Improved ICNN-LSTM Model Classification Based on Attitude Sensor Data for Hazardous State Assessment of Magnetic Adhesion Climbing Wall Robots

Zhen Ma,He Xu,Jielong Dou,Yi Qin,Xueyu Zhang

2024-12-30

Abstract:Magnetic adhesion tracked climbing robots are widely utilized in high-altitude inspection, welding, and cleaning tasks due to their ability to perform various operations against gravity on vertical or inclined walls. However, during operation, the robot may experience overturning torque caused by its own weight and load, which can lead to the detachment of magnetic plates and subsequently pose safety risks. This paper proposes an improved ICNN-LSTM network classification method based on Micro-Electro-Mechanical Systems (MEMS) attitude sensor data for real-time monitoring and assessment of hazardous states in magnetic adhesion tracked climbing robots. Firstly, a data acquisition strategy for attitude sensors capable of capturing minute vibrations is designed. Secondly, a feature extraction and classification model combining an Improved Convolutional Neural Network (ICNN) with a Long Short-Term Memory (LSTM) network is proposed. Experimental validation demonstrates that the proposed minute vibration sensing method achieves significant results, and the proposed classification model consistently exhibits high accuracy compared to other models. The research findings provide effective technical support for the safe operation of climbing robots

Robotics,Signal Processing,Instrumentation and Detectors