Open-set cross-working-condition diagnosis of marine engines for unknown faults

With the development of the marine shipping industry towards high efficiency and intelligence, marine internal combustion engines, as the core components of marine power, operate under complex and variable navigation conditions for long periods. Moreover, unknown fault types often occur in practical engineering, imposing higher requirements on the accuracy and adaptability of fault diagnosis. The existing traditional closed-set fault diagnosis methods are limited in performance in practical applications, as they cannot adapt to the feature distribution differences caused by working condition fluctuations and are difficult to identify unknown faults, making it challenging to solve the problem of open-set cross-working-condition fault diagnosis for marine engines. To address the above issues, this paper proposes an open-set cross-working-condition fault diagnosis method suitable for marine engines. Firstly, based on the fault physical mechanism and cross-working-condition data characteristics, a multi-scale separable convolution structure is constructed to perform deep feature extraction on the source domain fault data and capture the core discriminant features of faults. Secondly, a domain adversarial training mechanism is introduced to align the feature distributions of the source domain and the target domain, weaken the impact of domain shift, and an outlier detection module is embedded to model the optimal decision boundary for known fault categories, thus simultaneously achieving known fault classification and unknown fault identification. Through simulation experiments, the proposed method is verified in multiple sets of open-set diagnosis tasks based on the cross-working-condition fault data of marine engine lubricating oil systems, air intake and exhaust systems, and bearings, and compared with traditional methods. The results show that the proposed method achieves an unknown fault recognition rate of over 90%, and the average classification accuracy of known faults is more than 12% higher than that of traditional methods, which effectively overcomes the bottlenecks caused by variable working conditions and unknown faults. This study improves the fault diagnosis capability of marine engines under open-set cross-working conditions, provides a practical technical scheme for the condition monitoring and unknown fault identification of marine power plants, and has important engineering value for ensuring navigation safety and reducing operation and maintenance costs.