On the Optimal Self-Supervised Multi-Fault Detector for Temperature Sensor Data

Accurate detection of faults in sensor data is essential for monitoring and controlling industrial processes, environmental conditions, and infrastructure to ensure reliability and enable informed decision-making. Resolving these faults ensures measurement quality and unlocks process optimization opportunities, resulting in improved performance, energy efficiency, and cost savings. We present a transformer-based fault detection model which adopts the anomaly-attention mechanism. Experiments have been performed on the benchmark faults injected Intel temperature sensor datasets using precision, recall, and F1-score metrics. The result outperforms the other classical and complex algorithms, proving our method's effectiveness.