Machine Learning based Indicators to Enhance Process Monitoring by Pattern Recognition.

In industrial manufacturing, modern high-tech equipment delivers an increasing volume of data, which exceeds the capacities of human observers. Complex data formats like images make the detection of critical events difficult and require pattern recognition, which is beyond the scope of state-of-the-art process monitoring systems. Approaches that bridge the gap between conventional statistical tools and novel machine learning (ML) algorithms are required, but insufficiently studied. We propose a novel framework for ML based indicators combining both concepts by two components: pattern type and intensity. Conventional tools implement the intensity component, while the pattern type accounts for error modes and tailors the indicator to the production environment. In a case-study from semiconductor industry, our framework goes beyond conventional process control and achieves high quality experimental results. Thus, the suggested concept contributes to the integration of ML in real-world process monitoring problems and paves the way to automated decision support in manufacturing.