An Optimal Spatio-Temporal Hybrid Model Based on Wavelet Transform for Early Fault Detection

Area of Science:

• Industrial Process Monitoring
• Fault Detection and Diagnosis
• Signal Processing

Background:

• Early-stage faults in industrial systems are often subtle and masked by noise, hindering timely detection.
• Existing methods may lack the sensitivity to identify slowly evolving faults, impacting production safety and quality.
• Complex industrial environments require advanced techniques for robust fault detection.

Purpose of the Study:

• To develop an optimal spatio-temporal hybrid model (STHM) for enhanced early fault detection.
• To improve the sensitivity and accuracy of detecting slowly evolving, noise-submerged faults.
• To reduce false alarm rates and enhance industrial production safety and product quality.

Main Methods:

• Wavelet Transform (WT) for data denoising and noise reduction.
• Principal Component Analysis (PCA) and sliding window algorithm for spatial-temporal neighbor acquisition.
• Cumulative Sum (CUSUM) and Mahalanobis Distance (MD) for hybrid statistic reconstruction.
• Kernel Density Estimation (KDE) for optimizing the fault detection threshold.

Main Results:

• The proposed WT-based STHM effectively denoises data, reducing background noise interference.
• Hybrid statistics enhance the correlation between temporal dynamics and spatial information, improving fault detection precision.
• Simulations on the Tennessee Eastman (TE) process demonstrated a high fault detection rate (FDR) and low false alarm rate (FAR).

Conclusions:

• The optimal STHM provides a sensitive and accurate method for early fault detection in complex industrial systems.
• The model's ability to handle noise and subtle faults contributes to improved industrial safety and product quality.
• This approach offers a robust solution for real-time monitoring and fault diagnosis in industrial production.