Surface microdefect detection is a major challenge in advanced manufacturing, while the existing computer vision-based methods struggle to detect micrometer-scale defects, which are invisible to the naked eye. To overcome this problem, this article develops a novel system based on machine-tactile-sensation (MTS) to detect the microdefect. The MTS-based inspection system uses vision-based tactile sensor to convert the tactile signals into visual image, which can capture the fine geometric morphology of the microdefects. Furthermore, considering the limited resolution of the transformed signal, TouchNet, which integrates a label-guided diversity contrastive learning method with an adaptive receptive field selection module, is introduced for defect recognition, which can leverage prior knowledge to guide hyperspherical clustering and employs Gram regularization to prevent feature degradation. The experimental results on the HUST-Tactile dataset indicate that the proposed method can detect the microdefects as small as 0.01 mm with 97.06% accuracy. It also outperforms the state-of-the-art models by 1.33% and 8.46% on the public NEU-CLS and RSW-C datasets, respectively.

Zero-shot anomaly detection (ZSAD) aims to identify anomalies in new classes of images, and it’s vital in industry and other fields. Most current methods are based on the multimodal models CLIP and SAM, which have prior knowledge to assist model training, but they are highly dependent on the input of the prompts and their accuracy. We found that some diffusion model-based anomaly detection methods generate a large amount of semantic information and are very valuable for the ZSAD task. Therefore, we propose a diffusion model based zero-shot anomaly detection method, DZAD, and no additional prompt input is required. First, we propose the first diffusion-based zero-shot anomaly detection framework, which uses the proposed multi-timestep noise features extraction method to achieve anomaly detection in the denoising process of a latent space diffusion model with a semantic-guided (SG) network. Second, based on the detection results, we proposed a two-branch feature extractor for anomaly maps at different scales. Third, based on the difference between the anomaly detection task and other general image detection tasks, we propose a noise feature weight function for the diffusion model in the zero-shot anomaly detection task. Comparing with 7 recently state-of-the-art (SOTA) methods on MVTec AD and VisA datasets and analysis of the role of each component in ablation studies. The experiments demonstrate the validity of the method beyond the existing methods.

This paper studied the problem of wafer probing scheduling in wafer fabrication plants. As the final step in the front-end process of semiconductor manufacturing, wafer probing plays a critical role in wafer production. The wafer probing scheduling problem is modeled as a flexible job shop scheduling problem, considering resource constraints and batch processing. Firstly, a mixed-integer programming model is constructed to minimize the makespan. Then, an improved gray wolf optimizer is proposed to address the wafer probing scheduling problem. This improved gray wolf optimizer incorporates three improvement strategies within the framework of the original gray wolf algorithm: (1) an opposition-based learning approach is utilized to enhance the quality of the initial population; (2) leader agents mutation strategy is developed for local search; (3) a parameter adaptive adjustment is introduced to balance local and global search. In addition, an active decoding framework based on prior knowledge is designed to accelerate the convergence of the improved gray wolf optimizer. Finally, the effectiveness of the proposed improved gray wolf optimizer is verified on 15 instances of varying scales, and the results demonstrate that the proposed improved gray wolf optimizer outperforms other state-of-the-art algorithms.

Reconfigurable manufacturing systems offer enhanced flexibility to adapt to rapidly changing market demands. However, the reconfigurability of equipment introduces significant challenges to production scheduling, complicating optimization. This paper addresses the scheduling problem in reconfigurable manufacturing systems and proposes a discrete grey wolf optimizer algorithm with an active-decoding strategy (DGWO). A novel operation-configuration encoding scheme is proposed to comprehensively represent the solution space, accompanied by an active-decoding strategy that maximizes solution exploration and minimizes idle time. In the GWO, two crossover operators are introduced to enhance the search space, while the random walk strategy is introduced to prevent the algorithm from falling into premature convergence. Additionally, four neighborhood structures are defined based on the encoding space, and an efficient randomized enhanced local search is developed based on these structures to improve the algorithm's exploitation capability. The proposed DGWO algorithm is evaluated on 60 benchmark instances and compared with several related algorithms, demonstrating superior effectiveness and convergence performance.

Due to the complexity of production scheduling and increasing demand, various methods, including solvers, are widely applied to the job shop scheduling problem. Among these, using machine learning techniques to enhance solver quality has attracted significant attention. However, beyond the model, the characteristics of problems greatly influence solver performance. This study focuses on the classic job shop scheduling problem and explores methods to improve constraint programming model efficiency through machine learning. An automatic branching strategy selection method based on machine learning is proposed, consisting of two components: the problem features extraction and strategy selection identification. For features extraction, three feature extraction approaches are designed. In the strategy selection phase, a classification method based on the graph neural network is used to incorporate the set of three types of features, and the problem-related loss function is designed. We conducted experiments on the proposed method on 3500 training sets and 70 test sets (benchmark), and compared the experiments with the automatic selection strategy that comes with OR-Tools. The results show that the proposed method can obtain equal or better solutions on 81.42% of the instances.

Rolling bearings are the key components of various equipment and easily subject to failure, remaining useful life (RUL) prediction technology can grasp their health statuses to make a reasonable maintenance plan. Transfer learning methods for rolling bearing cross-domain RUL prediction focus on the direct alignment of the degradation process between the target domain and source domain. However, the working condition and degradation process of rolling bearing to be predicted are unknown, which are different from the target domain or source domain. The caused data distribution discrepancy has seriously affected the RUL prediction acceptance. To solve this problem, a multilevel feature alignment (MLFA) method for advancing RUL prediction without target domain data for training is proposed. Specifically, a proposed weighted TOPSIS is implemented for supervised label making to train a robust RUL prediction prior model. The three-level feature alignment (TLFA) strategy, which involves feature alignment by the proposed adaptive nonlinear state estimation, envelope spectrum (ES), ES-combined deep convolutional autoencoder, is developed to mitigate the data distribution discrepancy between the predicted and prior entities. TLFA transforms prediction tasks from cross-domain to different failure behaviors. Extensive experiments conducted on both public and industrial scene run-to-failure bearing datasets validated the superiority of the MLFA. These results from comparison experimental show that MLFA improves mean absolute error and root mean absolute error (RMSE) about 0.061 and 0.054 individually than some state-of-the-art methods.

Most industrial equipment works under variable conditions, variable conditions would increase the within-class difference and reduce the cross-class difference of fault samples. The existing methods mainly consider steady speed scenarios and the global domain adaptation while ignoring the within-class and cross-class distribution alignment, the fault distribution variation leads to the deterioration of diagnosis performance. In this article, a self-supervised pseudo-label learning-enabled (SPL) cross-domain diagnosis method is proposed for fault diagnosis under time-varying speeds. Specifically, the Cauthy maximum mean-square discrepancy is designed for global distribution-level feature alignment by reducing the domain discrepancy. The pseudo-label training and consistency regularization are established for decision boundary adjustment by optimizing the probability distribution difference between the target domain and its perturbed output. Besides, uncertainty-reweighted class confusion minimization is introduced in within-class and cross-class distribution alignment to decrease negative transfer caused by huge within-class discrepancies and small cross-class differences, which can effectively improve the diagnosis accuracy of the hard-to-identify confusion samples. Experiments on time-varying fault diagnosis tasks show the superior performance of the proposed method. The proposed SPL framework improves average diagnosis accuracies by at least 8%.

The flexible job scheduling problem with batch processing machines (FJSP-BPM) is an extension of the flexible job shop scheduling problem and batch scheduling problem in some engineering scenarios. It allows an operation to be processed by any usable machines, and multiple jobs can be processed on a batch in batch processing machines simultaneously. In this study, a mixed integer linear programming model is proposed to minimize the makespan. This study first designs a strategy of chromosome slicing based on the marginal cost to generate batches and a hybrid genetic algorithm (HGA) is proposed to solve the FJSP-BPM problem. Neighborhood structures are designed to search for better solutions. Finally, the experimental results demonstrate that the proposed HGA has obtained the best solutions for all instances and has effectively solved the FJSP-BPM problem.

Crossing different working conditions is a common scenario in rotating machinery fault diagnosis, which can be solved by cross-domain transfer learning. However, the existing diagnosis methods do not consider possibly new and unknown faults, i.e., open-set fault diagnosis scenarios, which would cause diagnosis performance degradation. To address this issue, in this article, the self-supervised-enabled open-set cross-domain (SEOC) approach is proposed for fault diagnosis of rotary machines under various working conditions. Specifically, open-set risk minimization and self-supervised contrastive learning are proposed to improve distinguishability and stability. A pseudolabel consistency self-training is designed to decrease the domain shift. A novel open-set identification strategy with the designed squeeze confidence rule is developed for unknown- and known-class fault detection. Experiments on three-phase motor and bearing datasets illustrate the superior and efficient performance of the proposed SEOC method. The proposed SEOC framework improves the overall classification accuracies by at least 9%, and the average accuracy of unknown fault identification is more than 97.68% in motor and bearing fault diagnosis.

Deep learning is an important method to deal with point cloud, but its ability is limited to extract local features of point cloud. Many deep learning networks are designed to capture the local information, but they ignore the importance of non-local features to the point cloud. This paper proposes an improved non-local neural networks for point cloud classification. The non-local module can extract local and non-local features of the point cloud simultaneously. The local information is obtained based on the feature distance between neighborhood points searched by k-nearest neighbor method. The extracted local features are integrated into the non-local network, which can capture non-local features from the entire point cloud. The designed non-local module can be easily inserted into the existing point cloud processing network. The proposed method is evaluated on well-known ModelNet40 shape classification benchmark. Experimental results show that the proposed method achieves a significant improvement in classification accuracy.

Surface defect is a serious problem in real-world manufacturing system and it is important to use vision-based recognition to ensure the surface quality of products. Currently, due to the ability of automatic feature extraction, deep learning models, such as convolutional neural network (CNN), have been widely used in this area. However, these CNN-based models may not solve a problem well - inter-class similarities and intra-class variations (ISIV), which affect their ability of feature extraction and thus influence their recognition performance. To address this problem, this paper introduces a graph guidance mechanism into CNN to improve the ability of feature extraction, called Graph guided Convolutional Neural Network (GCNN). Firstly, GCNN defines a graph by computing the similarities between training samples. Secondly, the graph is introduced into VGG11, a popular CNN structure, to increase the inter-class distances and decrease the intra-class distances between defect samples. Meanwhile, a learnable coefficient is introduced into the training process to balance the effect of graph guidance automatically. The experimental results on four famous surface defect datasets demonstrate that the graph guidance helps CNN models have better ability of feature extraction and thus achieve better performance. Compared with state-of-the-art models, the proposed method can achieve the best performance. Furthermore, the final discussion shows that the learnable coefficient can help the proposed model to gain better performance, and that the proposed model increases a little computation cost compared to its original CNN model.Note to Practitioners—This paper is motivated by the problem in real-world manufacturing process – inter-class similarities and intra-class variations. Most of current CNN-based models may not solve this problem well, which limits their applications in real-world system. This paper proposes a graph guidance mechanism and introduces it to the training of CNN. The proposal can improve the ability of feature extraction, and thus have better performance than those without the graph guidance. By applying the proposal to four examples, the results show that the proposal is more feasible and effective than the state-of-the-art models in surface defect recognition. Furthermore, the mechanism assists the training of CNN and is removed in the process of recognition, so it will not increase time and occupied memory in recognition, and meanwhile, only limited computation cost is increased in training process.

Complex equipment is important in industry and fault diagnosis is a key technology for maintenance. Convolutional neural network (CNN) is a common manner for fault diagnosis. But due to the imbalanced data, it cannot be applied to complex equipment directly. Generally, CNNs require balanced data, and imbalanced data will mislead the models and ignore actual faults. In complex equipment, since fault occurs rarely, the fault data are imbalanced. This impedes the application of CNNs greatly. To overcome this problem, a hierarchical training-CNN is proposed in this article. The proposed method uses an effective number-resampling to balance fault data, which avoids invalid samples, and introduces a magnet-loss pretraining to address the overlap of features between different faults. Based on these improvements, the proposed method achieves good diagnosis performances with an average accuracy of 96.56%, and has been developed into a real-world case successfully with an accuracy of 94.28%.

In vision-based defect recognition, deep learning (DL) is a research hotspot. However, DL is sensitive to image quality, and it is hard to collect enough high-quality defect images. The low-quality images usually lose some useful information and may mislead the DL methods into poor results. To overcome this problem, this article proposes a generative adversarial network (GAN)-based DL method for low-quality defect image recognition. A GAN is used to reconstruct the low-quality defect images, and a VGG16 network is built to recognize the reconstructed images. The experimental results under low-quality defect images show that the proposed method achieves very good performances, which has accuracies of 95.53-99.62% with different masks and noises, and they are improved greatly compared with the other methods. Furthermore, the results on PSNR, SSIM, cosine, and mutual information indicate that the quality of the reconstructed image is improved greatly, which is very helpful for defect analysis.