基于物理与统计特征融合的模拟电路故障检测方法OA

To address the need for high-precision circuit fault detection in embedded platforms,an embedded fault diagnosis and protection system integrating simulation program with integrated circuit emphasis(SPICE)physical simulation model with random forest machine learning model is proposed.In this circuit,signals from the circuit under testing are captured in real-time by means of multi-node voltage sampling circuit.The lightweight SPICE simulation and parallel inference of random forest are realized based an STM32G473 microcontroller.In the SPICE model,the modified Ebers-Moll1 model is used to incorporate secondary physical effects such as base charge storage,Early effect,and high-level injection effects,and the matching degree between simulated and measured voltages is calculated by means of weighted Euclidean distance to realize the initial screening for abnormal states.The random forest model is trained by means of NanoEdge AI,multi-node time domain,frequency domain and time-frequency domain features are analyzed to suppress noise interference and realize accurate fault location.When there are conflicts in the output results of the two models,the system can dynamically adjust weights according to SPICE matching degree and improve the overall diagnostic accuracy by means of the weighted probability fusion strategy.The experimental results show that,in the testing of single-stage amplifier circuits,the identification accuracy of this system for power disconnection,normal operation and three typical short-circuit faults ranges from 93%to 100%,and the average power-off response time is less than 342 ms.It provides a high-precision and low-latency fault diagnosis solution for embedded scenarios.