Ensemble semi-supervised Fisher discriminant analysis model for fault classification in industrial processes

Area of Science:

• Industrial Process Monitoring
• Machine Learning
• Fault Diagnosis

Background:

• Accurate fault classification is crucial for industrial process safety and efficiency.
• Traditional methods often struggle with complex, unlabeled industrial data.
• Semi-supervised and ensemble learning offer potential for improved fault detection.

Purpose of the Study:

• To develop an ensemble semi-supervised Fisher Discriminant Analysis (FDA) model for industrial fault classification.
• To enhance classification performance using K Nearest Neighbor (KNN) and adaptive weight adjustment.
• To improve model robustness and generalization by leveraging unlabeled data and diverse weak learners.

Main Methods:

• An ensemble Fisher Discriminant Analysis (FDA) model was developed.
• K Nearest Neighbor (KNN) algorithm was employed to merge outputs from sub-classifiers.
• An adaptive weighting method was introduced to further optimize classification performance.

Main Results:

• The proposed ensemble FDA model demonstrated effective fault classification capabilities.
• The adaptive weighting strategy enhanced the model's classification accuracy.
• The combined approach showed significant generalization performance on an industrial benchmark process.

Conclusions:

• Ensemble semi-supervised FDA, integrated with KNN and adaptive weighting, provides a robust solution for industrial fault classification.
• The method effectively utilizes unlabeled data to improve model generalization.
• The developed approach shows promise for real-world industrial monitoring applications.