Fault Detection in Engine Measurement Systems by a Model-Based Approach

Measurement systems are becoming more complex and test beds are usually automated; high measurement accuracy is also required. However it is common that measurement failures are only detected in the post-processing, resulting in important time and economic loss. Due to the huge amount of sensor signals, the online validation of the data is very time-consuming and infeasible without computer aid. In this study a failure detection framework is used for data plausibility analysis. This failure detection methodology is able to deal with generic models relating different measure channels. The general approach incorporates model and sensor inaccuracies in the evaluation procedure. Additionally, a useful set of physical equations applicable for failure detection in engine test bends is presented. These equations are combined with data-driven models allowing satisfactory detection rates while maintaining a low rate of false alarms. Validation results considering real-life data coming from engine test beds are included.