Answers are grounded in this paper and cite their sources.
Early and reliable detection of bearing faults in rotating machinery is critical for preventing unexpected failures, enhancing safety, and reducing maintenance costs. Although Temporal Convolutional Networks (TCNs) effectively extract discriminative features from vibration signals, their performance depends strongly on tuning interacting hyperparameters. Conventional approaches, including manual tuning and grid search, struggle to explore the high-dimensional, non-convex search spaces of deep learning models efficiently. This study proposes a hybrid optimisation framework integrating a TCN classifier with the Wound Healing Optimisation Algorithm (WHOA), a bio-inspired metaheuristic based on four tissue-repair stages: haemostasis, inflammation, proliferation, and remodelling. These stages are translated into adaptive computational operators that balance exploration and exploitation, control convergence, and preserve population diversity. Experiments on the CWRU bearing dataset demonstrate that the optimised TCN-WHOA achieves 99.9% accuracy, a macro-F1 score of 0.999, and an area under the ROC curve approaching 1.0, outperforming existing methods. Sensitivity analysis also identifies the hyperparameters most influential in model stability and generalisation. These findings indicate that TCN-WHOA offers an accurate and robust approach to vibration-based bearing fault diagnosis, supporting more effective predictive maintenance in industrial rotating machinery, subject to further validation across diverse operating conditions.
No takes yet. Share an insight, caveat, or question.
Kaya et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: