Hybrid Deep Learning-Based Fault Detection for IoT Sensors

An accurate method for early fault detection in sensor is critically important for smooth and reliable network operation in an IoT network. Most of the existing AI-based fault diagnosis approaches suffer from low interpretability with high computational complexities, which degrades their adoption in high-risk industrial applications. To address the aforementioned issues, this paper provides a hybrid deep learning-based approach that integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Graph Neural Networks (GNN) for fault detection in IoT sensors. The proposed method effectively detects various faults that may arise in sensors, including bias, drift, intermittent failure, precision loss, scaling, and linearity issues, and also reduces the challenge of data imbalance that may occur to prevent biased predictions. CNN extracts spatial features, LSTM captures temporal dependencies, and GNN models complex sensor relationships, where the accuracy of fault detection gets enhanced. Experimental results show that it is an accurate fault detector that provides a reliable, interpretable method for real-time IoT sensor fault diagnosis.