Fault diagnosis of marine diesel engine via control-aware domain-adversarial network: A control-diagnosis collaboration perspective

Aiming at the issue that existing fault diagnosis methods for marine diesel engines often neglect the diagnostic value of information contained in control signals, this paper proposes a control-aware domain-adversarial network for injector system fault diagnosis across maneuvering-regime distributions. Based on the diesel engine’s physical characteristic, where control signals dominate state signals, the control-aware domain-adversarial network employs a novel weighted fusion mechanism, which multiplies the features extracted from the state signals by the features extracted from the control signals using weight parameters that are generated by the control branch from the control signals and compressed into the interval (0, 1) through a sigmoid function. Meanwhile, the control-aware domain-adversarial network establishes a fine-grained domain-adversarial framework, in which the maneuvering-regime discriminator replaces the conventional binary classification—which distinguishes only between the source and target domains—with a multi-class formulation to maintain stable fault diagnosis performance across maneuvering-regime distributions. Experiments on simulation data from a thermodynamic model show that compared with mainstream methods, the proposed approach exhibits stronger robustness when control sensors drop out. The absence of real-engine validation is currently a limitation of this work. Consequently, future validation and optimization on a real engine would be highly desirable.