As automation control systems and field devices become more networked based, clients are being increasingly challenged with how to usefully use the huge amounts of data being generated by the control system. Network protocols like OPCuA, DNP3, and BacNet used by AIEglobal allow the information collected by the control system to be readily transferred to our clients management and corporate levels for archival or analysis.
It is a relatively trivial matter today to use the data in spreadsheets and plot all types of performance graphs, however what is more difficult is to analyse and model the data to determine key performance criteria, and then use that model for predicting current and future outcomes. The science behind this modelling is called “Multivariate Data Analysis”, and has been extensively used by us for NIR Spectroscopy. In simple terms the statistical modelling reduces multiple variables to one variable, thus Big Data becomes Little Data, with the potential of the latter to be extremely relevant and useful.
Multivariate data analysis can be applied to any data set and if the input variables are truly related to the measured outcome then modelling will be possible. As NIR Spectroscopy is an excellent example of using Multivariate Data Analysis to convert Big Data to Little Data to provide “qualitative” and “quantitive” measurements we have included an overview of its use for clarification of the process and its usefulness.
Application of qualitative measures using NIR/MIR Spectroscopy are…..
- Food – able to detect contamination, changes to the process, and missing ingredients
- Raw Materials – detection of incomming materials out of specification
- Waste Water – status determination of activated sludge reactor
The three dimensional graph shows how chemometrics maps the spectral data to three sets of new variables (PC1…PC3). The clouds P1 and P2 represent normal running processes, whilst F1 shows how a faulty process maps to a different region in the principal component space. A vector is generated which represents the distance from the center of the P1 or P2 cloud to the F1 cloud, this vector is then used to raise an alarm when above a specific threshold.
Examples of quantitive parameters measured using NIR/MIR Spectroscopy are…..
- Flour – protein, moisture, particle size, colour, starch damage and water absorption
- Cereal Grains – protein
- Milk (liquid) – fat, protein, lactose, and total solids
- Milk (powder) – moisture, fat, protein, lactose, and ash content
- Beer, Wine, Spirits – Alcohol content, orignal gravity a function of alcohol
- Fresh Fruit – Sugar(Brix), Dry Matter, internal cell breakdown, and internal colour
A validation set is used to quanitify how well a calibration model is working when predicting results for new samples not previously used in the model.