Generating process feedback from heterogeneous data sources in quality control
The goal of this three-years project under the scope of "IKT of the Future" funding line by FFG is to develop machine learning methods based on system identification, time-series models, causal relations and classifiers that merge all the heterogeneous data sources to automatically determine potential cause(s) for defects.
Additionally, changes in the process can be detected and monitored, before they actually
cause defects. This requires research on the following topics:
- Fusion concepts for appropriately handling multiple data sources and types (visual, design and process) in the subsequent stages.
- Feature extraction, data-driven system identification and predictive modelling to establish correlations (and – where possible - causal relations) over several steps in a multi-stage production process.
Early recognition of faults and prognostics of (upcoming) quality trends on the basis of identified models to initiate corrective actions to avoid the production of defective parts.
The results of the project will be methods for automatic data analysis implemented in
software that provide a set for tools for analysing heterogeneous data coming from
production processes and quality control, with the goal of generating short-term feedback to the process and to reduce the time needed for setting up new processes. The methods will be tested on 3 use cases from the production of micro-fluidic components with a real multi-stage character, but will aim at generic methods with a wide range of applications.