Practices for NMR Multivariate Quantitative Analysis
(Ref: ASTM Standard E1655-00)
The following practices describe the development of the multivariate NMR calibrations used to predict the chemical and physical properties of materials. These practices are applicable to the development of calibrations on any NMR instrument regardless of magnet field strength. The practices stated are encompassed by standard E1655-00 which although centered on Infrared spectroscopy is applicable to the NMR calibration development.
Procedures for collecting and treating data for developing NMR calibrations will be outlined. Also, the following summary of the calibration process can be given:
- Multivariate mathematics is applied to correlate the relative intensity and position of NMR resonance peaks measured for a set of calibration samples to validated laboratory data of chemical and physical properties. The resulting multivariate calibration models can be applied to NMR spectra of unknown samples to provide an estimate of the chemical and/or property values for the unknown sample.
- The multivariate mathematics referes to techniques such as multilinear regression (MLR), principal components regression (PCR), and partial least squares (PLS) which are used to develop the calibration models.
- The advatage of these techniques is that statistical tests can be applied to detect outliers during the development of the calibration model. Outliers include high leverage samples (samples whose spectra contribute a statistically significant fraction of one or more of the spectral variables used in the model), and samples whose reference values are inconsistent with the model.
- Validation of the calibration model is performed by using the model to analyse a set of validation samples and statistically comparing the estimates for the validation samples to reference values measured for these samples. This allows test for bias in the model and for agreement of the model with the reference method.
Referenced Documents:
D 1265 Practice for Sampling Liquified Petroleum (LP) Gases (Manual Method)
D 4057 Practice for Manual Sampling of Petroleum and Petroleum Products
D 4177 Practice for Automatic Sampling of Petroleum and Petroleum Products
D 4855 Practices for Comparing Test Methods
D 6122 Practice for Validation of Multivariate Process Infrared Spectrophotometers
D 6299 Practice for Applying Statistical Quality Assurance Techniques to Evaluate Analytical Measurement System Performance
D 6300 Practice for Determination of Precision and Bias Data for Use in Test Methods for Petroleum Products and Lubricants
E 386 Standard Practice for Data Presentation Relating to High-Resolution Nuclear Magnetic Resonance (NMR) Spectroscopy
E 131 Terminology Relating to Molecular Spectroscopy
E 456 Terminology Relating to Quality and Statistics
E 691 Practice for Conducting an Interlaboratory Study to Determine the Precision of a Test Method.
NMR Multivariate Calibration:
1) Selection of a calibrations set: see 1655, Section 17.
2) Samples from petrochemical processes can degrade with time unless careful handling and sample storage procedures are followed. For successful calibration development and calibration validation it is essential that the sample analysed in the laboratory to generate the reference data be representative of the sample analysed by the NMR to produce the spectrum to be correlated to the reference value. Sampling should be done in accordance with methods such as D1265, D4057 or D4177, whichever is applicable. Sample storage for extended periods before the test methods and NMR experiments are performed should be avoided.
3) When a calibration set has been determined all samples must be tested accurately to determine their physical and chemical properties. For complex mixtures typical to the petrochemical industry the physical and chemical property values must be determined by a reference analytical method. The accuracy of a multivariate model prediction will therefore be highly dependent on the accuracy and precision of the primary reference test method used in building the calibration. Biases and error in the test method will be present in the calibration by definition. The expected agreement between the NMR predicted value and the value from a single test method measurement is limited to the precision (repeatability) of the reference values. The precision of the reference data can be improved by using the average value of several repeated test method measurements on the same sample.
4) If reference values come from a single laboratory the ASTM test methods should be monitored for bias and precision by following quality assurance procedures presented in ASTM procedure D6299.
5) When performing the primary test methods one should keep the following ideas taken directly from E1655 Section 10 in mind:
“If the reference method used to obtain reference values for the multivariate calibration is an established ASTM method, then repeatability and reproducibility data are included in the method. In this case, it is only necessary to demonstrate that the reference measurement is being practiced in accordance with the procedure described in the method, and that the repeatability obtained is statistically comparable to that published in the method. Data from established quality control procedures can be used to demonstrate that the repeatability of the reference method is within ASTM specifications. If such data is not available, then repeatability data should be collected on at least three of the samples that are to be used in the calibration. These samples should be chosen to span the range of values over which the calibration is to be developed, one sample having a reference value in the bottom third of the range, one sample having a value in the middle third of the range, and one sample having a value in the upper third of the range. At least six reference measurements should be made on each sample. The standard deviation among the measurements should be calculated and compared to that expected based on the published repeatability. If the reference method to be used for the multivariate calibration is an established ASTM method, and the samples to be used in the calibration have been analyzed by a cooperative testing program (for example, octane values obtained from recognized exchange groups), then the reference values obtained by the cooperative testing program can be used directly, and the standard deviations established by the cooperative testing program can be used as the estimate of the precision of the reference data.
Reference methods that are not ASTM methods can be used for the multivariate calibration of infrared analyses, but in this case, it is the responsibility of the method developer to establish the precision of the reference method using procedures similar to those detailed in Practice E 691 and in Practice D 6300.
When multiple reference measurements are made on an individual calibration or validation sample, a Dixon’s Test (see D3764 Appendix X1.1) should be applied to the values to determine if all of the reference values came from the same population, or if one or more of the values is suspect and should be rejected.
6) NMR Experiments and Pre-Processing of Data: NMR experiments are performed following the practices describes in Practice E 386. All NMR settings such as optimum RF pulse width, lock channel optimisation, gain settings are set upon installation and unless there is catastrophic failure of an RF component they will not be adjusted again for that application. An 8 pulse averaged NMR free-induction-decay (FID) is acquired over a 4 kHz spectral window with an appropriate relaxation delay between pulses. The following post processing procedures are performed as the spectra are referenced and co-added to provide the final NMR spectrum:
- Remove dc
- Zero fill to 16k (resolution enhancement)
- FFT
- Symmetric autophase (proprietary algorithm)
- Remove dc
- Reference appropriate resonances
- Co-add spectra together
- Integrate spectrum every 0.1 ppm from 12 to –2 ppm and normalize sum of integrals to 100.
- Save “integral spectrum”
The “integral spectrum” represents the NMR spectral data that is regressed against the primary reference data to produce the calibrations. Before the data is regressed the NMR data is mean-centered. This is a prodedure where the average spectrum of the calibration set is subtracted from each of the individual spectra.
7) The multivariate mathematics that are used to generate the calibration models are described in Practice E 1655 Section 12, and the references therein.
8) The estimation of values from NMR spectra is described in Practice E 1655 Section 13, and the references therein.
9) Statistics used in evaluating and optimising calibration models is described in Practice E 1655, Section 15, and the references therein.
10) Outlier diagnostics and their use in improving calibrations is described in Practice
E 1655, Section 15, and the references therein.
11) Validation of a multivariate model is described in Practice E 1655, Section 18.
12) Precision of NMR estimated values can be calculated using the procedure described in E1655 Section 19. Replace all references to “Infrared” with “NMR”.
Upon initial installation laboratory NMR based models will be transferred to the on-line NMR system. After initial start up and calibration of the NMR a total of 10-20 samples will be required to further train and expand the NMR model base. After the original models have been updated with these new calibration samples NMR models will be ready for validation.
Implementation of On-Line Calibration Models and Model Validation for On-Line Process NMR Systems (ref: D3764, D6122, D6299, E1655)
Preconditions: Certain conditions shall be met before the on-line system validation is begun:
- Analyser has been installed in accordance with Foxboro NMR’s instruction.
- Multivariate NMR based calibration models (PLS) have been developed and validated according to methods described in Practice E1655.
- NMR spectrometer calibration has been performed at installation (Toluene Standard).
- A quality assurance program for the primary methods has been put in place in order to determine the precision and bias of the primary method. This quality assurance program should include following Practice D3764 to etablish the precision and bias of the laboratory tests. This procedure should be followed for 9 samples (3 from the bottom, 3 from the middle and three from the upper end of the range of reference values). Once the laboratory is shown to be performing to ASTM standards the procedure can be applied periodically as a check of continued laboratory conformance.
Collection of Line Samples: Withdraw samples in accordance with accepted sampling methods as given by Practices, D 1265, D 4057, and D 4177, whichever is applicable. The intent of the practice is to obtain a sample that corresponds directly to the spectrum generated by the on-line NMR analyser at that time. Sample should be collected at a point close to the NMR and at a time correlated to the NMR acquisition. For a more detailed description of the influence of lag time between sample collection and NMR measurement see Practice D3764. D 3764 also describes when samples should be grabbed during steady state operation.
Procedure for establishing reference sample value: For samples to be analysed multiple times to determine laboratory test bias and precision see D3764 Section 7.
Model Verification – Outlier Response: Outlier diagnostics should be applied to the initial process stream NMR spectra to determine that the model is applicable to the process stream sample. See Practice D 6122, Section 11 for details. If spectrum is found to be an outlier one should perform a standard sample (toluene) test to determine if there is a spectrometer problem. If the standard test is passed then the sample may represent a composition that is not currently found in the models. The models should be updated to include the sample in the calibration set. If the sample is readily included without flagging as an outlier a second sample should be predicted with the updated model to determine if the sample is an outlier. If it is not one should proceed to the initial validation procedure.
Initial Validation Procedure: The validation test shall follow exactly the procedures described in Practice D6122, Section 12.
Periodic Validation by Plotting Control Charts of the Differences Between NMR Predictions and Laboratory Methods: If the analyser passes the initial validation the stability of the differences between the NMR predictions and the lab results should be monitored using control charts. Details of the control charts are given in D6299 and D6122 Section 13. See also D6122, Section 9.5 for a detailed description of periodic validation tests.
Control Limit Updates: Continued monitoring of the agreement between NMR predictions and Lab values should be performed after the production of 20 lab values. Significant biases and variance should be monitored. If biases do appear then the root cause must be determined. Both the NMR and the laboratory should be checked. The laboratory should at all times be monitoring its conformance to ASTM repeatability/reproducibility through an established quality assurance program.
Online Analyzer Repeatability: can be determined following D6122, Section 15.
Model Updating: See E 1655 Section 24. If models are updated with additional calibration samples the outlier procedures described in E1655 must be applied to the updated models and the models must be revalidated.
Performance Tests: NMR does not require regular performance tests. If validation procedures indicate that outlier or “NMR-lab difference” limits are not being met then standard performance test should be performed using toluene. Another possibility is the use of a known process sample standard. The definition of the process sample standard used is to be determined by the customer. The sample should be similar in nature to the samples being analysed by the analyzer.