Applying artificial intelligence in process industry
A key contributor to better decisions is to be better informed. Better utilization of data has facilitated clear improvements in industrial Overall Equipment Effectiveness (OEE), e.g. in quality control, production optimization, and predictive maintenance. However, analysis results based only on measurement data are strongly dependent on the quality of available data. Practical data sets cannot cover all possible scenarios of even moderately large systems, especially as product portfolios are updated and processes are upgraded. The mere number of potential factors causing performance degradation is too large to allow training of computational models from measurement data alone. The state of the art is strongly focused on analysis of measurement data, mostly ignoring knowledge of physics and chemistry relevant to the phenomena being modelled.
Enhancing performance, quality and life time
VTT combines data analytics with understanding the physics and chemistry of industrial phenomena affecting process performance, product quality, and component life times. More of the available information is utilized than in mere data-driven approach. Even though the data-driven approach allows relatively easy identification of models describing the available data set reasonably well, such models can be quite unreliable in describing situations not covered by the available data set. Including knowledge of the physical and chemical phenomena improves the reliability of the models especially in extrapolation to new process conditions and in predictive analytics.
Combining data-driven and first principles based elements in computational models balances the relative advantages of these modelling approaches with the business goals of each modelling activity. Adequate model accuracy and reliability are reached with constrained modelling resources.