Leveraging Information Flow-Based Fuzzy Cognitive Maps for Interpretable Fault Diagnosis in Industrial Robotics

In Industry 4.0, Artificial Intelligence (AI) is revolutionizing manufacturing with innovations such as automated fault detection in robotics. However, many current AI models are opaque, obscuring decision-making processes and reducing worker trust. Additionally, these models rely on correlative learning, making them susceptible to adopting spurious correlations that affect their reliability and generalizability. This paper presents the use of Information Flow-Based Fuzzy Cognitive Maps (IF-FCMs) for fault detection and diagnosis in industrial robotics, aiming to overcome these challenges. IF-FCMs, building on FCMs known for their intuitive causal structure and interpretability, integrate Liang-Kleeman Information Flow analysis for rigorous data-driven causality analysis. This approach effectively distinguishes authentic causal links from spurious correlations, enhancing the predictive and explanatory power of FCMs. Moving beyond previous studies that used synthetic data, which often lack real-world complexity and variability, this study employs actual industrial robot data. Numerical simulations demonstrate that IF-FCMs outperform traditional FCMs in terms of both diagnostic accuracy and interpretability, underscoring their potential for tackling manufacturing challenges.