H. Gerdes, M. Mahdhaoui, S. Singh, D. Barton, R. Bandorf, Fraunhofer IST, Braunschweig, Germany
In contemporary efficiency-driven industrial environments, rigorous process validation is indispensable. Exclusive reliance on human operators for end-to-end process monitoring may be viable in isolated scenarios, but it is labor-intensive, strongly dependent on individual expertise, costly, and ultimately constrains scalability. Transitioning from predominantly manual observation to largely automated validation mitigates operational risk, reduces expenses, and enables operators to concentrate on systematic process improvement rather than continuous routine oversight.
Advances in data acquisition, processing, and storage, combined with the widespread availability of machine learning frameworks, now permit the cost-effective deployment of relatively simple models for fault detection in equipment such as coating systems. Trained on historical operational data, these lightweight algorithms characterize normal system behavior and identify deviations indicative of potential faults.
This presentation provides a concise overview of data acquisition strategies, introduces the training and validation of machine learning models for anomaly detection, and illustrates their use in alerting human operators to potentially erroneous system states. In addition, it addresses the application of machine learning techniques for optimizing process input parameters.