Comparison and Optimization of Generalized Stamping Machine Fault Diagnosis Models Using Various Transfer Learning Methodologies

Abstract

The integration of artificial intelligence (AI) with stamping technology has become increasingly critical in smart manufacturing, driven by advancements in both fields. Total clearance, a crucial determinant of both process and product quality in stamping operations, significantly impacts cutting precision, material deformation, and the longevity of stamping equipment. Consequently, real-time monitoring and prediction of total clearance are essential for effective process control and fault diagnosis. However, the heterogeneity of stamping machine designs necessitates the development of numerous machine-specific models, posing a significant challenge for practical implementation. This research addresses this challenge by developing a generalized fault diagnosis model applicable across multiple stamping machine types. Specifically, the model is designed to monitor four distinct machine models: OCP-110, G2-110, G2-160, and ST1-110. Vibration data, acquired using accelerometers strategically placed at two distinct sensor locations on each machine, serve as the primary input for the model. Four prominent deep learning architectures—a 10-layer convolutional neural network (CNN), a CNN with residual connections (CNN-Res), VGG16, and ResNet50—were rigorously evaluated in conjunction with fine-tuning strategies to determine the optimal model architecture. The resulting generalized fault diagnosis model achieved an average accuracy, recall rate, and F1 score exceeding 99%, demonstrating its efficacy and reliability for real-world applications. This proposed approach offers the potential for scalability to additional stamping machine types and operational conditions, thereby streamlining the deployment of predictive maintenance systems by equipment manufacturers.