Contact Line Insulator Detection and Washing Method Based on Multi-Sensor Perception

In response to the high labor costs and low level of automation faced by current contact network insulator cleaning operations, this paper innovatively proposes an integrated method for insulator target detection and flushing based on multi-sensor fusion. The method mainly consists of a target detection part and a tracking control part. For object detection, we propose an improved BEVFusion-based multi-modal data fusion algorithm. This approach effectively fuses lidar features and camera features projected onto the Bird's Eye View (BEV) through an optimized feature integration strategy, achieving higher fusion accuracy. The method enables complementary advantages of different sensors while significantly enhancing the perception and recognition capability for target insulators. To further improve detection accuracy, this paper designs a custom neural network joint calibration model, which can effectively enhance the calibration accuracy of each sensor and reduce the cumulative error in the calibration process. On the control level, a composite Model Predictive Control (MPC) tracking flushing control strategy is proposed in line, which comprehensively controls the position and speed status of the servo motor to achieve precise and stable control of the flushing water cannon. Finally, a contact network insulator water flushing robot platform is built for experimental verification. The experimental results show that the method can achieve precise detection and stable tracking flushing control of the target insulator. More importantly, the method has all-weather operation capabilities, which can significantly improve the efficiency and level of automation of cleaning operations, showing a broad application prospect.