Transparent information fusion network: An explainable network for multi-source bearing fault diagnosis via self-organized neural-symbolic nodes

Abstract

In recent years, the integration of Artificial Intelligence (AI) into Intelligent Fault Diagnosis (IFD) through multi-source signal fusion has advanced significantly. However, the inherent opacity of AI-driven IFD models often hampers explainability, a crucial factor for fostering deeper collaboration between humans and intelligent systems to enhance the safety and reliability of industrial assets. This study introduces a novel and explainable methodology termed the Transparent Information Fusion Network (TIFN). TIFN incorporates multiple self-organized Neural-Symbolic Nodes (NSNs) equipped with signal processing and statistical operators, as opposed to conventional black-box neural networks or manually crafted expert systems. NSNs are interconnected through learnable signal-wise gates to construct the Signal Operator Layer (SOL) and Feature Operator Layer (FOL). The entire network is trainable using gradient-based learning methods in a self-organized manner. This enables accurate representation of fault signals and establishes a fully comprehensible semantic framework with an explainable decision-making process. The transparency, generalization, and capability to learn from limited samples of TIFN are demonstrated through two case studies on rotating machinery equipped with different sensors. Case 1 fuses vibration and piezoelectric signals, while Case 2 integrates piezoelectric and triboelectric signals to achieve comprehensive information fusion at both signal and feature levels. By incorporating learnable NSNs, signal-wise gates, TIFN achieves superior diagnostic performance with fewer parameters compared to traditional expert-organized models. This research underscores the potential of TIFN as a fully explainable tool for industrial diagnostics with multi-source signals, paving the way for enhanced human-machine collaboration in Industry 5.0, with a focus on trustworthiness, transparency, and accountability.