Improved ICNN-LSTM model classification based on accelerometer sensor data for hazardous state assessment of magnetic adhesion climbing wall robots

Abstract

The magnetic adhesive crawler-type climbing wall robot is widely used in high-altitude inspection, welding, and cleaning tasks. However, during operation, the influence of self-weight and payload may generate a flipping moment, leading to detachment of the magnetic pads and consequently posing safety hazards. To address this issue, this paper proposes a data acquisition strategy based on micro-electromechanical system (MEMS) accelerometer sensors, integrated with a deep learning-based classification approach for real-time monitoring of the attachment state recognition of the climbing wall robot and prevention of potential risks. First, a high-precision data acquisition strategy was developed for MEMS accelerometer sensors that is capable of effectively capturing subtle vibration information. Subsequently, an innovative feature extraction and classification model combining adaptive convolutional neural networks (ICNN) and long short-term memory networks (LSTM), referred to as ICNN-LSTM, was proposed. Experimental results indicate that the proposed method accurately extracts features from subtle vibrations and demonstrates superior classification accuracy compared to other models. This study provides an effective technical solution for ensuring the safe operation of magnetic-adhesion crawler-type climbing wall robots, showcasing significant practical value.