Model-Based Fault Detection in Multi-Sensor Measurement Systems

In the process and manufacturing industries, there has been a large push to produce higher quality products, to reduce product rejection rates, and to satisfy increasingly forceful safety and environmental regulations. Hence, the increasing complexity of measurement systems inside modern industrial processes with a rising amount of actuators and sensors demands automatic fault detection algorithms which can cope with a huge amount of variables and high-frequented dynamic data. Indeed, humans are being able to classify sensor signals by inspecting by-passing data, but this classifications are very time-consuming then and also have deficiencies because of underlying vague expert knowledge consisting of low-dimensional mostly linguistic relationships. In this paper we propose a model-based fault detection algorithm which is generic in the sense, that any model correctly describing a functional dependency inside a system can be enclosed easily almost without adjusting any thresholds or other essential parameters. This advanced 'residual view' fault detection includes aspects for incorporating sensor inaccuracies and model qualities as well as processing further normalized residuals for obtaining fault probabilities. Validation results with respect to data coming from engine test benches are included at the end of the paper.