Dependency-based models are widely used to extract semantic relations in text. Most existing dependency-based models establish stacked structures to merge contextual and dependency information, which encode the contextual information first and then encode the dependency information. However, this unidirectional information flow weakens the representation of words in the sentence, which further restricts the performance of existing models. To establish bidirectional information flow, a dual attention graph convolutional network (DAGCN) with a parallel structure is proposed. Most importantly, DAGCN can build multi-turn interactions between contextual and dependency information to imitate the multi-turn looking-back actions of human beings. In addition, multi-layer adjacency matrix-aware multi-head attention (AMAtt), including context-to-dependency attention and dependency-to-context attention, is carefully designed as a merge mechanism in the parallel structure to preserve the structural information of sentences and dependency trees during interactions. Furthermore, DAGCN is evaluated on the popular PubMed dataset, TACRED dataset and SemEval 2010 Task 8 dataset to demonstrate its validity. Experimental results show that our model outperforms the existing dependency-based models.

Quality prediction is committed to predicting the key quality-related variables to obtain real-time feedback information for process control. To achieve the robust and transferable quality prediction of products processed in complex and uncertain manufacturing processes, deep learning methods have been developed. However, the training process of deep learning methods requires a large amount of annotated data to avoid overfitting, and the labeling process of quality-related variables is often time-consuming and labor-intensive. Therefore, few-shot quality prediction in a multistage manufacturing process is formalized to address the lack of annotated data and deal with previously unseen tasks without an additional training process. In addition, a novel contrastive decoder generator (CDG) is proposed to enable few-shot quality prediction, which consists of a machine feature encoder, a contrastive stage, and task feature generator, and an instance-specific decoder generator. Experiments are conducted on a public quality prediction dataset collected from an actual production line. The CDG achieves state-of-the-art results on this dataset for few-shot quality prediction settings, which proves the effectiveness of the CDG. Additionally, detailed experiments are performed to evaluate the roles of different modules in the CDG.

The class imbalance problem has a huge impact on the performance of diagnostic models. When it occurs, the minority samples are easily ignored by classification models. Besides, the distribution of class imbalanced data differs from the actual data distribution, which makes it difficult for classifiers to learn an accurate decision boundary. To tackle the above issues, this article proposes a novel imbalanced data classification method based on weakly supervised learning. First, Bagging algorithm is employed to sample majority data randomly to generate several relatively balanced subsets, which are then used to train several support vector machine (SVM) classifiers. Next, these trained SVM classifiers are adopted to predict the labels of those unlabeled data, and samples that are predicted as minority class are added to the original dataset to reduce the imbalance ratio. The critical idea of this article is to introduce real-world samples into the imbalanced dataset by virtue of weakly supervised learning. In addition, bidirectional gated recurrent units are used to construct a diagnostic model for fault diagnosis, and a new weighted cross-entropy function is proposed as the loss function to reduce the impact of noise. Besides, it also increases the model's attention to the original minority samples. Furthermore, experimental evaluations of the proposed method are conducted on two datasets, i.e., Prognostics and Health Management challenge 2008 and 2010 datasets, and the experimental results demonstrate the effectiveness and superiority of the proposed method.

Quality prediction, as the basis of quality control, is dedicated to predicting quality indices of the manufacturing process. In recent years, data-driven deep learning methods have received a lot of attention due to their accuracy, robustness, and convenience for the prediction of quality indices. However, the existing studies mainly focus on the quality prediction of a single machine, while ignoring dependency relationships among multiple machines in multistage manufacturing process. To tackle the above issues, a novel path enhanced bidirectional graph attention network (PGAT) is proposed in this article. PGAT models the dependencies among machines into directed graphs and introduces graph attention network to encode the dependencies. Nonetheless, it is difficult for graph neural networks to encode long-distance dependencies. Hence, dependency path information is introduced into the features of machines. Moreover, in order to solve the label noise problem that often occurs in actual industrial dataset, a masked loss function is devised. With its help, batch training with noisy labels can be achieved, which improves the training efficiency. Furthermore, experiments are conducted on a public quality prediction dataset collected from an actual production line. PGAT achieves the state-of-the-art results on this dataset, which confirms the effectiveness of PGAT. Additionally, the experimental results demonstrate the critical role of modeling dependency relationships among machines.

Deep learning plays an increasingly important role in industrial applications, such as the remaining useful life (RUL) prediction of machines. However, when dealing with multifeature data, most deep learning approaches do not have effective mechanisms to weigh the input features adaptively. In this article, a novel feature-attention-based end-to-end approach is proposed for RUL prediction. First, the proposed feature-attention mechanism is directly applied to the input data, which gives greater attention weights to more important features dynamically in the training process. This helps the model focus more on those critical inputs, and the prediction performance is therefore improved. Next, bidirectional gated recurrent units (BGRU) are used to extract long-term dependencies from the weighted input data, and convolutional neural networks are employed to capture local features from the output sequences of BGRU. Finally, fully connected networks are used to learn the above-mentioned abstract representations to predict the RUL. The proposed approach is validated in a case study of turbofan engines. The experimental results demonstrate that the proposed approach outperforms other latest existing approaches.

A novel encoder-decoder model based on deep neural networks is proposed for the prediction of remaining useful life (RUL) in this work. The proposed model consists of an encoder and a decoder. In the encoder, the Bi-directional Long Short-Term Memory Networks (Bi-LSTM) and Convolutional Neural Networks (CNN) are used to capture the long-term temporal dependencies and important local features from the sequential data, respectively. Besides, single 1*1 convolution filter in the last convolutional layer is used for dimensionality reduction. In the decoder, the fully connected networks are employed to decode the feature information to predict RUL. In addition, the proposed data-driven method can achieve end-to-end prediction, which does not need feature engineering. To evaluate the proposed model, experimental verification is carried out on a commonly used aero-engine C-MAPSS dataset. Compared with other state-of-the-art approaches on the same dataset, the effectiveness and superiority of the proposed framework are demonstrated. For example, the scoring function value of the second subset is reduced by up to 64.99% compared with the best existing result.