Prognostics and Health Monitoring for ECU Based on Piezoresistive Sensor Measurements

This dissertation presents a new approach to prognostics and health monitoring for automotive applications using a piezoresistive silicon stress sensor. The stress sensor is a component with promising performance for monitoring the condition of an electronic system, as it is able to measure stress values that can be directly related to the damage sustained by the system. The primary challenge in this study is to apply a stress sensor to system-level monitoring. To achieve this goal, this study firstly evaluates the uncertainties of measurement conducted with the sensor, and then the study develops a reliable solution for gathering data with a large number of sensors. After overcoming these preliminary challenges, the study forms a framework for monitoring an electronic system with a piezoresistive stress sensor. Following this, an approach to prognostics and health monitoring involving this sensor is established. Specifically, the study chooses to use a fusion approach, which includes both model-based and data-driven approaches to prognostics; such an approach minimizes the drawbacks of using these methods separately. As the first step, the physics of failure model for the investigated product is established. The process of physics of failure model development is supported by a detailed numerical analysis of the investigated product under both active and passive thermal loading. Accurate FEM modeling provides valuable insight into the product behavior and enables quantitative evaluation of loads acting in the considered design elements. Then, a real-time monitoring of the investigated product under given loading conditions is realized to enable the system to estimate the remaining useful life based on the existing model. However, the load in the design element may abruptly change when delamination occurs. A developed data-driven approach focuses on delamination detection based on a monitoring signal. The data driven methodology utilizes statistical pattern recognition methods in order to ensure damage detection in an automatic and reliable manner. Finally, a way to combine the developed physics-of-failure and data-driven approaches is proposed, thus creating fusion approach to prognostics and health monitoring based on piezoresistive stress sensor measurements