Research and Development of Offline Circuit Breaker Status Detection Based on Deep Learning and ONNX

Avionics testing constitutes a pivotal phase in aircraft production, wherein circuit breakers serve an essential function in the assessment of various subsystems. These components enable the verification of onboard logic functions and critical performance indicators against design specifications. Traditionally, manual visual inspection has been employed for this purpose; however, this approach is both time-intensive and susceptible to errors. To address these limitations, this study proposes a novel lightweight two-stage deep learning detection method. In the first stage, the DSAM-YOLOv8 model is utilized to rapidly localize switch areas. Subsequently, the LSMA-MobieNetV3 model is applied in the second stage to perform fine-grained classification of switch states. This methodology enhances the model’s capability to discern subtle features of switch states while maintaining a low computational overhead. To facilitate operation in environments devoid of mobile network connectivity, the trained deep learning models are converted and optimized via the ONNX (Open Neural Network Exchange) framework. This enables offline deployment and real-time detection on resource-constrained Android devices. Experimental results demonstrate that the proposed method achieves an accuracy exceeding 99% in real-world circuit breaker state detection. Furthermore, it delivers immediate results on mobile terminals, with inference times constrained to under 2 s per operation, thereby substantially enhancing on-site operational efficiency. This research presents an efficient and accurate automated solution for circuit breaker state detection, offering significant improvements over conventional methods.