Slow feature-based feature fusion methodology for machinery similarity-based prognostics

Area of Science:

• Machine Learning
• Predictive Maintenance
• Signal Processing

Background:

• Similarity-based prediction methods are crucial for industrial predictive maintenance, especially for machines with unknown failure mechanisms.
• Current methods often fail to account for discrepancies in degradation trends from multi-sensor data and lack automatic normalization of operating regimes during feature fusion.

Purpose of the Study:

• To propose a novel feature fusion methodology for enhanced degradation trend analysis.
• To address limitations in current methods regarding multi-sensor data discrepancies and operating regime normalization.

Main Methods:

• A feature fusion methodology based on a signal-to-noise ratio metric is proposed.
• Slow Feature Analysis (SFA) is leveraged to quantify degradation trend discrepancies and automatically filter operating regimes during feature fusion.

Main Results:

• The proposed method effectively quantifies degradation trend discrepancies in Degradation Indicators (DIs).
• Automatic normalization of operating regimes during feature fusion is achieved, improving prognostic accuracy.
• Demonstrated effectiveness and superiority on aero-engine and rolling bearing datasets.

Conclusions:

• The developed feature fusion methodology offers a superior approach to predictive maintenance.
• This method enhances the reliability of prognostics for machines with complex degradation patterns.