Meta learning based residual network for industrial production quality prediction with limited data

Area of Science:

• Industrial Engineering
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning models for production quality prediction struggle with limited real-world industrial data.
• Model degradation during training is a significant challenge in deep learning networks.

Purpose of the Study:

• To propose a Meta-Learning based on Residual Network (MLRN) model for production quality prediction with limited data.
• To address model degradation issues in deep learning training processes.

Main Methods:

• Developed an MLRN model trained on diverse learning tasks to acquire predictive knowledge.
• Utilized an enhanced residual network with Effective Channel Attention (ECA) for feature extraction and to mitigate gradient issues.
• Implemented a multi-batch and multi-task data input strategy to prevent overfitting.

Main Results:

• The MLRN model demonstrated availability and effectiveness in production quality prediction.
• Performance was validated through comparisons with other models on both numerical and graphical datasets.

Conclusions:

• The proposed MLRN model offers a viable solution for production quality prediction in data-scarce industrial environments.
• The integration of meta-learning, enhanced residual networks, and attention mechanisms improves model robustness and predictive accuracy.