Fault Diagnosis

Fault diagnosis studies how to detect abnormal behavior, identify its source and severity, and provide evidence that can guide repair in systems such as machinery, power grids, vehicles, networks, and software. Common approaches learn patterns from vibration, electrical signals, images, or logs using classifiers, anomaly detectors, and graph or sequence models; physics-based models, digital twins, causal graphs, and knowledge retrieval can add structure when data are scarce or explanations matter. Recent systems also use language-model agents to investigate incidents and communicate findings, often alongside explicit checks that ground conclusions in telemetry or maintenance records. Strong results on known faults and benchmark data do not always carry over to new operating conditions or unseen failures, making reliable evaluation, uncertainty handling, and real-world validation central challenges.