Towards Effective Interpretability In IoT Fault Diagnosis

In IoT systems, it is of utmost importance to develop a fault diagnosis method that reliably detects faults, identifies their underlying causes, and enables prompt fault resolution. However, while current diagnostic methods excel at identifying faulty data, they cannot often provide a clear interpretation of the underlying cause of the fault. This creates a significant challenge in identifying the appropriate mitigation action promptly, as faults can originate from various sources such as sensor mal-functions, network disruptions, power supply issues, or software errors, each requiring a unique remedy solution. To address this gap, this research proposes a novel rule-based method that enables an interpretable fault diagnosis technique, uncovering the root causes of anomalies through a transparent procedure. In the proposed approach, we consider multiple factors to enable a clear understanding of anomaly sources. Hence, after detecting a fault through sensor data analysis, we evaluate the impact of the fault on the operational status of the edge device. It includes monitoring and assessing the execution time of code on the microcontroller and function call sequences. Additionally, we consider the effects of the fault on in-situ sensors and neighboring IoT edge devices sharing the same access point. By integrating these elements, the approach enables a comprehensive interpretation of the root cause of the anomaly in a fault diagnosis procedure, distinguishing it from other methods. The effectiveness of the proposed solution in enhancing the visibility of the fault diagnosis process is demonstrated through a proof of concept involving various real-world IoT application scenarios.