Detection of Peak Points for Wear Control of Band Saw Blades

Industrial band saw cutting machines are widely used in metalworking and mass production processes due to their high precision and efficiency. These machines offer significant advantages such as reducing labor costs, increasing productivity, ensuring occupational safety, and saving energy. However, the wear or breakage of band saw blades can negatively impact production quality and machine performance. This study compares four different edge detection algorithms for detecting wear and fractures in the blades of industrial band saw cutting machines. These algorithms are LDC, HED, Sobel, and Canny. The selected four algorithms were applied to a dataset obtained from a project supported by the 1711 Artificial Intelligence Ecosystem Call of TÜBİTAK. The performance of the edge detection algorithms was evaluated using statistical metrics such as Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). Experimental results showed that deep learning-based algorithms (Lightweight Dense CNN (LDC) and Holistically-Nested Edge Detection (HED)) performed with higher accuracy compared to image processing-based algorithms (Sobel and Canny). In particular, the LDC algorithm demonstrated the best performance with shorter processing times and fewer parameters. These findings reveal the potential of using deep learning-based edge detection algorithms for real-time fault detection and predictive maintenance in industrial cutting machines. The results obtained in this study indicate that deep learning-based methods can be effectively utilized to enhance the efficiency and reliability of industrial cutting machines. In this context, the applicability of the cost-effective and highly efficient LDC algorithm is particularly noteworthy for resource-limited systems.