Fault Diagnosis and Prognosis With Inferential Sensors: A Hybrid Approach Integrating Symbolic Regression and Information Theory

We propose a fault diagnosis approach that integrates symbolic regression and information theory to optimize inferential sensors, which, along with traditional detection techniques and hard sensors, contribute to a comprehensive fault detection and diagnosis algorithm. Additionally, we introduce an advanced predictive scheme to estimate time-dependent fault evolution functions that enable us to compute the remaining useful life (RUL) of the system. Our methodology employs symbolic regression for parameter trending via a genetic programming algorithm, integrating the system model to discover the fault’s dependence on time. This relation analytically describes the fault progression, providing valuable insights into predicting system health. For comparison, we apply a dynamic degradation-based regression model using the inferential sensors as health indicators. The effectiveness, robustness and superiority of the proposed diagnosis and prognosis methods over traditional power function-based degradation models are demonstrated through experiments on a dynamic cross-flow plate-fin heat exchanger system with varying levels of fault progression, measurement noise, and uncertainty.