Routine spectroscopic analysis generates thousands of data points per sample set. Analytical methods are routinely used in CAPPA in order to get the most information out of a given data set. Multivariate analytical methods are well suited for work with spectroscopic data sets examples of which include principal component analysis (PCA) and multivariate curve resolution (MCR). These are unguided methods that allow for identification of groups within a sample set (PCA) or allow for spectroscopic data taken from mixtures to be separated so that spectral response of the constituents of the mixture can be identified (MCR). Guided multivariate techniques are also used such as partial least squares regression (PLS) which allows for the analysis of the relationship between a physical parameter of a sample and its spectroscopic response. Classification of samples by their spectroscopic response is possible by the application of ‘soft independent modelling of class analogy’ (SIMCA) or through simpler methods such as correlation analysis. Many of the processes mentioned here are also useful in the development of process analytical techniques (PAT) such as those that incorporate machine vision.
Data Analytics in Contamination Analysis
Contamination analysis is carried out regularly at CAPPA and is enhanced by the ongoing research activities of the group. Identification of the contaminant is the priority in every case of contamination analysis. High-resolution imaging as well as chemical and elemental fingerprinting are essential tools for identification of contaminants. High-resolution imaging can reveal valuable information when investigating the origin and identity of a contaminant. CAPPA operate and maintain optical, elemental and chemical imaging microscopes that reveal the material properties of the contaminant. This information is then used to identify the contaminant.
The application of computational techniques to contamination analysis is available at CAPPA. This is necessary in cases where samples are mixtures and it is difficult to ascertain the identity of the contaminants under examination separately. Measurements relating to the chemical fingerprint of a mixed material will produce data where the fingerprint of one or more of the materials are overlapping. The fingerprint of each of the compounds in the mixture may be isolated using computational techniques such as multivariate curve resolution. Once the fingerprint of each material in the mixture is isolated, identification is simplified.
Principal component analysis (PCA) enhances the sensitivity of chemical fingerprinting techniques to changes in materials resulting from contamination or changes in processing conditions. Information calculated using PCA might identify clusters of samples within a set that are related to each other and provide physical data that allows for the interpretation of the observed clustering.
Data Analytics in Process Analysis and Optimisation
The application of multivariate methods to process analysis is a research area of the CAPPA group. The combination of spectroscopic measurement and data analytics provides a powerful tool for modelling, measuring and predicting the physical and chemical parameters of materials used in manufacturing. There are opportunities to improve or replace existing chemical and physical analysis methods with faster, non-destructive methods that are based on the combination of photonic sampling and multivariate analysis. Process analysis systems designed in this way offer online and timely control of process parameters. Online monitoring facilitates cost savings by:
- Optimising material consumption
- Predictive maintenance-early warning for processes drifting out of specification
- Replacing time-consuming and expensive laboratory analysis methods