Intelligent state assessment of molded case circuit breaker based on multi-source sensor data fusion and machine learning

Molded case circuit breaker (MCCB) is widely used in the power system. With the increase in the number of current-breaking, the contacts of the MCCB will gradually deteriorate until failure, severely limiting the electrical life and the breaking reliability. Precise identification and assessment of the contact ablation state were required. Firstly, the travel and terminal voltage were selected as multi-source sensing quantities. The accelerated ablation experiment platform was established to obtain the real-time degradation experimental data of the circuit breaker in its whole life cycle. The action process of the contact was observed and analyzed with a high-speed camera. Then, four electromechanical hybrid features were extracted from multi-source sensor data. A novel multi-source feature fusion model was designed based on the unsupervised machine learning of the stacked auto-encoder, and the contact health index was successfully obtained, which showed the unique ‘three-stage decline’ trend. Furthermore, an assessment strategy was set, and the contact ablation states were divided into four states: ‘Healthy’, ‘Sub-Healthy’, ‘Abnormal’, and ‘Dangerous’. The experimental results show that the proposed method can effectively identify the contact ablation state of the MCCB and has strong universality, demonstrating the significant engineering application value.