Model Maintenance
Theory of Operation
Model Maintenance and Customizing a Starter Model to Meet Your Operational Needs
Theory of Operations
A complete online process monitoring system includes an analyzer, sample interface, and fiber optic cables. The instrument also requires calibration for each application. We often provide a starter or factory calibration. Given the variation in process stream composition, it is highly recommended to perform model updates to customize the factory calibration for each installation.
As an example, we have developed Partial Least Squares Regressions using fuel standards to generate starter models (calibrations) for various applications in petrol and diesel refining. When purchased along with a GUIDED WAVE™ NIR-O™ process analyzer spectrometer, these models provide a starting point for analyzer operation.
Best Practices for Model Development will Include the Following:
- Take care to collect the physical sample at a port (location) close to the optical probe and at a time that corresponds to the time that the spectrum is collected and saved. Follow all recommended sampling procedures to maintain sample integrity.
- Sample storage for extended periods is not recommended if there is a likelihood that samples degrade with time. Chemical changes occurring during storage will cause changes in the spectrum, as well as changes in the property or quality parameter measured by the primary method.
- Perform careful data evaluation for suitability to model use. This includes sample spectrum noise level, absorbance patterns, and reference values (laboratory) data.
- Comparison of new data to existing model data to verify that the new data is part of the same general population of samples.
- Evaluation of model results to demonstrate no over-fitting of data. In general, these steps require some time and practice to become proficient. General statistics and mathematics knowledge is helpful. Knowledge of the product’s chemical characteristics can be useful. The ASTM provides two guideline practices to help with the development and validation of Near Infrared (NIR) multivariate calibration models using techniques such as those in a third party calibration software. These are: ASTM E1655- Standard Practices for Infrared Multivariate Quantitative Analysis ASTM D6122 – Standard Practice for Validation of Multivariate Process Infrared Spectrophotometers.
Model Updates Included Free with Purchase
When a starter model is purchased from Process Insights, the terms include three model updates to be performed by us. To access this free benefit, simply send an email with spectra collected with the NIR-O analyzer and the corresponding lab data to us for analysis. Once completed we will email you back an updated model.
How Many Samples are Required to Update a Starter Model?
We recommend a minimum of 20 samples that span the normal operating range of the process. For example, if your refinery is blending petrol and then have an octane value which can range from 87-93, then the 20 or samples need to include samples that span that range. If all 20 samples are for 91-92 octane, the model cannot be truly optimized. We recommend at least 40 samples be included for a final online (production ready) NIR calibration.
If I Have Light Ends in my Product, are there any Special Requirements?
Grab samples for Reid Vapor Pressure (RVP) NIR models do require special attention. If the low molecular weight compounds such as butanes (C4) or alkylates are allowed to boil off before the sample is analyzed there will be a bias in the lab result. That is to say, the lab value will be based on a sample with a different composition than what the NIR analyzer measured. This mismatch can cause an issue during model maintenance. If the bias is large, the collected data cannot be used to optimize the model. Some users have incorporated the use of special airtight containers to ensure that the light ends do not evaporate before the lab analysis is complete.
What does ASTM say about Multivariate Model Maintenance?
The summarized findings of ASTM D6122-10 Standard Practice for Validation of the Performance of Multivariate Online, AtLine, and Laboratory Infrared Spectrometer Based Analyzer Systems are:
1. Contain samples which provide examples of all chemical components which are expected to be present in the samples which are to be analyzed using the model, thereby ensuring that analyses involve interpolation of the model.
2. Contain samples for which the range of variation in the concentrations of the chemical components exceeds the range of variation expected for samples which are to be analyzed using the model, thereby ensuring that analyses involve interpolation of the model.
3. Contain samples for which the concentrations of chemical components are uniformly distributed over their total range of variation.
4. Contain a sufficient number of samples to statistically define the relationships between the spectral variables and the component concentrations or properties to be modeled.
The number of samples that are required to calibrate a multivariate model depends on the complexity of the samples being analyzed. If the samples to be analyzed contain only a few components that vary in concentration, then there will be a small number of spectral variables, and a relatively small calibration set is adequate to define the relationship between the variables and the concentrations or properties. If a larger number of components vary in the samples to be analyzed, then a larger number of calibration samples are required for the model development. When artificially planning a calibration sample population, consider a box-shaped distribution rather than a normal bell-shaped distribution. Using process samples can make it a challenge to have a flat (box) distribution, but this should be kept in mind during data collection.
More Topics
- Why you should use NIR spectroscopy
- NIR Benefits for Online Process Measurements
- How does the Flow Rate, Viscosity, and Pipe Diameter, Effect Accurate (good) NIR Measurements?
- What is pathlength and why is it important when selecting a sample interface?
- Double down on double beam
- Why NIR is better than GC
- Difference between FT-NIR and dispersive dual-beam (DG-NIR) process analyzers
- Benefits of FT-NIR for online process management
- How safety standards in explosive or incendive processes apply to our probes and flow cells
- What is Diffraction Grating?
- Chemometric Calibration Development
- Model Maintenance and Customizing a Starter Model to Meet Your Operational Needs