Adaptive Convolution Sparse Filtering Method for the Fault Diagnosis of an Engine Timing Gearbox

Area of Science:

• Mechanical Engineering
• Signal Processing
• Fault Diagnosis

Background:

• Convolution Sparse Filtering (CSF) offers robust outlier signal detection without prior knowledge.
• Current CSF methods are limited by suboptimal filter number and length selection.
• Effective health monitoring and fault diagnostics are crucial for industrial machinery.

Purpose of the Study:

• To propose an Adaptive Convolution Sparse Filtering (ACSF) method for end-to-end health monitoring and fault diagnosis.
• To introduce a novel entropy-time function (He-T) for evaluating filtered signal accuracy and efficiency.
• To optimize filter selection using particle swarm optimization for improved diagnostic performance.

Main Methods:

• Developed the Adaptive Convolution Sparse Filtering (ACSF) algorithm.
• Introduced the entropy-time function (He-T) as a performance metric.
• Employed particle swarm optimization to find the optimal He-T.
• Utilized envelope spectrum analysis for failure mode diagnosis.

Main Results:

• The ACSF method demonstrated effectiveness and efficiency in gearbox fault diagnosis.
• The proposed He-T metric successfully guided the selection of optimal filters.
• ACSF successfully extracted characteristic failure signatures from gearbox signals.
• Experimental validation confirmed the ACSF's capability in identifying gearbox faults.

Conclusions:

• ACSF provides an improved approach to signal filtering for fault diagnosis.
• The method overcomes limitations of traditional CSF by adaptively selecting filter parameters.
• ACSF shows significant potential for enhancing machinery health monitoring systems.