Traceable Algorithm Unrolling Network: An Interpretable Deep Sparse Representation Model for Mechanical Fault...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Intelligent fault diagnosis (IFD) methods achieve high accuracy but lack interpretability due to excessive parameters.
• Uninterpretable architectures lead to ambiguity and unclear decision-making bases in mechanical fault diagnosis (MFD).

Purpose of the Study:

• To propose a traceable algorithm unrolling (TAU) network for interpretable MFD.
• To enhance the credibility and transparency of fault diagnosis decision-making processes.

Main Methods:

• A mechanism-driven feature extractor (FE) is developed by unrolling sparse coding for interpretable feature encoding from vibration signals.
• A theory-based feature clustering (FC) algorithm utilizes capsule network (CN) dynamic routing, with inner product measuring feature association.
• A post hoc interpretability strategy with a coupling matrix analyzes how TAU generates results and validates feature-fault associations.

Main Results:

• TAU demonstrates diagnostic decisions based on high-dimensional feature mapping linked to fault characteristic frequencies.
• The proposed method enhances the credibility of fault diagnosis results through verified feature-fault associations.
• Simulation and experimental results confirm TAU's decision-making mechanism and superior fault diagnosis performance.

Conclusions:

• The TAU network provides an interpretable framework for MFD, overcoming limitations of traditional IFD methods.
• TAU's approach ensures that diagnostic decisions are grounded in meaningful, interpretable features.
• The study validates TAU's effectiveness and credibility in mechanical fault diagnosis applications.