Fault prediction method of large forging press based on a multi scale and multi model integrated method.

Area of Science:

• Mechanical Engineering
• Manufacturing Technology
• Artificial Intelligence in Industry

Background:

• Large forging presses are crucial in automotive, shipbuilding, and aerospace manufacturing.
• Intricate component relationships in these presses make early fault detection challenging.
• Traditional single-model prediction methods struggle with nonlinear, stochastic fault characteristics.

Purpose of the Study:

• To develop an advanced fault prediction model for large forging presses.
• To overcome limitations of traditional single-model approaches in complex industrial settings.
• To improve the accuracy and generalization capability of fault signal prediction.

Main Methods:

• Proposed a multi-scale Autoregressive-Support Vector Regression-Long Short-Term Memory (AR-SVR-LSTM) ensemble prediction approach.
• Integrated AR, SVR, and LSTM models with weight constraints for hybrid prediction.
• Validated the model using brake oil pressure data from an 80MN electric screw press.

Main Results:

• The proposed AR-SVR-LSTM ensemble model showed significant advantages over comparative models.
• The model's accuracy, while not meeting stringent industry standards (MAE < 5%, RMSE < 1%), demonstrated effectiveness.
• Case study on an 80MN electric screw press confirmed the model's practical applicability.

Conclusions:

• The multi-model ensemble approach offers a superior alternative for fault prediction in large forging presses.
• Despite not meeting all industry benchmarks, the model's enhanced performance warrants further investigation.
• This research contributes to advancing predictive maintenance strategies in heavy-duty manufacturing.