In this study, we consider surfacing control problems for the autonomous underwater vehicle in three-dimensional space under emergencies. Most model-based controllers cannot effectively solve these problems due to the unknown task environment model. Moreover, existing deep reinforcement learning (DRL) methods have some limitations, such as slow convergence or overestimation bias. For these purposes, we propose a model-free DRL algorithm based on the Deep Deterministic Policy Gradient within the paradigm of deep learning as a service (DLaaS). The algorithm combines existing expert episodes as demonstration and transitions from the interaction between the agent and task environment, which potentially reduce risk from adversarial attack. We propose a variable delay learning mechanism to improve the performance of the proposed algorithm. The simulation results show that our method can converge faster and has a more robust performance with adversarial attack than the existing DRL method under the premise of solving the AUV surfacing control tasks for safety.

Intelligent fault diagnosis of rolling bearings has advanced significantly with the increase in labeled industrial data. However, the limited data for unknown working conditions poses a challenge to the generalization capabilities of current deep learning methods. Therefore, this article proposes a novel approach to domain generalization, leveraging a combination of covariance loss and graph convolutional networks to realize feature augmentation for intelligent fault diagnosis. This method employs random receptive field layers in feature extractors to project inputs from each source domain into distinct feature spaces. Moreover, a covariance loss is incorporated to ensure the dissimilarity of feature representations. Consequently, the augmented features contribute to the construction of an expanded adjacency matrix and prototypes within graph convolutional networks, thereby enhancing the model's capacity to generalize to unknown domains. Results on both a public dataset and an experimental dataset of rolling bearings have shown the superiority of the proposed approach.

A proton exchange membrane fuel cell (PEMFC) has great application prospects due to its low emission and high efficiency. An accurate model to predict the dynamic output voltage is essential for the optimal control of the PEMFC for the applications on vehicles and power stations. In this article, a novel deep learning framework with the long short-term memory (LSTM) and artificial neural network (ANN) fusion is proposed to develop the PEMFC dynamic model by extracting both the historical and current information. The LSTM extracts the temporal information from the past PEMFC states with its order determined with autocorrelation and partial autocorrelation functions, while the influence of the system current inputs is learnt by the ANN. Then, the outputs of LSTM and ANN are concatenated with the multiple information fused to predict the PEMFC dynamic output voltage. After validated by the operating data from a laboratory-scale PEMFC system, the LSTM and ANN fusion model is compared with the existing models, such as support vector regression, ANN, and LSTM methods. The comparison results show that the proposed LSTM and ANN fusion model can provide the best prediction performance with the lowest mean square error of 1.303. The proposed LSTM and ANN fusion model can be helpful to develop the optimal control strategy of the PEMFC.

Currently, with the development of the Internet of Things (IoTs) and artificial intelligence, a new IoT structure known as the artificial Intelligence of Things (AIoTs) comes into play. With the development of AIoT, a large amount of unlabeled industrial big data has been accumulated. The analysis of large amounts of unlabeled data is labor-intensive and time-consuming for diagnostic personnel. To improve this situation, a novel two-stage unsupervised fault recognition algorithm, namely, deep adaptive fuzzy clustering algorithm (DAFC) is proposed for unsupervised fault clustering in this article. DAFC amalgamates stacked sparse autoencoder (SSAE) into adaptive weighted Gath–Geva (AWGG) clustering to form an unsupervised fault recognition framework for clustering analysis of unlabeled industrial big data. SSAE can extract the highly abstract features of the original data, and adopt different unsupervised strategies to fine-tune the network in two stages. AWGG is an improvement of Gath–Geva clustering, and can adaptively obtain optimal clustering results without presetting the number of clusters. Experimental results on two different datasets show that the proposed DAFC can stably extract fault features from unlabeled data, and automatically obtain the optimal clustering results without knowing the number of clusters in advance. To the best of our knowledge, this article is the first attempt to fine-tune SSAE in an unsupervised manner, and to propose an unsupervised fault recognition framework that requires no prior knowledge or data labels at all. DAFC can be a feasible industrial big data application for collaborative AIoT. Diagnostic personnel analyze the clustering results obtained by DAFC instead of the original unlabeled data, greatly saving time and labor costs.

Massive industrial data collected from the Industrial Internet-of-Things (IIoT) assets improve data-driven methods for prognostics and health management (PHM) systems. As an important role in PHM, remaining useful life (RUL) prediction is essential to maintain the reliability and safety of industrial manufacture. However, recent data-driven approaches for bearing RUL prediction do not weight the contributions of data from different sensors and time steps, which decreases the efficiency in the big data era. In this context, we present a deep learning-based RUL prediction method with an attention mechanism to weight sequence data representations. Specifically, the proposed method weights different industrial sensors and time steps, respectively, based on the distributed attention mechanism. Then, temporal convolution modules with the shared weights are used for feature extraction of time series. Finally, the performance of the proposed method is verified using a popular data set C-MAPSS. The experimental results reveal that the proposed method has high accuracy and is efficient in practical applications.

Nowadays, the industrial Internet of Things (IIoT) has been successfully utilized in smart manufacturing. The massive amount of data in IIoT promote the development of deep learning-based health monitoring for industrial equipment. Since monitoring data for mechanical fault diagnosis collected on different working conditions or equipment have domain mismatch, models trained with training data may not work in practical applications. Therefore, it is essential to study fault diagnosis methods with domain adaptation ability. In this article, we propose an intelligent fault diagnosis method based on an improved domain adaptation method. Specifically, two feature extractors concerning feature space distance and domain mismatch are trained using maximum mean discrepancy and domain adversarial training respectively to enhance feature representation. Since separate classifiers are trained for feature extractors, ensemble learning is further utilized to obtain final results. Experimental results indicate that the proposed method is effective and applicable in diagnosing faults with domain mismatch.

Industrial Internet of Things (IIoT) obtains big data from industrial facilities. Based on these data, health conditions for facilities can be predicted using machine learning methods, which in turn improves the trustworthiness of IIoT. Intelligent fault diagnosis is developing with this process. Since machine damages always happen under different circumstances, the designed intelligent fault diagnosis methods should have domain adaption ability. In this article, a data-driven fault diagnosis method called domain adaption network (DAN) is present, in which the difference between training and testing data can be minimized while the maximum training accuracy can be maintained. First, DAN uses a combined loss for consistent feature learning and optimum training classification performance. Then, a DAN retraining (DAN-R) strategy is employed based on weighted pseudo-labeled testing dataset. Finally, three experiments are performed on two fault diagnosis datasets. The results reveal that DAN-R is applicable and has outstanding domain adaption ability.