Most rolling element bearing (REB) fault diagnosis algorithms are evaluated on the Case Western Reserve University (CWRU) bearing dataset for its popularity and simplicity. However, the diagnosis accuracy on CWRU bearing dataset is overly saturated; it is nearly up to 100%. In this study, an input feature mappings (IFMs)-based deep residual network (ResNet) is proposed to conduct detailed and comprehensive fault diagnosis on REB with complicated bearing dataset. Firstly, a new data preprocessing method named as a signal-to-IFMs method is proposed to automatically extract features from raw signals without predefined parameters. Then, a deep ResNet is used as the fault classifier to learn the discriminative features from IFMs and identify the faults of REB. Finally, the proposed model is evaluated on the artificial, real, and mixed damages of the Paderborn university bearing dataset. The proposed method yields the average testing accuracies of 99.7%, 99.7%, and 99.81% in artificial, real, and mixed bearing damages, which outperforms other methods.

In this paper, a Self-learning Collective Intelligence Differential Evolution (SLCIDE) algorithm was proposed to optimize both the architecture and parameters of a Feedforward Neural Network (FNN). In order to improve the exploration-exploitation capability, a new Collective Intelligence (CI) based mutation operator was proposed by mixing some promising donor vectors in the current population. Besides, a self-learning mechanism which was designed to adaptively select m top ranked donor vectors was developed by using a widely used unsupervised learning method, k-means. As a result, the proposed approach can be more adaptive and statistically powerful on versatile problems. Then, we evaluated the performances of the proposed SLCIDE approach by studying some numerical optimization problems of CEC 2013 with D = 30 and D = 50. Further, the proposed SLCIDE method was applied to train a FNN on four most popular datasets, resulting in very competitive performances. The comprehensive experimental results have demonstrated that the presented SLCIDE method obtain better results compared with other state-of-the-art algorithms.

With the growing demand for emission reductions and fuel efficiency improvements, alternative energy sources and energy storage technologies are becoming popular in a ship microgrid. In order to balance the two non-compatible objectives, a new differential evolution variant, which is named as SaCIDE-r, was proposed to solve the optimization problem. In this algorithm, a Collective Intelligence (CI) based mutation operator was proposed by mixing some promising donor vectors in the current population. Besides, a self-adaptive mechanism which was developed to avoid introducing extra control parameters. Further, to avoid being trapped in local optima, a re-initialization mechanism was developed. Then, we have evaluated the performances of the proposed SaCIDE-r approach by studying some numerical optimization problems of Congress on Evolutionary Computation (CEC) 2013 with D = 30, compared with seven stateof-the-art DE algorithms. Moreover, the proposed SaCIDE-r method was applied for economic scheduling of a shipboard microgrid under different cases compared with other multi-objective optimizing methods, resulting in very competitive performances. The comprehensive experimental results have demonstrated that the presented SaCIDE-r method might be a feasible solution for such a kind of optimization problem.

Marine diesel engines are the preferred power equipment for ships and are the most important component among the numerous electromechanical devices on board. Accidents involving these engines can potentially cause immeasurable damage to the vessel, making fault detection in marine diesel engines crucial. This design enables the detection and reporting of faults in marine diesel engines at the earliest possible time through the computation of convolutional neural networks, which is of great significance for ensuring the safe navigation of ships. For this functionality, the Xilinx ZYNQ-7000 XC7Z010 is selected as the main control chip, and the LoRa wireless network is used as the transmission module. The FreeRTOS embedded operating system is ported, with sensor data collection completed on the PS side of the ZYNQ chip and algorithm acceleration calculations on the PL side. Data are then transmitted to the host computer via the LoRa module paired with a custom protocol. Experimental test results show that the program provides stable data transmission, with each module of the algorithm generally accelerating by more than 95% and an accuracy rate of 92.86%. Additionally, the host computer can display the received data in real time. The custom protocol’s header also allows for precise judgments about the completeness and origin of messages, facilitating the expansion of other SOC’s message uplink and the host computer’s message downlink.

The reliability of marine diesel engines is crucial to the safety of ships. However, condition monitoring data under actual working conditions often exhibit multi-source heterogeneity, including frequency-domain features of vibration signals and statistical indicators of cylinder pressure, and are often constrained by class imbalances and strong noise interference. To address these challenges, this paper proposes a dual-path fusion diagnostic model CNN-MLP-CBAM, which integrates a one-dimensional convolutional neural network (CNN), a multilayer perceptron (MLP), and a convolutional block attention module (CBAM). The core innovation of this study lies in constructing an asymmetric dual-path structure: it breaks the limitation of conventional single-network input, utilises dedicated branches to process time-series waveforms and statistical features in parallel, and then performs deep feature fusion; meanwhile, adaptive synthetic sampling (ADASYN) is introduced to generate minority class samples to eliminate data bias. Experimental results on a public diesel engine fault dataset indicate that the model achieves diagnostic accuracies of 99.43% and 99.71% at 15db and 30db signal-to-noise ratios, respectively. Compared with mainstream baseline models such as LSTM and SVM, this method improves diagnostic accuracy by more than 5% under strong noise conditions, demonstrating outstanding robustness and potential for engineering applications.