Deployment-Oriented Deep Learning for Predictive Maintenance of Electric Motor-Driven Rotating Systems: A Review of Robustness, Trustworthiness, and Industrial Readiness

Recent developments in predictive maintenance (PdM) for motor-driven rotating systems have improved fault detection, fault diagnosis, anomaly detection, and remaining useful life (RUL) prediction. Despite this progress, industrial deployment remains constrained by scarce labeled data, variable operating conditions, noise, domain shift, computational and energy limits, privacy requirements, and the need for trustworthy decision support. This review examines deployment-oriented deep learning for electric motor-driven rotating systems, with electric motors treated as the central electromechanical assets and bearings, shafts, gears, and coupled loads considered as mechanically coupled subsystems. The review combines a structured literature mapping with a challenge-driven synthesis of 119 recent research papers and 30 review papers, following the PRISMA 2020 guidelines, and is organized around three key themes: data and model efficiency, reliability and generalization, and system-level integration. The literature is organized around data-efficient learning, robust and generalizable models, lightweight and deployable architectures, explainability and uncertainty quantification, foundation models, large language model-assisted frameworks, multi-sensor fusion, and federated learning. By linking methodological advances to practical deployment constraints, this review identifies the current maturity of recent PdM methods, clarifies the remaining gap between benchmark validation and industrial readiness, and outlines research priorities for robust, scalable, and trustworthy PdM solutions in motor-driven powertrain systems.