Accurate and timely fault diagnosis for the diesel engine is crucial to guarantee it works safely and reliably, and reduces the maintenance costs. A novel diagnosis method based on variational mode decomposition (VMD) and kernel-based fuzzy c-means clustering (KFCM) is proposed in this paper. Firstly, the VMD algorithm is optimized to select the most suitable K value adaptively. Then KFCM is employed to classify the feature parameters of intrinsic mode functions (IMFs). Through the comparison of many different parameters, the singular value is selected finally because of the good classification effect. In this paper, the diesel engine fault simulation experiment was carried out to simulate various faults including valve clearance fault, fuel supply fault and common rail pressure fault. Each kind of machine fault varies in different degrees. To prove the effectiveness of VMD-KFCM, the proposed method is compared with empirical mode decomposition (EMD)-KFCM, ensemble empirical mode decomposition (EEMD)-KFCM, VMD-back propagation neural network (BPNN), and VMD-deep belief network (DBN). Results show that VMD-KFCM has advantages in accuracy, simplicity, and efficiency. Therefore, the method proposed in this paper can be used for diesel engine fault diagnosis, and has good application prospects.

Fault diagnosis can improve the safety and reliability of diesel engines. An end-to-end method based on a multi-attention convolutional neural network (MACNN) is proposed for accurate and efficient diesel engine fault diagnosis. By optimizing the arrangement and kernel size of the channel and spatial attention modules, the feature extraction capability is improved, and an improved convolutional block attention module (ICBAM) is obtained. Vibration signal features are acquired using a feature extraction model alternating between the convolutional neural network (CNN) and ICBAM. The feature map is recombined to reconstruct the sequence order information. Next, the self-attention mechanism (SAM) is applied to learn the recombined sequence features directly. A Swish activation function is introduced to solve “Dead ReLU” and improve the accuracy. A dynamic learning rate curve is designed to improve the convergence ability of the model. The diesel engine fault simulation experiment is carried out to simulate three kinds of fault types (abnormal valve clearance, abnormal rail pressure, and insufficient fuel supply), and each kind of fault varies in different degrees. The comparison results show that the accuracy of MACNN on the eight-class fault dataset at different speeds is more than 97%. The testing time of the MACNN is much less than the machine running time (for one work cycle). Therefore, the proposed end-to-end fault diagnosis method has a good application prospect.

Transfer learning (TL) is a powerful approach that enhances the generalizability of cross-domain fault diagnosis. However, the challenge of acquiring high-quality mechanical fault signals limits its application. This article introduces the extreme class imbalance problem in the cross-domain diagnosis, restricting the label space of the target domain while relaxing the restrictions of unsupervised learning. The study proposes a novel generative TL method called fast sparse neural style, which employs sparse representation to capture the domain-invariant fault features as well as the Gram matrix to measure the domain-specific features. Fault features and domain features are proven to be separable in mechanical signals and are fused in the data generation process. Compared to other methods through various cross-domain diagnostic tasks on a piston aero-engine, the proposed method has obvious advantages in tasks with substantial inter-domain differences, demonstrating the potential and research value of generative TL.

Transfer learning is an effective approach to address the decline in generalizability of intelligent fault diagnosis methods. However, there has been a persistent lack of comprehensive and effective metrics for assessing the transferability of cross-domain data, making it challenging to answer the fundamental question in transfer learning: “When to transfer”. This study proposes a novel hybrid transferability metric (HTM) based on weighted correlation-diversity shift. The metric introduces a correlation shift measurement based on sparse principal component analysis, effectively quantifying distribution differences in domain-invariant features based on the sparse representation theory. It also designs a diversity shift measurement based on label space differences, addressing the previously overlooked impact of label variation on transferability. The proposed transferability metric is validated on four types of cross-domain diagnosis tasks involving piston aero engines. The results show that in diagnostic scenarios involving both supervised transfer learning and extreme class imbalance problems, HTM accurately predicted the transferability of the target tasks, which aligned with the actual diagnostic accuracy trends. It provides a feasible method for predicting and evaluating the applicability of transfer learning methods in real-world scenarios.

Engine fault detection is conducive to improving equipment reliability and reducing maintenance costs. In practical scenarios, high-quality data is difficult to obtain. Usually, only single-sensor data is available. This paper proposes a fault detection method combining Variational Mode Decomposition (VMD) and Random Forest (RF). At first, the spectral energy distribution is obtained by decomposing and statistic the engine data of multiple working conditions. Based on the spectral energy distribution, the overall optimal mode number was identified, and the quadratic penalty term was optimized using SNR. The improved VMD (IVMD) improves mode aliasing and iterative efficiency and unifies feature dimensions. Decomposition of real signals demonstrates the effectiveness. The paper designs a feature vector composed of seven types of attributes, including unit bandwidth energy, center frequency, maximum singular value and so on. The feature vector is then fed to RF for classification. Features are selected in order of importance to classification to improve the training efficiency. By comparing with various algorithms, the proposed method has higher accuracy and faster training efficiency in single-speed, multi-speed and cross-speed single-sensor data diagnosis. The results show that the method has application prospects with little training data and low hardware requirements.