Chemometric Modeling and Best Practices
Process instruments uses Solo, Chemometric modeling package from Eigenvector Research, Inc. Solo is seamlessly integrated into PROspect for rapid live-time results. PROspect with Solo is able to work with the following model types:
- PLS: partial least squares regression
- PCA: principal components analysis
- PCA: principal components regression
- SECV: standard error of cross-validation
With a conservative choice of factors from SECV table, Solo works to find outliers from leverage verses studentized residuals, where normalization method depends upon sample type. Solo also touts robust cross validations options of both Venitian blinds and only one pass through.
With the help of GLSW filtering, solo can reduce subtle differences between instruments and process streams. GLSW filtering accomplishes its goal by greatly reducing the number of factors used, which improves model transference and model performance.
Normally, one model is sufficient to cover the seasonal blend changes observed in most properties. Creating hierarchical chemometric models that can easily be tailored to smaller seasonal changes can improve the SEP for critical parameters with the following: octane, RVP, Distillation points, etc. The PROspect software delivers accurate results when it automatically runs a particular parameter through the correct target model by using a broader trigger model. The graphic above shows how PROspect is able to choose which model to fit to any given spectra using this state-of-the-art technique.
Process Instruments Inc. has successfully integrated Solo (Eigenvector Research, Inc.) into their own PROspect for the best real-time results with hierarchical modeling available on the market. Modeling services with Process Instruments most experienced engineers is included for 12 months following installation of an instrument. During this period, Process Instruments can train process and lab personnel to confidently handle modeling duties.
After 12 months, Process Instruments can continue to handle modeling on a fee-for-service basis, where typical model turnaround is about 3 business days after receipt of a customer’s data.
For a more clear picture about how Process Instruments’ chemometric modeling could better fit your organizational needs, please contact Process Instruments Inc.