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

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Stamping operations require precise total clearance for quality and equipment longevity.
• Diverse stamping machine designs hinder the development of universal fault diagnosis models.
• Real-time monitoring of total clearance is crucial for process control and fault detection.

Purpose of the Study:

• To develop a generalized fault diagnosis model for stamping machines.
• To enable effective process control and predictive maintenance across different machine types.
• To overcome the challenge of machine-specific model development in smart manufacturing.

Main Methods:

• Utilized vibration data from accelerometers on four distinct stamping machine models (OCP-110, G2-110, G2-160, ST1-110).
• Evaluated four deep learning architectures: CNN, CNN-Res, VGG16, and ResNet50, with fine-tuning strategies.
• Developed a generalized fault diagnosis model applicable across multiple stamping machine types.

Main Results:

• The generalized fault diagnosis model achieved average accuracy, recall, and F1 scores exceeding 99%.
• Demonstrated high efficacy and reliability in real-world stamping fault diagnosis.
• Validated the model's performance across OCP-110, G2-110, G2-160, and ST1-110 machine models.

Conclusions:

• The developed generalized model effectively diagnoses faults in diverse stamping machines.
• This AI-driven approach streamlines predictive maintenance deployment in smart manufacturing.
• The model shows potential for scalability to more machine types and operational conditions.