A Coal-Rock Cutting Hardness Recognition Model Based on CNN-PN

Existing methods for accurate recognition of coal-rock cutting hardness by shearer machines in complex environments face challenges in timely adjusting the drum speed and traction velocity to adapt to changes in coal-rock cutting hardness. In this paper, we propose a Coal-Rock Cutting Hardness Recognition Model based on Convolutional Neural Network (CNN) - Prototype Network (PN). Firstly, multiple sensor signals are utilized to generate high-dimensional data, which is then fed into CNN for feature extraction. Subsequently, the features extracted by the convolutional neural network are input into the PN, which calculates the embedding center distances for each category, obtaining representative prototype vectors. The similarity between prototype vectors is measured using Euclidean distance, and these prototype vectors form different groups. Thirdly, the Adam algorithm is employed to optimize the network model, enabling better adaptation to data distribution differences and improving training effectiveness. Finally, the proposed model is validated through experiments using collected data. Unlike traditional machine learning approaches that require extensive manual feature extraction and data preprocessing for coal-rock cutting hardness recognition, which affects recognition accuracy and efficiency, the proposed method overcomes these limitations and can directly classify raw multi-sensor data signals during coal-rock cutting by shearer machines while effectively recognizing coal-rock cutting hardness. Experimental results demonstrate an average recognition rate of 99.64% for the proposed model, indicating high recognition accuracy and good generalization performance.