Concurrent control chart pattern recognition in manufacturing processes based on zero-shot learning.

Area of Science:

• Industrial Engineering
• Machine Learning
• Statistical Process Control

Background:

• Collecting labeled concurrent abnormal control chart patterns (C-CCPs) is difficult in industrial settings.
• Existing C-CCP recognition (CCPR) methods are limited by the lack of training data.
• Zero-shot learning offers a potential solution for recognizing novel C-CCPs.

Purpose of the Study:

• To propose an intelligent zero-shot learning model for recognizing concurrent abnormal control chart patterns (C-CCPs).
• To address the challenge of limited labeled C-CCP data in industrial quality control.
• To improve the effectiveness of CCP recognition (CCPR) methods.

Main Methods:

• Introduced zero-shot learning into quality control for C-CCP recognition.
• Developed a multiscale ordinal pattern (OP) feature to capture data sequential relationships.
• Established an attribute description space (ADS) using expert knowledge to bridge single CCPs and C-CCPs.
• Integrated an attribute classifier to associate features with attributes.

Main Results:

• Achieved 98.73% accuracy for 11 unseen C-CCPs without C-CCP training samples.
• Attained an overall accuracy of 98.89% for all 19 CCPs.
• Demonstrated superior recognition performance of the multiscale OP feature on unseen C-CCPs compared to other features.

Conclusions:

• The proposed zero-shot learning model effectively recognizes unseen C-CCPs.
• The multiscale OP feature and ADS are crucial for successful C-CCP recognition.
• This approach significantly advances CCPR in industrial quality control, especially with limited data.