XAI-LCS: Explainable AI-Based Fault Diagnosis of Low-Cost Sensors
Aparna Sinha, Debanjan Das
Abstract
An accurate technique for early detection of sensor faults proves useful in the uninterrupted supply of correct monitoring data across the Internet of Things (IoT) network. Most of the existing AI-based fault diagnosis techniques have a high computational burden, and their “black-box” nature creates challenges in generating adequate trust in high-risk industrial applications. To address the existing drawbacks, a unique IoT-based method, i.e., XAI-LCS, has been proposed that uses eXtreme gradient boosting algorithm for detecting different types of sensor faults, such as bias, drift, complete failure (CF), and precision degradation. This method is also capable of handling imbalanced data distribution to prevent biased predictions. The fault detection method identifies four types of sensor faults with 99.8% validation accuracy. The explainable AI interprets the prediction outcome and increases the trustworthiness of the used AI model.