Mean value model-assisted dual transfer: a cross-domain fault diagnosis framework in diesel engines from simulation domains to experimental domains

Marine diesel engine fault diagnosis is critical to the safety and reliability of ship power systems. This paper proposes a mean value model-assisted dual transfer (MVMA-DT) framework for cross-domain fault diagnosis of marine diesel engines. The framework integrates two synergistic methods: 1) data level transfer (DLT) method driven by the diesel engine mean value model and an information fusion architecture, which generates simulated response data closer to the distribution of actual measured data to enhance the source domain feature characterisation capability; 2) model level transfer (MLT) method, which employs the dense connectivity neural network (DCNN) to extract the fault features and integrates a comprehensive optimization strategy, efficiently enhances the model's cross-domain diagnostic capability, achieving more accurate cross-domain fault diagnosis. Diagnostic experimental results demonstrate that the framework achieves an average accuracy of over 97.70 % in cross-domain fault diagnosis tasks for diesel engine lubricating-oil systems, which is 61.04 % and 14.86 % higher than non-Model level transfer method and non-Data level transfer method. The framework enables dual cross-domain fault diagnosis for marine diesel engines across types and operating conditions, offering a new hybrid digital-model driven paradigm for diesel engine fault diagnosis.