Nested named entity recognition aims to identify multi-granularity entities with hierarchical inclusion relationships in text, and is a core task in the field of information extraction. To address the boundary insensitivity issues caused by pooling or concatenation operations in existing span-based methods, as well as the problem of error propagation in cascaded architectures, this paper proposes a label-aware global pointer model (LAPointer). This model embeds entity category information into the text representation via a label-aware module and utilises a global pointer mechanism to construct a cross-category correlation matrix for precise entity boundary localisation; simultaneously, a classification-balanced loss function is introduced to mitigate the issue of class imbalance. Experimental results on the ACE04, ACE05 and GENIA benchmark datasets demonstrate that LAPointer achieves superior performance to existing methods in nested entity recognition tasks, exhibiting particularly strong boundary discrimination capabilities when handling complex multi-level nested structures.

Machining accuracy of machine tools is a core indicator for measuring the level of manufacturing equipment, directly determining component quality and overall machine performance. However, various errors such as thermal and wear during the machining process significantly affect machine tool accuracy. In recent years, digital twin-driven data modeling methods have shown significant advantages in error prediction and closed-loop feedback, yet challenges remain in the compensation stage, including difficulty in ensuring sensing data quality, inability to correct compensation delays, and static and rigid compensation strategies, which correspond to the bottlenecks of accuracy, timeliness, and intelligent adaptive compensation, respectively. To address these issues, this study proposes a digital twin-based timely self-healing framework for machining accuracy. First, a cascaded “prediction-identification-correction” strategy is designed to obtain high-fidelity sensing data. Then, a cross-scale time window feature aggregation method is developed to accurately deduce key physical characteristics of machine tools, providing forward time for compensation decision-making. Finally, a geometric error propagation model incorporating reverse clearance is constructed to evaluate the positioning capability through a dual-stage qualitative and quantitative strategy, enabling intelligent compensation command generation to achieve accuracy self-healing. On a self-constructed dataset containing anomalies, the proposed cascaded correction framework improves data quality by at least 36.5%. The proposed forward deduction method achieves at least 5.7% improvement compared with six mainstream methods on the self-constructed feed axis thermal characteristic dataset and the PHM2010 wear characteristic dataset. Furthermore, case studies on multi-axis machine tools validate the effectiveness of the intelligent adaptive compensation method.

Epoxidized natural rubber (ENR) is a high-performance natural rubber derivative with superior elasticity, oil resistance and gas barrier properties, showing great potential in high-end seals and intelligent tires. Accurate control of its self-healing performance is essential for high-end applications, but traditional trial-and-error experiments are inefficient in revealing multi-parameter coupling effects and structure–property relationships. In this work, a dataset was constructed using full factorial experiments with six core variables. Cheminformatics tools were employed to achieve micro–macro multiscale correlation. The hybrid screening method was used to identify key features, and linear regression (LR), support vector regression (SVR), and extreme gradient boosting (XGBoost) models were established. Results showed that XGBoost achieved a test-set coefficient of determination (R2) of 0.956 for tensile strength before repair (TSBR), approximately 0.9 for tensile strength after repair (TSAR), and 0.837 for self-healing efficiency (SHE), which outperformed LR and SVR. SHapley Additive exPlanations (SHAP) values interpreted feature mechanisms, and the proposed experiment–data–model–interpretation framework supports efficient formulation design.

Existing cross-domain mechanical fault diagnosis methods primarily achieve feature alignment by directly optimizing interdomain and category distances. However, this approach can be computationally expensive in multi-target scenarios or fail due to conflicting objectives, leading to decreased diagnostic performance. To avoid these issues, this paper introduces a novel method called domain feature disentanglement. The key to the proposed method lies in computing domain features and embedding domain similarity into neural networks to assist in extracting cross-domain invariant features. Specifically, the neural network architecture designed based on information theory can disentangle key features from multiple entangled latent variables. It employs the concept of contrastive learning to extract domain-relevant information from each data point and uses the Wasserstein distance to determine the similarity relationships across all domains. By informing the neural network of domain similarity relationships, it learns how to extract cross-domain invariant features through adversarial learning Eight multi-target domain adaptation tasks were set up on two public datasets, and the proposed method achieved an average diagnostic accuracy of 96.82%, surpassing six other advanced domain adaptation methods, demonstrating its superiority.

The strong noise existing in the vibration signals has a negative impact on rotating machinery fault diagnosis. To solve the noise problem in the engineering applications of fault diagnosis, a deep residual network, named nonnegative garrote shrinkage network with adaptive Swish (NNGSN-AS), is proposed for rotating machinery fault diagnosis under noisy environment. In the NNGSN-AS, the non-negative garrote shrinkage function (NNGSF) is integrated into residual building blocks as nonlinear transformation layers, and the residual building block is named the non-negative garrote shrinkage building unit (NNGSBU). In the NNGSBU, the threshold of the NNGSF is adaptively learned by the thresholding module, so that different thresholds can be assigned to different data samples. The thresholding module is close to the NNGSBU input, enabling early noise handling. The depthwise convolutions with wide kernels in the thresholding module increase the receptive field and lead to a one-to-one correspondence between the learnable threshold and the input feature elements of the NNGSBU, reducing the negative influence of the noise. Additionally, an adaptive Swish (ASwish) activation function module is developed, enabling adaptive nonlinear transformation of each feature channel. The experimental results on a public dataset and our laboratory dataset indicate that the NNGSN-AS is superior to the existing methods for rotating machinery fault diagnosis under noisy environment. Given that the performance gain of the NNGSN-AS stems from multiple components for deep feature extraction, the ablation experiments are conducted to demonstrate the improvement effect of each component.

Accurate monitoring of tool wear states and wear values is crucial for reducing machine tool failures and ensuring machining accuracy and efficiency. However, wear monitoring faces significant challenges due to the imbalance of wear samples and the dynamic changes in the coupling relationships among multi-source sensing signals. Additionally, varying processing conditions further complicate the accurate tracking of wear. To address these challenges, an evolutionary spatio-temporal parallel network model is proposed. The model first employs a cyclic consistency classification enhancement network to accurately identify the real-time wear state of the tool. Then, it utilizes a parallel network to uncover the spatio-temporal coupling relationships within multi-source sensing data. Based on this, an evolutionary monitoring mechanism drives the continuous evolution and update of the model, adapting to real-time wear state and working condition changes, thus achieving precise tool wear monitoring under variable working conditions. Our self-built grinding wheel wear dataset and PHM2010 milling public dataset are used to verify the effectiveness of the method. Experimental results demonstrate that the proposed method improves prediction accuracy by 55.85 %, 10.26 %, and 50.14 % over existing methods on the C1, C4, and C6 datasets of PHM2010, respectively, while achieving a remarkable accuracy advantage of over 96.63 % in grinding wheel wear prediction.

Tool wear prediction is vital for enhancing machining accuracy and ensuring production safety. However, challenges arise from non-processing data interference and missing tool wear samples, complicating the construction of accurate prediction models. Additionally, the complexity of collaborative multi-tool operations on precision computer numerical control (CNC) machine tools, where varying tool types and complex working conditions exist, further exacerbates the difficulty of achieving precise wear prediction. To address these challenges, this paper introduces a digital twin architecture for tool wear prediction, based on knowledge embedding. The proposed architecture is designed to predict the wear of multiple tools, incorporating modules for processing data screening, missing value completion, wear state classification, and so on. On the basis of obtaining high- quality sensing data and complete tool wear values, the wear state and machining process knowledge are embedded into the prediction process. A tool wear prediction model is then constructed based on a KolmogorovArnold integrated time convolutional network (KA-TCN), so as to achieve accurate prediction of multi-tool wear. The effectiveness of the method is validated using data from two grinding wheel wear test platforms and two milling datasets, PHM2010 and NASA. Experimental results demonstrate that the knowledge embedded KA-TCN model outperforms existing approaches, improving prediction accuracy by over 22.4 % on the milling dataset, and by 76.4 % in grinding wheel wear prediction compared to classical methods.

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.

Real-time data may undergo distribution drift due to changes in operating conditions and other factors, which can affect the classification accuracy of online fault diagnosis models and potentially lead to serious consequences. Therefore, understanding the classification accuracy of the model on real-time data holds substantial significance. However, the absence of labels in real-time data presents a challenge for evaluating classification accuracy. Furthermore, the real-time nature of fault diagnosis necessitates a swift and straightforward evaluation method. For the above reasons, this paper proposes a method for evaluating the classification accuracy of a model on real-time data, which is done in the absence of labels for the real-time data. The proposed label-free evaluation method transforms the model’s output into a scalar that measures the degree of matching between the classification probabilities, termed the average free energy. It then establishes a mapping between the average free energy and the classification accuracy using an auxiliary dataset. Finally, it predicts the model’s classification accuracy on the real-time data through this mapping and the average free energy of the real-time data. The proposed method is experimentally evaluated on public datasets, demonstrating its superiority in various aspects.

With the rapid development of sensing technology, deep learning (DL) methods have gained great popularity in remaining useful life (RUL) prediction of equipment. However, DL-based prediction methods generally cannot quantify uncertainties or provide reliable prediction intervals (PIs) for RUL. To tackle this problem, this article proposes an ensemble framework for uncertainty quantification and interval prediction of RUL based on semisupervised learning. First, the prediction variance is used to represent the aleatoric uncertainty, and a weighted negative log-likelihood loss function is designed to help the DL model to provide both the prediction value and the variance, thereby quantifying the aleatoric uncertainty in RUL prediction. For the issue that variance labels are difficult to obtain, the DL model is trained in a semisupervised manner to learn the prediction variance caused by data noise. Next, the Bootstrap method is applied to sample the training data with replacement to obtain several subsets, and several DL models are trained by these subsets, thereby quantifying the epistemic uncertainty in RUL prediction. Furthermore, PIs with 95% confidence are calculated based on these uncertainties to achieve RUL interval prediction of equipment. Finally, experiments are conducted on the turbofan engine dataset and CNC machine tool dataset, and the validity of the proposed method is confirmed.

Conceptual design plays an important role in determining the basic characteristics and final product performances. However, there are many uncertainties in the conceptual design, such as the vagueness of customer requirements, uncertainty of design parameters, and diversity of the decision-making, which will lead to fluctuations in the performance or even failure of the scheme in many cases. To solve this problem, a conceptual design model was constructed by considering multi-uncertainties. First, taking satisfaction, performance, and production cost as design objectives, a conceptual design optimization model was established. Secondly, an improved non-dominated sorting genetic algorithm-II (NSGA-II) based on the expectation-possibility-probability hybrid model was proposed to search the Pareto solutions of conceptual design. Finally, the optimal conceptual design scheme was selected from the Pareto set by using the intuitionistic fuzzy λ-Shapely Choquet integral method. The effectiveness and efficiency of the proposed method were validated by the polishing machine conceptual design.

We design a complete technical chain for developing a Q&A interaction system between Chinese ancient figures and modern users. The system is built on end-cloud collaboration, and all the users need to do is uploading a Chinese ancient painting figure face image. The interactive ability is realized by a large language model. Chinese ancient paintings often emphasize vivid expression and lack realism. Therefore, to enhance the time-travel experience, we restore the Chinese ancient painting figure faces as modern faces with realistic style, which is also the most critical and challenging part in the entire technical chain. We solve this problem by the StyleGAN2 generator in the encoder4editing (e4e) algorithm. Our system is very expected to be deployed in application scenarios such as museums.

Feature engineering is one of the most important and time-consuming steps of machine learning algorithms. In recent years, automatic feature engineering (AFE) methods have received a lot of attention due to their low cost and scalability. However, the existing AFE methods do not take into account the interactions of features in the evolution process and cannot achieve effective feature selection, which weakens the performance of these methods. To tackle the above issues, a novel visible-hidden hybrid automatic feature engineering (VHAFE) method is proposed in this work. Specifically, a visible-hidden hybrid feature transformation graph (VHFTG) is devised to represent both various kinds of feature transformation functions and the feature selection process. Afterward, a multi-pointer state identification method is proposed, which enables the interactions among derived features and the efficient utilization of historically derived features. Furthermore, the multi-agent reinforcement learning algorithm is introduced to optimize the evolution process of VHFTG, where an input variable selection network is devised as the auxiliary policy network to avoid the generation of excessive noise features. Additionally, the VHAFE achieves the state-of-the-art results on a total of 19 public datasets and 1 dataset collected from the actual industrial operation process of gas turbines, which demonstrates the effectiveness of the proposed method.

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Copy DOI

Spindle axial error is the main factor restricting machining accuracy improvements of machine tools. Monitoring the machining process of computer numerical control (CNC) machine tools is challenging due to inability to reserve sensor space and interference from high-pressure coolant spray with sensor readings. This paper proposes a method for predicting machining accuracy of CNC machine tools and adaptive compensation for the absence of sensing during machining. The residual and skip connection enabled adaptive cosine annealing learning rate physics informed neural networks model reconstructs the temperature field of spindle warm-up process, with computational speed improving by 82.10% and stability by 40.68%, respectively, versus PINN models. The temperature node most correlated with axial error is identified, and its temperature process is predicted using a long short-term memory model with hyper-parameter optimization. The end condition of the warm-up occurs when the temperature reaches a specific threshold, determined by the preset machining accuracy requirement. Subsequently, the transition characteristics of temperature-error mapping relationship in the warm-up process are identified and an error prediction model is developed according to the sensing information after the turning point. Timely compensation is then performed before the accumulated prediction errors exceed the limit. Prediction and compensation effectiveness are verified on the factory machine tool, with results demonstrating that prediction accuracy improves with extended warm-up time, and machining precision enhances by 96.8% compared to conventional machining.

To address the challenges of solving the many-objective flexible job-shop scheduling problem, this study proposes a loose non-dominated sorting genetic algorithm III (LNSGA-III), an enhancement of the non-dominated sorting genetic algorithm III (NSGA-III). First, a loose dominance principle is proposed to overcome the shortcomings of low selection pressure and slow convergence under the Pareto dominance principle. Next, a novel crossover operator without repair, named improved order crossover, is presented to fully preserve the characteristics of exchanged operations and enhance the exploration capability of the algorithm. Experimental studies involve testing algorithms on some typical scheduling instances with six simultaneously optimized objectives. The primary metric for algorithm comparison is the hypervolume, with additional investigation for statistical significance. Other metrics, including coverage, convergence and diversity, are also used for comparison. The experimental results demonstrate the effectiveness of the proposed enhancements, showcasing the significant superiority of the algorithm over some state-of-the-art alternatives.

Insufficient data, lack of labelled data, and limited data sharing hinder deep-learning-based fault-diagnosis methods. Most of the existing methods focus on addressing only one of these issues and consequently lack practical applicability. The method proposed in this study addresses these issues simultaneously. First, a temporal-context contrastive learning method is proposed that combines the concepts of few-shot and self-supervised learning. This method assists the feature extractor in learning fault data representations from unlabelled data through a specially designed loss function, thereby enabling the training of deep-learning models on small-scale, label-deficient datasets. Next, a federated learning framework is introduced to train a global model from multiple datasets without data sharing. In addition, a novel client contrastive loss function is proposed to address the issue of model performance degradation caused by different distributions among client datasets. Finally, experimental evaluations are conducted on public datasets, and the results demonstrate the effectiveness and superiority of the proposed method.

Restricted by the scarcity of labeled samples, the transfer domain adaptation has been applied to thermal error prediction of machine tools under complex industrial practices. However, extant studies largely rest on the assumption that the target distribution is given and invariant, which violates the fact that the working conditions may change over time in the real production. To this end, this paper presents a novel subspace metric-based dynamic domain adaptation (SMDDA) scheme for real-time prediction of thermal error. Firstly, a practical thermal feature extractor is constructed to capture both local and global features of temperature sequences. Then, domain adaptation of thermal features is achieved by aligning each source-target domain pair and the outputs of each regressor. In particular, instead of directly aligning the original thermal features, we align their angles and scales in a specific subspace generated by the pseudo-inverse Gram matrix of the two domains to improve the characterization of feature correlations. To fit real-time temperature streams with dynamic conditions, a model updating strategy with buffered weighted incremental time windows is proposed, which achieves dynamic prediction of thermal errors via pseudo-values generated by the target network and its asynchronous update with the online network. Extensive evaluations and comparisons with state-of-the-art methods under exhaustive experiments covering seven different spindle thermal error transfer tasks show that the proposed SMDDA performs quite competitively in terms of both prediction accuracy and stability.

The rationality of product module partition is crucial to the success of modular design. The correlations between components of complex products are complex, increasing the difficulty of module partition. Thus, many existing methods of module partition have difficulty realizing this process effectively for complex products with a large number of components. This paper proposes a module partition method for complex products based on stable overlapping community detection and overlapping component allocation. The correlations between components are analyzed to obtain a comprehensive correlation strength matrix. The undirected weighted network is used to represent components and the correlations between them. A stable overlapping community detection algorithm based on the improved judgement of within-community Shapley values is proposed to generate multiple preliminary schemes of module partition. Overlapping components among modules are allocated to the most suitable modules by adopting a genetic algorithm (GA). The scheme with the largest modularity measure Q is selected as the final scheme of module partition. The proposed method is applied to a computer numerical control (CNC) grinding machine. The proposed module partition method for complex products is demonstrated to be superior to other effective methods.