ABSTRACT Addressing the shortcomings of traditional reliability‐centred maintenance (RCM) for reciprocating compressors, namely subjective risk assessment, inadequate generalisation of small‐sample modelling, and maintenance decisions reliant on experience, this paper proposes an optimised RCM model driven by the synergistic application of ‛Pre‐trained Language Model (PLM)—Entropy‐weighted Failure Mode and Effects Analysis (EW‐FMEA)—Multi‐layer Decision Tree (MLDT)—Domain‐adaptive Transfer Learning (DATL)’. This model achieves standardised fusion of multimodal operational data via PLM; employs EW‐FMEA to objectively assign weights for severity (S), occurrence probability (O), and detectability (D), thereby resolving the insufficient discrimination of traditional risk priority numbers (RPN); leverages MLDT to adapt to differentiated risk scenarios and generate targeted maintenance strategies; and overcomes small‐sample constraints through DATL to enhance model generalisation performance. Experimental validation demonstrates a model risk assessment stability rate of 0.997, with core risk dimension prediction accuracy approaching 1.0. Risk level‐to‐maintenance recommendation alignment exceeds 98%, delivering significantly superior comprehensive performance compared to conventional approaches. This research proposes a multi‐technology collaborative RCM optimisation theoretical framework, clarifying the integration logic and mechanism. It extends the boundaries of small‐sample industrial equipment maintenance technology, achieves full‐process automation of RCM, and provides an implementable academic paradigm and technical solution for intelligent health management of reciprocating compressors and similar equipment.
Hou et al. (Thu,) studied this question.