PMD500: SX-Plus
Contents
1. Sartorius AG Weender Landstrasse 94 108 37075 Goettingen Germany Phone 49 551 308 0 Fax 49 551 308 3289 www sartorius com Copyright by Sartorius AG Goettingen Germany All rights reserved No part of this publication may be reprinted or translated in any form or by any means without the prior written permission of Sartorius AG The status of the information specifications and illustrations in this manual is indicated by the date given below Sartorius AG reserves the tight to make changes to the technology features specifications and design of the equipment without notice Status August 2009 Sartorius AG Goettingen Germany Printed in Germany on paper that has been bleached without any use of chlorine W1A000 KT Publication No WPM607 1 e090822. effectively be used to cancel out unwanted effects such as physical properties e g particle size and to linear rise the response Correctly applied pre treatment often yield a more precise and robust model NOIMAGE Removes image data from the observations if present MEAN Subtracts the mean of your vector SMOOTH 2 Smoothing over 5 data points along each vector DG 1 5 3 Computes first derivative using 5 data points and a smoothing window of three data points DG 2 3 5 Compotes second derivative using 3 data points and a smoothing window of five data points 12 SNVT Subtracts the average and divides by the standard deviation of the vector AVG Calculated the average of repeated measurements Repeated measurements hold same sample ID AVG 2 Calculated the average of two consecutive x and y values irrespective of sample ID OSC Computes a model of the data that is perpendicular to the modelling target and subtracts it from the observations EMSC 1 Computes a second order fit of to the spectra and remove it from the data EMSC 2 Computes a second order fit of the means spectrum to the spectra and remove it from the data X2 X0 5 Power transformation 7 4 5 Regression methods Determine the regression method Examples PCR Standard PCR Method PLS Standard PLS Method PLS2 Standard PLS2 Method XLS PLS with second derivative RR Ridge Regression MLR If a plus sign is added to the method the software
3. variables are a linear combination of them selves this can also be seen as an ambiguity in solving the equation since many solutions give approx the same result Extreme colinearity will in many cases lead to a poor model in these cases you will need to select the appropriate wavelengths by some means 4 3 7 Interference Term that a measured signal is disturbed 4 3 8 Pre treatment Pre treatments are mathematical roformating functions of the original measured signal these are sometimes need to e g cancel particle size effects 4 3 9 Report abrevations In the report window SX Plus lists various number these are XCVAR Calibration set observation variation XYVAR Validation set observation variation SEC RMS of calibration residuals CMAX Maximum calibration residual SECY RMS of cross validation residuals iie residual RMS of validation residuals VMAX Maximum validation residual OVAL Standardized ratio of residual when in and outside of calibration set negative if outside yields lesser residual than in Number of instances X Observations Y Target population 5 Questions amp Answers Applying NIR and before applying NIR as a measurement method many questions are raised prior to use Some can be answered directly but a majority need to be included as a part of the assessment of the method 5 1 How accurate can measure This is probably the most frequently asked question
4. Unfortunately there is no exact answer since there are many parameters that affect the answer If we separate the answer two measuring subtances that are known to absorb in the measured region e g Protein Moisture or Fat It is often found that the measurement accuracy is equal to the reference method error One should however always conduct a study and determine the error for each application since other effect as impurity sample presentation may have significant effect on the accuracy obtained 5 2 How many samples do need to make a calibration A round number is 100 and this i soften not a bad number of needed samples Again the complexity of the sample composition determines the actual number of needed samples An estimate can be calculated by taking 20 samples with various concentration covering the future expected range and adding 3 samples for each expected disturbance variation of other parameters that should not cause change in estimate To optimize your model you will need a further 10 samples plus 3 samples for each disturber and a further set of at least 20 real samples to validate your model In total you will calculate the need for about 100 samples 5 3 Transfer a calibration from one equipment to another The accuracy obtained when a calibration is transferred is one of the most important properties this is partly affected by instrument to instrument differences as well as the inner properties of the
5. calibration In general a calibration that works good will transfer a calibration that works poor due to inner properties will not transfer It is often a benefit if the calibration is developed using data from more than one instrument A transferred calibration often needs a new defined intercept since this property isn t a part of the calibration model 5 4 What to do when it doesn t work Reference values Determine the error of your reference values by sending same sample to your laboratory with different names Sampling Validate that your sample corresponds too your spectra are the sample uniquely identified 6 Collecting data Measurements made using SX Center can directly be exported and used from withing SX Plus Samples analysed in the Today View require a separate ID file where the target values of the samples are defined Using the Journal of SX Center enables the user to enter the reference values directly and export a table of data contining all necessary data for calculation 6 1 Collecting data using a separate ID List 1 Start SX Server Configure sample recognition Start SX Center Create a Recipe Measure the samples and enter ID s Export the data Create a Tab separated file using excel with reference values NOM fwWNH 6 2 Collecting data using SX Center Journal 1 Start SX Center 2 Select Journal 3 Enter target values 4 Select entry for begin of export 5 Export
6. graphically displayed 3D graphs may be rotated by holding down the left mouse button and moving the mouse By holding the Ctrl down and clicking with the left mouse button references can be selected By clicking the right mouse button still holding the Ctrl key down copies these references to the clipboard To e g delete the marked sample enter the delete leaf of the project and paste the sample references If not samples have been selected clicking the right mouse button will copy a white background graph into the clipboard This may be pasted into Word or Excel 7 8 Export your calibration From the menu select Project Ex port Select a location and enter the name of your calibration It is common practice to keep a version number parameter name and range 7 9 Settings Further settings can be modified selecting View Options from the main menu 8 Using the calibration To use the calibration place them in the Calibrations folder of your SX Suite installation It will automatically appear in the list of available calibrations when you edit a Recipe 9 Calibration maintenance It is in the beginning often necessary to add data to an earlier project Open the project select the data table From the menu select File Merge and select the file containing the new data The new data will be added to the end of the table Re compute and export the model
7. the data 7 SX Plus 7 1 Users interface erases LOI ie m gt c ow Tos amt amd Prcgettienane Table of Dota cm y ma at Catwrferst or tn n us ua wu NN 4 gt e D Rent Gah Esgetersse wath 7 2 Projects SX Plus maintains all data in project files The project file contains references to other files for e g calibration data and TD List It is recommended to store data in subdirectories relative to the location of the project file in this way you may easily copy data from one location to another e g create a backup on a server 7 2 1 New Project To create a new project select File New in the main menu A Empty project is created named Project1 Allways begin by saving this project to a desired name 10 7 2 2 Saving your Project You may save the project by selecting any item in the project window and click the save button You may also save it under a different name by selecting File Save As in the main menu 7 2 3 Project window The project is displayed as a tree Each entry is defined by its location and or its description showed behind each entry in parentheses A entry is only regarded if the check box is checked To edit an entry select the entry text with a single left mouse click pause and click again a entry box will be displayed Some entry s will also show a drop
8. to use 7 5 Project results Results of the computations can be viewed as a numerical table or graphically by selecting one of the bold entries in the project 7 5 1 Pre treat Wavelengths SNRs WLCs Means Variations 7 5 2 Modell Wavelengths Indexes IDs Center Target Zero Observations Weights Loads Regressions Press Scores Bias Estimates Residuals 1DResiduals TEstimates TResiduals Validation 14 Pretreat ector of used wavelengths Vector of instrument serial numbers Vector of instrument wavelength calibrations Matrix of spectral means for each data set Matrix of variance for each data set Vector of used wavelengths Vector of references to used samples Vector of sample 1D s Average Target Vector of Target Vector of mean spectra Matrix of spectras Matrix of projected X values Matrix of projected Y values Matrix of regression vectors 2 Vectors representing residuum for calibration and validation Estimated Y values Zero levels Estimated Target values Residuals of estimates Estimates over time Residuals over time In this group you find above entries for your validation data 7 6 Results By selecting Compute from the main menu the results of the computation is shown in the report window This report discloses information about used data statistics and possible abnormalities 7 7 Graphs By selecting various matrices vectors in the project the data is
9. Project results 7 5 1 Pre treat 7 5 2 Modell 7 6 Results 7 7 Graphs 7 8 Export your calibration 7 9 Settings 8 Using the calibration 9 Calibration maintenance The following symbols are used in these instructions indicates required steps indicates steps required only under certain conditions describes what happens after you have performed a particular step indicates an item in a list indicates hazard 1 Revision history 1 1 V1 4 0 Valid for software V1 5 0 or later Projects must now be named Parameter prj e g Mix Protein 10 15 prj a warning will be displayed if name doesn t follow this convention The project name will automatically change when you select a parameter and start the computation Clicking a File will automatically add selectors based on values in field Recipe Clicking a file will automatically add existing parameters and their range AutoDelete function has been moved into project and overrides any settings in the Options Format is Calibration Mahalanobis Validation Calibration of classes using now automatically creates a usable calibration file upon export You may now denote files as 1 2 3 and 4 This will load the exported journal files from the respective instrument Any field named Recepies will be renamed to Recipes The AutoDelete function did sometimes not work properly problem fixed License entering dialogue is now visible in the task b
10. able in the range of 10 to 20 Output is formatted using 2 decimal digits followed by a percent sign Ash 0 3 0 5 0 5 0 9 0 000 Target is defined by the field Ash in your data table The calibration defines two zero levels one for samples between 0 3 and 0 5 and one for samples between 0 5 and 0 9 Output is formatted with 3 decimal digits Class 0 4 Flour Wheat Durum The field Class in your data table is used to define targets Flour will be represented by a numerical one Wheat by 2 and Durum by 3 Class 0 4 Flour Wheat Durum The field Class in your data table is used to define targets Flour will be represented by a numerical one Wheat and Durum by 2 Note Range limits must be formatted using a decimal not comma independent of regional settings 7 4 3 Standard pre treatment The elements of a diode array are statically aligned along the wavelength range of the spectrometer Therefore it is required to recompute the absorbencies determined by measurement to a fixed scheme and allow combination of data from several units and transfer of a model from one equipment to another This pre treatment is mandatory Example Spline2 1 2 3 1050 1750 10 Recalculates the measured spectrum to a vector representing the absorbencies from 1050 to 1750nm with a spacing of 10 nanometres 7 4 4 Special pre treatments Mathematical pre treatments are used to reformat reshape the data These can
11. ar A zoomed Wavelength graph now has the capability to display component absorbance lines Components can be modified in file Bands txt located in the install folder Upon export and overwriting of a calibration file only year month day is used this means that only one calibration per day is traceable Traceable file is only created ones In the Mahalanobis Residium and Spectral graph the sample reference rather than 1D is used when a item is selected in the graph Removed the need to right click the graph to get selected objects into the clipboard 1 2 V1 3 9 Relevant for SX Plus V1 4 3 or later Corrected graph of Residuals Cross validation option is now stored in the project A Project using option validation or cross validation will export without checking number of factors Model factors are defined by minimum validation or cross validation error Cross validation can be maximized for large data sets resulting in leaving out more than one sample for each cross validation E g setting 30 creates 30 sub sets for validation sample ID is persistent across validation groups Fixed condition when saving Project to new location and creating local copy of data Default graph using cross validation is CVEstimates When saveing a project matrix dialog title has been changed to Save Matrix Project is now blocked for changes when computing the model Incomplete editing of project settings when starting the
12. calibration process are automatically completed 1 3 V1 3 8 Relevant for SX Plus V1 4 0 or later Corrected import of ISI exported data Chapter 7 4 2 Note on range limit Note to check Bias removed 1 4 V1 3 7 Relevant for SX Plus V1 3 124 or later Correction of table of content 7 4 2 Corrected missing semicolons 7 4 4 Added OSC EMSC X2 X0 5 7 4 6 Added option 9 3 2 Introduction Prior to using NIR it is necessary to create a calibration this calibration re computes the measured spectrum into a desired property The creation of a calibration model is not equal to the adjustment of intercept and slope though this i soften referred to as calibrating the unit NIR is a secondary method thus it in almost all cases require prior knowledge of the composition of the samples measured in the calibration development phase To create a calibration model the samples are anlaysed using the NIR spectrometer as well as applicable primariy methods In most cases a calibration is used to determine a concentration property e g Moisture or Protein In some cases the sample is classified as Type Good or Bad To develop a calibration the samples need to cover all future expected variations not only in the target properties but also other such as sample temperature or particle size 3 NIR Applying NIR spectroscopy is based upon physical fact that causes energy of certain wavelength to interact with atom to atom bonds The
13. down box with pre defined entries 7 3 Files 7 3 1 Adding data files By double clicking Er a new entry is created By double clicking an entry you may select a file to use By single click on an entry the file will be displayed in the Table window 7 4 Calculation parameters There are a number of settings that control the way your date is computed The following are the minimum used 1 Query Defines the mask used to query the data in your data tables 2 Y Defines the target parameter 3 Pretreat1 Basic pretreatment 4 Pretreat Optional further pretreatments 5 Type Regression method 6 Validate Samples used to validate your model 7 4 1 Sample selector The selection of samples is defined by a query A field in each Table is compared to a mask The following examples apply Recipe Salt selects all samples that have been measured with the Recipe Salt Recipe selects all spectra s Recipe anr selects all spectra s measured with Recipes containing an a n Or T 7 4 2 Selecting and formatting the target parameter The target parameter can be defined and formatted in the following way TD Protein 0 100 0 0 Target values are read by comparing the data tables ID field with your ID List values in the field Protein of your ID list defines the target Output is formatted with one decimal digit followed by a percent sign Moisture 10 20 0 00 Targets are selected from your data t
14. ocedural way by which the equation is determined 4 3 1 MLR Abbreviation for Multiple Linear Regres sion Is the most common method of solving equations where you have many variables that in a linear combination represent the target Various methods are use depending on numerical and stability properties of your data SVD Singualr value decomposition or RR Ridge Regression 4 3 2 PCR Abbreviation for Principal Component Regression This method is a combination of data reduction and MLR where by the data is approximated by a fewer number of variables that are later used to solve the equation 4 3 3 PLS Abbreviation for Partial Least Squares is the most common method applied to calculate calibration models This method is similar to PCR though the eigenvectors are calculated to maximize the covariance to the target target residuum 4 3 4 Other methods There is a long list of other methods used in special cases where MLR PCR PLS fails to produce a satisfactory result Examples ANN Artificial Neural Networks LWR Locally weighted regression SIMCA Soft independant model of classes specialized PCR 4 3 5 Standard deviation errors Standard deviations are often used to determine the quality of a model Given a normal distribution the value represents the spread of 68 of the population examined 4 3 6 Colinearity Colinearity can be viewed as redundancy in the data where many of the observed
15. se bonds are brought into vibration causing selective absorption along the wavelength axis The amount of energy converted is proportional to the concentration of this molecule bond Thereby a measurement of the chemical composition is possible Since a molecule is build of a plurality of bonds it is necessary to measure more than one wavelength to analyse the sample These molecules are also subject to measurable changes due to e g temperature Measurement in different wavelength regions causes different pros and cons Measurement above 2 5um causes the bands to be well separated and a relatively small sample set may be used for calibration the con is that the penetration depth of these wavelengths is low and that the sample needs to be prepared prior to measurement by extraction using a solvent or grinding the sample this makes long wavelength non suitable for on line analysis The shorter wavelengths suffer from more interference requiring a larger sample set for training the pro is that the measurement may be conducted without sample preparation suitable for online analysis 4 Terms amp Abbreviations 4 1 Chemometrics Is an expression often used in conjunction with NIR It signifies the interpretation of spectral data to determine chemical composition with the use of mathematical methods 4 2 Regression Is a mathematical term for solving an equation for its un knowns 4 3 Regression methods The mathematical or pr
16. will automatically select significant wavelengths 68 Relevance for each factor E g PLS or RR a double plus E g PLS will select 95 relevance a tripple plus is not supported 7 4 6 Validate With this setting you determine the samples used to optimize validate the calculation during model development 2 Selects every second spectra for validation 4 2 Selects 2 samples in each group of four samples for validation 9 3 Selects the last 3 samples in a group of nine for validation 10 Selects the last 10 spectra for validation 7 4 7 Cross validation Determine the number of sections to divide the calibration set into where each section is left out once in order to determine statistics for these left out samples during regression 7 4 8 Auto Delete Determine what samples to remove from the calibration and validation data 3 25 4 Removes any samples from the calibration data with a standardized residual larger than 3 removes any sample from the calibration or validation data with a Mahalanobis distance larger than 25 and any samples in the validation data with a standardized residual larger than 4 Note When using Automatic recalibration these deletes are non persistent thus re evaluation will occur upon each re calibration 7 4 9 Factors Determine the number of factors for your model Leaving this option unchecked during calculation will lead the software to optimize the number of factors
17. ze sartorius User Manual Sartorius PMD500 SX Plus V1 4 0 Software Program me DEDE LIE LETT 98646 003 07 Contents 1 Revision history 1 1 V1 4 0 1 2 V1 3 9 1 3 V1 3 8 1 4 V1 3 7 PPWWW e 2 Introduction e 3 NIR 4 Terms amp Abbreviations 4 1 Chemometrics 4 2 Regression 4 3 Regression methods 4 3 1 MLR 4 3 2 PCR 4 3 3 PLS 4 3 4 Other methods 4 3 5 Standard deviation errors 4 3 6 Colinearity 4 3 7 Interference 4 3 8 Pre treatment 4 3 9 Report abrevations Ooccoo000000000 eo 5 Questions amp Answers 5 1 How accurate can measure 5 2 How many samples do 1 need to make a calibration 8 5 Transfer a calibration from one equipment to another 8 5 4 What to do when it doesn t work co co 9 6 Collecting data 9 6 1 Collecting data using a separate ID List 9 6 2 Collecting data using SX Center Journal 10 10 10 10 10 10 11 11 11 11 11 12 12 13 13 13 13 13 13 14 14 14 14 14 14 15 15 e O gt A 7 SX Plus 7 1 Users interface 7 2 Projects 7 2 1 New Project 7 2 2 Saving your Project 7 2 3 Project window 7 3 Files 7 3 1 Adding data files 7 4 Calculation parameters 7 4 1 Sample selector 7 4 2 Selecting and formatting the target parameter 7 4 3 Standard pre treatment 7 4 4 Special pre treatments 7 4 5 Regression methods 7 4 6 Validate 7 4 7 Cross validation 7 4 8 Auto Delete 7 4 9 Factors 7 5
Download Pdf Manuals
Related Search