMATION The complexity and expense of chemical methods for determining SoC rule out their use in the REV project Other methods are far cheaper and more suitable for the needs of the project 5 1 3 Artificial Neural Networks Over the last ten years Artificial Neural Networks ANN have found increasing use in battery estimation problems 55 56 57 58 59 ANNs are mathematical models whose structure is partly inspired by biological neurons The greatest strength of these networks is their ability to approximate non linear functions through machine learning and have been described as universal approximators 60 Given a good set of training data it is possible to train an ANN to approx imate battery dynamics Neural networks have several advantages and disadvantages that will be detailed For the purposes of brevity information in this section is restricted to feed forwar
78. sions 52 7 Future Work 53 8 References 54 A Appendix 59 A 1 BMS Protocol Source Ivan Neubronner 60 A2 GPS Protocol Source NMEA 5 ic Rx a a E s ur 61 A 3 Telemetry Protocol Source John Pearce 63 AA CAN SPY Breakout Board 64 A Tachometer Amplifier Board a ds Ga we a SS 65 A 6 Motor Controller Breakout Board Source Watts 66 A 7 AVR CAN Schematic Source Olimex 67 A R Computer Listings pa Rose aa Des a a Ea 68 1 INTRODUCTION 1 Introduction 1 1 The REV Project The Renewable Energy Vehicle REV project 2 is a multidisciplinary effort to construct and evaluate electric passenger vehicles for the private market It began as a restart of earlier research into hydrogen fuelled vehicles which had been discarded primarily due to costs Since it s incep tion in 2008 the REV Project has successfully converted a number of conventional Internal Com bustion Engine ICE vehicles to full electric systems The vehicles converted include a Hyundai Getz a Lotus Elise and a formula SAE Vehicle An extra conversion has been done for the Elec tric Vehicle Trial for a fleet of Ford Focus but the physical conversion was undertaken by a third party with members of the REV project in consulting positions Each of these vehicles represents a different aspect of the goals of the project The converted Hyundai Getz
79. specific meaning It is very difficult to infer meaning from layer weights and it not normally possible to create a model of battery dynamics from understanding the network structure As such neural networks are best viewed as black boxes Due to the ease of implementation and versatility of neural networks an implementation is used for charge estimation in this project 5 1 4 Kalman Filter The Kalman filter 63 is one of the most famous developments in control systems theory It is a mathematical model that allows measurements to be taken and computed to form a more accurate reading and reject noise It does this by taking measurements comparing them to a model of the system and computing a more accurate reading based on the noise of the model and the actual measurements The Kalman filter is simple to implement and provided an appropriate model can be constructed tends to produce stable and accurate readings Although there are many variants of the Kalman filter namely unscented and extended variants for systems that are non linear functions of the system states this discussion will restrict itself to the standard Kalman filter The Kalman filter can be thought of as a two step process the first of which is commonly referred to as the predict state This system is displayed graphically in Figure 27 During this step a prediction of the next state is made using the associated model of the actual system An update of estimate covari
80. stimating state of charge and state of health of lithium ion batteries Applied energy vol 86 no 9 pp 1506 1511 2009 J Wang B Cao Q Chen and F Wang Combined state of charge estimator for electric vehicle battery pack Control Engineering Practice vol 15 no 12 pp 1569 1576 2007 G Sikha R White and B Popov A mathematical model for a lithium ion bat tery electrochemical capacitor hybrid system Journal of the Electrochemical Society vol 152 p A1682 2005 J Wang L Xu J Guo and L Ding Modelling of a battery pack for electric vehicles using a stochastic fuzzy neural network Proceedings of the Institution of Mechanical Engineers Part D Journal of Automobile Engineering vol 223 no 1 pp 27 35 2009 C Bo B Zhifeng and C Binggang State of charge estimation based on evolutionary neural network Energy conversion and management vol 49 no 10 pp 2788 2794 2008 N Cui C Zhang Q Kong and Q Shi A combined method of battery soc estimation for electric vehicles in Industrial Electronics and Applications ICIEA 2010 the 5th IEEE Conference on IEEE pp 1147 1151 S Qingsheng Z Chenghui C Naxin and Z Xiaoping Battery state of charge estimation in electric vehicle using elman neural network method in Control Conference CCC 2010 29th Chinese IEEE pp 5999 6003 M Charkhgard and M Farrokhi State of charge estimation for lithium ion batteries us
81. sy that is detailed in Who Killed the Elec tric Car 5 whereby car companies produced EVs to be allowed to participate in the lucrative 4 1 3 Automotive Instrumentation 1 INTRODUCTION California market but did not advertise them and or put them on unpopular lease terms refusing to sell outright to the consumers and repossessing the vehicle at the end of the term Those critical of the petroleum industry point to the lobbying by oil heavyweights as large reason for this behaviour by manufacturers Unfortunately despite the research into EVs during this time low oil prices meant that consumers favoured large sports utility vehicles The economic crisis and the global warming debate has fuelled recent interest in EVs as oil prices rise and make owning less efficient vehicles less affordable In 2004 Tesla Motors a small up start California car company began work on the Roadster which was eventually available in 2008 The Roadster went in an area where electric cars had not been before use as a high per formance sports car This vehicle caught the attention of the media and the big car companies and renewed their interest in EVs Following this GM introduced the Chevrolet Volt and Nissan produced the LEAF The future of EVs looks to be an exciting time as the existing car companies fight to gain a competitive advantage in a lucrative new market 13 Automotive Instrumentation Automotive instrumentation is a slowly evolving
82. t emitting serial messages is the most challenging element Setting the timer to 100 microseconds should be all that is required to transmit messages and drive the tachometer Most important is of course capturing the relevant message Fortunately most CAN controllers off the ability to filter for specific messages by specifying a message and a mask In doing so only the selected message or range of messages are written to the CAN controllers buffers The particular message that is to be filtered for is displaying in Figure 20 31 4 2 Design 4 CAN BUS SPY PDU Format PF PDU Specific PS Source Address SA Bit31 29 Bit28 26 25 24 Bits 23 16 Bits 15 8 Bits 7 0 ba ACCURATE FEEDBACK OXO4EFRRSS OR OXRR9 Signed Torque Signed Voltage Signed Current Signed Speed Byte 1 Byte 2 Byte 3 Byte 4 Byte 5 Byte 0 Byte 7 Byte 8 Transmission repetition rate Increments of 0 006 seconds 0 102 seconds default user adjustable Data length 8 bytes Data page 0 PDU format 239 PDU specific DA Default priority 1 Byte 12 Signed Torque Feedback 1 10 34 Signed Voltage Feedback 1 12 5 6 Signed Current Feedback 18 7 8 Signed Speed Feedback is Figure 20 Accurate Feedback CAN Message Source UOM CAN Communication Summary 4 2 3 Alternate Hardware Design Cost is an important consideration of the design and one that wasn t looked at too closely in the preceding hardware design section The platform selected was the AT90CANI12
83. tart of Frame SOF Ex tension Identifier Bit IDE Remote Transmission Request RTR Reserve Bit r0 Data Length 26 4 1 Background 4 CAN BUS SPY CAN Device A CAN Device B CAN Device C CAN H Line uiuo OCT WYO OCT CAN LLine Figure 15 High Speed CAN Network SOF Identifier A IDE Identifier B RTR r0 DLC Data CRC ACK EOF Figure 16 CAN Extended Frame 21 4 1 Background 4 CAN BUS SPY 25900000004 MOOZ 1 5 Signal Ground 2 4 Chassis Ground 3 6 CAN High J 2284 4 7 ISO 9141 2 K Line 5 14 CAN Low J 2284 6 10 J1850 Bus 7 2 J1850 Bus 8 15 ISO 9141 2 L Line 9 16 Battery Power Figure 17 OBDII Pinout Diagram Source http www pinout net Code DLC CRC Cyclic Redundancy Check Acknowledgement Bits ACK and End of Frame EOF OBDII is the current diagnostic standard for vehicles All cars sold in the United States past 2008 are required to implement ISO 15765 4 a variant of the CAN network bus Typically it is a connector underneath the dashboard that may implement a variety of different network standards Figure 17 is a typical diagnostic connector pin out diagram 4 1 2 Tachometer Tachometer implementation varies between year to year Today CAN bus driven tachometers confirming to a messaging standard are reasonably common However the Lotus was constructed in 2002 It wasn t until 2005 th
84. ter System Convergence Tests 50 51 35 Block Diagram of Full ANN amp Kalman Filter System LIST OF TABLES LIST OF TABLES List of Tables I Brief History of Automotive Advances 6 II Engine Sounds TCP Commands 19 II Popular CAN high level protocols and their common usage 25 IV Bit rates for specific bus lengths Source 1 26 V Frequency vs Tachometer Value 29 VI Connector specifications in system 30 VII Costs of Various Chips Source Mouser 33 VIL CAN SPY Testing List o os se sos acaos pe almost ae ue Gate gs ae 34 IX Statistical Data of Training Set 45 Contents 1 Introduction Ll Che REV Project lar ne SC NOE QUON SU UE QURE DIRE S EHE ae A 1 2 Electi Vehicles et et Seu eS tS eet SUIT Deke aed See Syed By 1 3 Automotive Instrumentation Lan eot x oboe Roto a E A 1 4 REV Instrumentation amp Document Structure 2 Overall System Design 2 Background 8 29592 50999 0990 BoseD a ad HN aS a uo eono AE 2 1 2 Hardwate des uo gan e A ere edo ee a Se 22 System Dpdates 2001 4 525 cd Roe E a eT UN R e ON RUN E 2 2 DEDO CAOS sermon S enint RE NC RE NGA RR INCUN 2 2 2 Battery Status Panel 4 uude RE e keen 2 2 3 Enhanced Logging Functionality 2 2 4 GPS BMS Core Modific
85. tes what data was obtained as well as statistical information of the training set Parameter Mean Median Minimum Maximum Std Deviation Min Cell Voltage per BMS cycle 2 99 3 04 2 67 2 26 0 16 Ave Cell Voltage per BMS cycle 3 26 3 26 3 19 3 34 0 026 Current 10 1 44 128 4 28 4 13 99 SOC 0 81 0 8 0 71 1 0 0 01 Table IX Statistical Data of Training Set The parameters were sampled every 100 ms The average cell voltage and minimum voltage per BMS cycle were selected for the model as this is roughly equivalent to the model whereby a load on the battery will produce a voltage drop The minimum cell in the pack at any instant in time if 45 5 2 Design amp Evaluation 5 BATTERY ESTIMATION the pack is balanced well enough is the cell that is sampled when drive current is at its maximum The difference between the average cell voltage and minimum cell voltage is not precisely the voltage of any one cell under a particular load but it is equivalent Finally the system was designed and evaluated using the python programming language Ex tensive use of the PyBrain 68 and NumPy 69 packages has been employed 5 2 1 Artificial Neural Network The first step is to build and evaluate neural networks in an effort to determine what a reasonable network structure for our purposes will be In doing this several different neural networks were constructed and trained on a set of data acquired from the a test drive of the Lotus
86. the Twitter Application Programming Interface APD which can then respond with an XML response This response can then be parsed to format the data The HTTP requests are sent using a C library called Twitcurl which includes several wrappers for HTTP requests that are sent using cURL Figure 11 shows a graphical outline of this system whilst Figure 12 shows the Twitter API panel in operation A decision was made to allow the user to refresh the information on command as opposed to by a timer in an effort to reduce data costs incurred by the vehicle s 3G connection MAIN Call Library Function 4 T Parse XML response C TWITCURL LIBRARY Construct HTTP request 4 T Pass XML response cURL SendHTTP request 4 T Receive XML response TWITTER Figure 11 Twitter Program Structure 13 2 3 Summary 3 ENGINE SOUNDS Figure 12 Lotus Twitter Feed Display 2 3 Summary Several changes were made to the system in 2011 in order to improve robustness along with the addition of several new features The BMS GPS and telemetry systems now operate with added reliability and are capable of recovering from system errors The BMS and GPS systems also received enhanced logging functionality for use in testing and emulation The UI has gained an improved look along with features that will increase the usability of the system Finally the addition of the Twitter module enables traffic information to be relayed to the driver 3 Engine Sounds
87. this overhead is quite small and indistinguishable cur rently if the program gets more complicated and it likely will it will increase Currently because data sent from the BMS is used to drive engine sounds the response is quite poor The use of the RPM value from the CAN bus which is described in the CAN spy section will improve this as by it s nature it operates approximately 10 100 times faster than the BMS serial line To understand how pop effects occur is to understand how the program operates it s sound buffers The program has access to a set of sound files based on RPM values at intervals of 100 RPM from 2000 RPM to 7000 RPM As an example if accelerating from 2000 to 2100 RPM continuously the system will frequency shift the 2000 value to 2049 and at the mid point will switch tracks to a 2100 RPM sound frequency shifted to 2050 RPM At the point that a buffer switches tracks a small popping noise can be heard caused by sound wave discontinuity The technique employed to reduce this is to utilise multiple sound buffers with varying amplitudes 21 3 4 Summary 3 ENGINE SOUNDS The last RPM track sent by the program is the the loudest and sound tracks that were previously played are slowly reduced in volume so that upon switching sound tracks the popping effect is indistinguishable This leaves us with one major design decision to make how many buffers do we use Using too many buffers even with a reasonably large decay on
88. through serial whilst also generating a PWM signal Olimex provides the AVR CAN development board 45 of which uses an AT90CANI128 chip from ATMEL as well as the MCP2551 chip a popular CAN transceiver This board provides a DE9 CANBUS connector as well as a DEO serial connector In addition it provides several input output pins and 4 PWM channels As such it is perfect for our requirements Summary data sheets for the AVR CAN AT90CAN128 and MCP2551 can be find in Appendices The AVR CAN board includes two 34 pin IDC plugs to access various pins In order to access these lines a breakout board is constructed in order to access these pins through screw terminal connections A schematic of this breakout board is included in the appendix Finally the board supplies PWM signals at 5 volts This signal must be amplified to between 7 12 volts A schematic is included in the appendix Figure 18 illustrates the overall system design Table VI lists the various connectors 29 4 2 Design 4 CAN BUS SPY UOM Powerphase Motor Controller Breakout Board CAN H CAN L TID 12v Supply AVR CAN Breakout L Amplifier Circuit keb To Tachometer Board Extl Ext2 AVR CAN gen gt Serial Line to PC Figure 18 CAN SPY System Design Line Connectors Motor Controller Breakout Board MC Am19 to Am19 Breakout Board MC AVR CAN AVR Stripped to DE 9 AVR
89. ty in the most depleted cell Before discussing the most popular methods for estimating SoC it is important to outline some criteria that must be met in order to estimate SoC on the Lotus The Lotus has some special require ments that are not present on commercial EVs due to the way in which varying electrical elements of the car are designed to operate The car PC BMS battery pack and car electrics are all operating on different power circuits 44 As such it it possible for the BMS and the car PC to be turned off whilst the rest of the car is on A consequence of this is that the BMS may not be able track every event that is occurring to the pack and as such tracking the SoC by knowledge of it s previous state alone will not accurately determine SoC Likewise if the car PC is switched off it obviously will not receive data from the BMS Most importantly this means SoC must be determined from solely from parameters at a particular instant in time or at the very least converge to an accurate estimate after some period of time Secondly online computation of SoC should be kept to a mini mum Ideally the driver wants accurate estimates of remaining capacity as often as possible Most research papers regarding battery estimation methods for electric vehicles perform tests on one cell and suggest creating estimators for every cell in a pack A lot of estimation methods use integrals derivatives multiplication and division which by their nature ar
90. utput this over the serial line 3 Alter a Pulse Width Modulated PWM signal to drive the tachometer Following on now is background information about the CAN bus Afterwards the system design in both hardware and software is discussed finishing with the evaluation of the system 23 4 1 Background 4 CAN BUS SPY 4 1 Background 4 1 1 CAN Bus In 1986 Bosch 36 introduced the CAN bus to the world It was designed to allow several different car components to communicate with either and to support messaging priority Motors motor controllers instrument panels etc are often designed to send and receive standard CAN messages Most cars being produced today include a CAN bus and it is common interface to com plicated fleet management systems used in the transport industry Although originally designed for passenger and commercial vehicles it has seen uptake in the agricultural and manufacturing industries The CAN bus layer is based on several layers of abstraction In it following layers are defined Application Object Transfer Physical layers and Data Link Layers These are based on the fol lowing ISO specifications ISO 11898 1 2003 Data Link Layer amp Physical Layer ISO 11898 2 2003 High Speed CAN Bus ISO 11898 3 2006 Low Speed Fault Tolerant CAN Bus ISO 11898 4 2004 Time Triggered Communication Protocol ISO 11898 5 2007 High Speed CAN Bus Extension Typically the physical layer is only referred to in abstract te
91. w elements of this protocol are not relevant to the REV vehicles or are not tracked by the server telemetry statistics and therefore can be omitted or implement only a basic state One such element is the reason codes for which only the journey start finish and time interval update codes are implemented Appendix A3 includes the telemetry protocol by Pearce The structure of the original program was incredibly complicated for the function and has un dergone some significant simplification to ease maintenance of the telemetry function The 2010 version worked by building a buffer of messages and attempting to send these to the server If the server did not acknowledge a message it would clog the buffer up until it could be sent and removed from the queue This has been greatly simplified by running a separate timer in Log ger class which constructs and sends a message every x seconds where x is the timer interval Messages that are not received by the server are destroyed 2 2 6 Main Roads Twitter Panel Another minor addition to the code is the addition of a panel displaying information from Main Roads Twitter account 17 In effect this allows the car to display information about traffic incidents in Perth that are posted to Twitter forewarning the user of road closures accidents traffic light malfunctions and other road issues that can occur in the Perth metropolitan area It operates by sending HTTP requests that are specified through

REV Performance Vehicle Instrumentation

Contents

Download Pdf Manuals

Related Search

Related Contents