An ANN-based Digital Twin for PCB Temperature Monitoring: Development and Validation

In recent years, digital twins have been gaining growing attention among researchers across academia and industry, as they can provide information about physical assets that is often unattainable through traditional methods, due to physical or budgetary limitations. In microelectronics, this technology can be implemented to provide information useful for control, maintenance, and optimization activities. The present paper describes an early study of a digital twin designed to forecast the temperature of a simplified prototype PCB, equipped with resistive heat sources, operating under varying electrical loads and changing forced-air cooling conditions. The prototype was intentionally designed and operated to provide a fully controllable means of mimicking the thermal behavior of a generic automotive electronic control unit. The developed digital twin was based on neural networks, specifically a long short-term memory network. It was further optimized to improve prediction accuracy and limit its size and the number of input features necessary for successful forecasting of the outputs. The trained and optimized digital twin was evaluated in a number of test scenarios, yielding a mean absolute error below 2 K across all predictions. Additionally, the same optimized network architecture required the second-smallest number of neurons in the input and hidden layers among all the investigated alternatives. The results obtained in this study pave the way for industry-ready implementation of a digital twin enabling thermal monitoring of a PCB assembly operating in the automotive environment.