Early Fault Diagnosis Method for Batch Process Based on Local Time Window Standardization and Trend Analysis.

Area of Science:

• Chemical Engineering
• Process Control
• Data Science

Background:

• Batch processes are economically valuable but complex, leading to high failure rates.
• Sensor data in batch processing is often noisy and exhibits multistage variations, hindering accurate fault diagnosis.
• Early fault detection is critical for operational efficiency and safety in these processes.

Purpose of the Study:

• To develop a robust fault diagnosis method for batch processes that overcomes noise and multistage variations.
• To enhance the extraction and utilization of fault information from sensor data.
• To improve the diagnostic performance and efficiency compared to traditional methods.

Main Methods:

• An adaptive standardization method based on a time window was developed.
• Quadratic fitting was employed to extract data trends within the time window.
• A novel trend recognition method utilizing Euclidean distance calculations was created.

Main Results:

• The proposed method was validated using penicillin fermentation data.
• Two test datasets (existing and unknown batches) were used, achieving average diagnostic rates of 100% and 87.5%, respectively.
• The mean diagnosis time was consistently low at 0.2083 hours.

Conclusions:

• The developed fault diagnosis algorithm demonstrates superior fault diagnosis and feature extraction capabilities.
• The method effectively handles noisy and complex data typical of batch processes.
• This approach offers a significant improvement for early fault detection in industrial batch operations.