Fault Detection and Prognostics Framework for Technical Systems Subjected to Time-Varying Operating Conditions

Time-varying operating conditions can obscure condition monitoring signals, either causing changes in operating conditions to be misclassified as faults, which leads to false alarms, or causing actual faults to be unidentifiable due to the influence of operating conditions on the condition monitoring signals. False alarms typically result in economic losses, while undetected faults can have serious environmental and human consequences. To enable reliable fault detection and accurate prediction of the remaining useful life (RUL) for technical systems operating under time-varying operating conditions, this work provides a comprehensive framework comprising five distinct strategies to leverage information about such time-varying operating conditions.The proposed methodology is evaluated using experimental data from accelerated run-to-failure tests on piezoelectric bending actuators and ball bearings subjected to time-varying operating conditions. Piezoelectric bending actuators are electromechanical components utilized in numerous household, medical and industrial production applications, whereas ball bearings are widely used mechanical components within more complex technical systems. The evaluation results from both application examples demonstrate that incorporating the operating conditions, for instance either via condition-invariant features or by including the operating conditions as additional inputs to machine learning models, enhances the effectiveness of fault detection and prognostics methods. Although evaluated on only two application examples, this study contributes to advancing the field of predictive maintenance, particularly for technical systems operating under time-varying operating conditions.