Recent trends in automotive electronics such as automated driving will increase the number and complexity of electronics used in safety-relevant applications. Applications in logistics or ridesharing will require a specific year of service rather than the conventional mileage usage. Reliable operations of the electronic systems must be assured at all times, regardless of the usage condition. A more dynamic and on-demand way of assuring the system availability will have to be developed. This article proposes a thermomechanical stress-based prognostics method as a potential solution. The goal is achieved by several novel advancements. On the experimental front, a key microelectronics package is developed to directly apply the prognostics and health management concept using a piezoresistive silicon-based stress sensor. Additional hardware for safe and secure data transmission and data processing is also developed, which is critically required for recording in situ and real-time data. On the data management front, proper data-driven approaches have to be identified to handle the unique dataset from the stress sensor employed in this study. The approaches effectively handle the massive amount of data that reveals the important information and automation of the prognostic process and thus to be able to detect, classify, locate, and predict the failure. The statistical techniques for diagnostics and the machine learning algorithms for health assessment and prognostics are also determined to implement the approaches in a simple, fast, but accurate way within the capacity of limited computing power. The proposed prognostics approach is implemented with actual microelectronics packages subjected to harsh accelerated testing conditions. The results corroborate the validity of the proposed prognostics approach.
- electronic packages
- machine learning (ML)
- piezoresistive stress sensor
- prognostics and health management
- recurrent neural network (RNN)