The rotating machinery is continuously monitored in practical application. However, the historical life-cycle data cannot be always preserved due to the limited storage resource; meanwhile, the on-site computing platform cannot process a large number of monitoring samples. It brings a great challenge for the remaining useful life (RUL) prediction. Thus, continuous learning (CL) is introduced into RUL prediction model for achieving its knowledge accumulation and dynamic update. To improve the performance of continuous RUL prediction, this article presents a new RUL prediction methodology with a multistage attention convolutional neural network (MSACNN) and knowledge weight constraint (KWC). First, an improved multihead full-channel sight self-attention (MFCSSA) mechanism is proposed to capture the global degradation information across all channels. MSACNN is then constructed by embedding MFCSSA, squeeze-and-excitation (SE) mechanism, and convolutional block attention module (CBAM) into different stages of feature extraction, which enables it to capture the global degradation information and refine the feature representations progressively. The KWC mechanism based on the importance of weight parameters and gradient information is proposed and integrated into MSACNN to achieve the continuous RUL prediction task. The proposed KWC can effectively alleviate catastrophic forgetting in CL. Finally, the experimental results on the life-cycle bearing and gear datasets demonstrate that MSACNN has a higher accuracy than the existing prediction methods. Moreover, the KWC mechanism performs better than typical CL methods in retaining the previously learned knowledge while acquiring the new task knowledge. Therefore, the proposed methodology can be better applied to the continuous RUL prediction tasks than the advanced methods of the same kind.

Aiming at the large differences between tasks in continuous remaining useful life (RUL) prediction and the limited information capturing capability of the existing continuous learning (CL) methods, this article develops a novel multibranch horizontal augmentation network (MBHAN). First, a hierarchical self-attention (HSA) mechanism is proposed to capture the local degradation features and dependencies at different scales and enhance the representation capacity of RUL prediction model. Based on HSA and temporal convolutional network (TCN), a time-frequency fusion TCN (TFFTCN) is designed to mine the hidden degradation information from the time-domain and frequency-domain data. Then, a memory weight constraint (MWC) regularization term is built to control the update of important parameters for pervious tasks during the learning of new task. A horizontal network augmentation rule based on the task similarity and MWC is proposed, including the augmentation of a task branch network for small task difference and the augmentation of a feature extraction backbone network for large task difference. On this basis, the MBHAN is proposed to continuously predict RUL of machinery. Finally, the experimental results on the life-cycle bearing and gear datasets demonstrate that TFFTCN achieve an average accuracy of 93% across both datasets, surpassing the existing prediction methods.

To accurately predict the remaining useful life (RUL) of rotating machinery while continuously providing the task data, a novel continuous RUL prediction methodology was proposed. The methodology comprises a self-guided attention convolutional neural network (SGACNN) and memory consciousness adjustment (MCA) mechanism. First, a multihead focal channel-wise self-attention (MFCWSA) mechanism was implemented to effectively capture the degradation information across all the channels and achieve the attentional focus. Next, the SGACNN was constructed using the MFCWSA, squeeze-and-excitation mechanism, and convolutional block attention module. A new network gradient direction was synthesized by leveraging the gradients from both the previous task and the current task. Further, a weight constraint loss term based on the gradient magnitude was designed to constrain the learning process of important parameters. With the new network gradient direction and weight constraint loss, a novel MCA mechanism was proposed and integrated into the SGACNN for implementing the continuous RUL prediction tasks. Finally, various RUL prediction experiments on the life-cycle bearing and gear data sets were carried out, and its outcomes were compared to those of the advanced methods of the same kind. The comparative results validated the superiority of the proposed methodology.

Many domain adaptation (DA) models have been explored for fault transfer diagnosis. However, their successes completely rely on the availability of target-domain samples during the training process. As target domain is usually unseen, the domain-adaptation-based diagnostic models cannot meet the requirement of real-time diagnosis in actual engineering. To achieve the domain confusion in the actual diagnosis scenario, a novel relationship transfer (RT) diagnosis framework is first proposed, which can indirectly measure and reduce the distribution discrepancy between the source domain and unseen target domain. Based on the proposed RT framework, a new domain generalization transfer method, called relationship transfer domain generalization network (RTDGN) is constructed. RTDGN is divided into two phases including task-irrelevant domain adaptation (TIDA) and task-relevant domain generalization (TRDG). In the TIDA phase, a DA adversarial network with several domain discriminators is built to enhance the domain confusion of RT framework. Furthermore, to bring the adversarial network a more general domain confusion ability, a new inverse entropy loss is designed. In the TRDG phase, a residual fusion classifier is constructed to improve the generalization ability of fault classifier. Finally, the experimental results on the wind turbine planetary gearbox dataset and bearing dataset verify the effectiveness and superiority of the proposedRTDGN.

Remaining useful life (RUL) prediction can provide a foundation for the operation and maintenance of industrial equipment. In order to improve the predictive ability for the complex degradation trajectory, a new dual-thread gated recurrent unit (DTGRU) is explored. It uses a dual-thread learning strategy to mine the stationary and nonstationary information from the input data and the difference of hidden states at two adjacent time steps. Then the state transition updating formulas of DTGRU are derived. Using the collected gear vibration signals and degradation-trend-constrained variational autoencoder, the gear health indicator (HI) is constructed. Based on the constructed HI and DTGRU, a novel RUL prediction method is developed. Via multiple gear life-cycle datasets, the effectiveness of the DTGRU-based RUL prediction approach is verified. Furthermore, compared with the existing typical prediction methods, the experimental results show that DTGRU has higher predictive ability in terms of HI fitting precision and RUL prediction performance.

Facing the gap in the unsupervised construction of health indicator (HI) with a uniform failure threshold, a new unsupervised HI construction approach is developed. First, the distribution of the raw vibration signal is estimated by the Gaussian mixture model, then a distribution contact ratio metric (DCRM) is designed to compute the distance between two arbitrary distributions. With DCRM, a distribution contact ratio metric health indicator (DCRHI) is innovatively constructed for well representing the degradation process and obtaining a uniform failure threshold. Next, aiming at the challenge of prediction under limited samples, a novel consolidated memory gated recurrent unit (CMGRU) is proposed by making full use of the historical state information, and it can effectively slow down the forgetting speed of important trend information. Combing the proposed DCRHI and CMGRU, a novel remaining useful life (RUL) prediction methodology is put forward for enhancing the predictive performance. Via two public bearing datasets, several contrast experiments are implemented, and the comparative results show that DCRHI can better describe the degradation process of bearing than other typical unsupervised HIs, and CMGRU has a stronger prediction ability than other classical time series processing networks. Thus, the proposed methodology has great application value in the RUL prediction.

In constructing a smart court, to provide intelligent assistance for achieving more efficient, fair, and explainable trial proceedings, we propose a full-process intelligent trial system (FITS). In the proposed FITS, we introduce essential tasks for constructing a smart court, including information extraction, evidence classification, question generation, dialogue summarization, judgment prediction, and judgment document generation. Specifically, the preliminary work involves extracting elements from legal texts to assist the judge in identifying the gist of the case efficiently. With the extracted attributes, we can justify each piece of evidence’s validity by establishing its consistency across all evidence. During the trial process, we design an automatic questioning robot to assist the judge in presiding over the trial. It consists of a finite state machine representing procedural questioning and a deep learning model for generating factual questions by encoding the context of utterance in a court debate. Furthermore, FITS summarizes the controversy focuses that arise from a court debate in real time, constructed under a multi-task learning framework, and generates a summarized trial transcript in the dialogue inspectional summarization (DIS) module. To support the judge in making a decision, we adopt first-order logic to express legal knowledge and embed it in deep neural networks (DNNs) to predict judgments. Finally, we propose an attentional and counterfactual natural language generation (AC-NLG) to generate the court’s judgment.

Domain adaptation has been widely used for knowledge transfer. However, the aligning targets of the existing domain adaptation mechanisms dynamically vary during the training, which leads to the loss oscillation, slow convergence, and poor robustness. To overcome this main problem, a novel domain adaptation mechanism named intermediate distribution alignment (IDA) is proposed. For implementing the end-to-end diagnostic tasks, a feature extractor based on deep convolutional neural network with wide first-layer kernel is first built to fit the posterior distributions of source and target domains. Then through the KL divergence, IDA maps the learned features from the source and target domains into a specific intermediate distribution. It is proved theoretically that IDA can align the prior distributions of two domains. The proposed IDA mechanism is successfully applied to the fault transfer diagnosis of planetary gearboxes without labeled target-domain samples. The comparative results show that the proposed IDA mechanism has higher diagnostic performance than the typical domain adaptation mechanisms.

Users' clicks on Web search results are one of the key signals for evaluating and improving web search quality and have been widely used as part of current state-of-the-art Learning-To-Rank(LTR) models. With a large volume of search logs available for major search engines, effective models of searcher click behavior have emerged to evaluate and train LTR models. However, when modeling the users' click behavior, considering the bias of the behavior is imperative. In particular, when a search result is not clicked, it is not necessarily chosen as not relevant by the user, but instead could have been simply missed, especially for lower-ranked results. These kinds of biases in the click log data can be incorporated into the click models, propagating the errors to the resulting LTR ranking models or evaluation metrics. In this paper, we propose the De-biased Reinforcement Learning Click model (DRLC). The DRLC model relaxes previously made assumptions about the users' examination behavior and resulting latent states. To implement the DRLC model, convolutional neural networks are used as the value networks for reinforcement learning, trained to learn a policy to reduce bias in the click logs. To demonstrate the effectiveness of the DRLC model, we first compare performance with the previous state-of-art approaches using established click prediction metrics, including log-likelihood and perplexity. We further show that DRLC also leads to improvements in ranking performance. Our experiments demonstrate the effectiveness of the DRLC model in learning to reduce bias in click logs, leading to improved modeling performance and showing the potential for using DRLC for improving Web search quality.

In search and recommendation, diversifying the multi-aspect search results could help with reducing redundancy, and promoting results that might not be shown otherwise. Many previous methods have been proposed for this task. However, previous methods do not explicitly consider the uniformity of the number of the items' classes, or evenness, which could degrade the search and recommendation quality. To address this problem, we introduce a novel method by adapting the Simpson's Diversity Index from biology, which enables a more effective and efficient quadratic search result diversification algorithm. We also extend the method to balance the diversity between multiple aspects through weighted factors and further improve computational complexity by developing a fast approximation algorithm. We demonstrate the feasibility of the proposed method using the openly available Kaggle shoes competition dataset. Our experimental results show that our approach outperforms previous state of the art diversification methods, while reducing computational complexity.

To support complex search tasks, where the initial information requirements are complex or may change during the search, a search engine must adapt the information delivery as the user’s information requirements evolve. To support this dynamic ranking paradigm effectively, search result ranking must incorporate both the user feedback received, and the information displayed so far. To address this problem, we introduce a novel reinforcement learning-based approach, RLIRank. We first build an adapted reinforcement learning framework to integrate the key components of the dynamic search. Then, we implement a new Learning to Rank (LTR) model for each iteration of the dynamic search, using a recurrent Long Short Term Memory neural network (LSTM), which estimates the gain for each next result, learning from each previously ranked document. To incorporate the user’s feedback, we develop a word-embedding variation of the classic Rocchio Algorithm, to help guide the ranking towards the high-value documents. Those innovations enable RLIRank to outperform the previously reported methods from the TREC Dynamic Domain Tracks 2017 and exceed all the methods in 2016 TREC Dynamic Domain after multiple search iterations, advancing the state of the art for dynamic search.