Modern spectrometers provide a large amount of data. A maximum of relevant information can only be obtained from the measurement data through efficient data analysis. Fraunhofer IPM uses chemometric methods to extract and enhance chemical information from experimental data.
Methods such as multivariate linear regression, principal component analysis or support-vector-machines help to correlate the spectral data with chemical quantities like concentrations or to identify unknown substances. The algorithms must be trained based on measurement data (or artificially generated "measurement data") of known pure substances and mixtures in order to be able to deliver the relevant process variables within the production environment. In addition to the application of mathematical algorithms, it is crucial to understand the production process and the measurement conditions. In many casesit is then possible to improve the data basis prior to mathematical treatment e.g. by clever reference measurements, focusing on certain spectral ranges, or applying means of noise suppression in such a way that the chemometric methods themselves become rather secondary in the evaluation and simple methods such as linear regression become sufficient to determine the process variables. Thereby, traceability of the results and a significantly lower risk of error in the algorithms are ensured.
Fraunhofer IPM has a lot of experience in the evaluation and presentation of extensive measurement data and the development of algorithms for the error tolerant determination of essential process parameters. In addition to data evaluation and the development of chemometric algorithms, our range of services also includes the optional coding into PC software or the implementation of algorithms into an FPGA.