ThermoVision-MOSFET: A Multimodal Thermal–Radiographic Machine Learning Framework for Fault Diagnosis and Maintenance Decision Support in Power MOSFETs
Year: 2026
Type: article
Abstract: The reliability of power-electronic, automotive, and industrial systems relies on reliable fault detection in the Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs). Single-modality inspection is still limited as infrared thermography has only surface thermal anomalies, while X-ray radiography has internal structural defects. The paper proposes an integrated diagnostic framework, called ThermoVision-MOSFET, which integrates thermal infrared imaging, X-ray imaging, pre-processing, hybrid segmentation, multi-domain feature extraction, and machine-learning classification. The segmentation pipeline combines the Otsu and adaptive thresholding with the Canny and Sobel edge detection and marker-guided watershed refinement to enhance fault-region localization. Random Forest is used to classify a 28 feature representation that contains statistical, texture, thermal-gradient, morphological, X-ray structural and fusion features. The framework was evaluated on a dataset of 400 MOSFET samples from four diagnosis classes, yielding a segmentation accuracy of 94.7%, classification accuracy of 96.8%, an F1-score of 95.9%, and an AUC of 0.978. The segmentation method proposed in this work outperformed the next best individual method by 9.5 percentage points, and the hybrid thermal-radiographic analysis showed a detection rate of 93.7%–97.6% for six physical fault manifestations. With the highest noise level, segmentation accuracy was 84.1% and the processing time was about 0.47 s per image. The statistical comparisons revealed that the results were significantly better than the individual segmentation baselines (p < 0.01), and the measured throughput suggests the application of moderate throughput diagnostic inspection and assessment for maintenance could be realized in near real-time. The results illustrate the benefit of using multimodal evidence for the diagnosis of MOSFETs and for maintenance-oriented decision making based on condition-based inspection, planned maintenance and prioritized maintenance.
Source: Computer and decision making.
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7222794726