Journal of Intelligent Manufacturing最新文献

To thoroughly investigate the impact of beam shaping on melt pool behavior and accurately predict the microstructure and mechanical properties of the final product in laser powder bed fusion (LPBF) for metal additive manufacturing (AM), it is crucial to efficiently model the temperature profiles of melt pools subjected to different laser beam shapes. Numerical methods necessitate significant computational resources and time. Machine learning (ML) based surrogate models, on the other hand, are incapable of precisely predicting three-dimensional temperature profiles and lack generalizability in modeling distinct beam shapes beyond the Gaussian beam. To address these limitations, this paper introduces the Melt Pool Temperature Profile Network (MPTP-Net), a novel model developed to efficiently predict the three-dimensional temperature profile of the melt pool based on laser beam parameters, including power, scan velocity, standard deviation of power distribution, and ring radius (applicable to ring beams). By incorporating an auxiliary geometry branch alongside the temperature profile head, our constructed multi-task learning framework is capable of learning the underlying connection between the laser beam parameters and melt pool morphology in the latent space. Hence, the proposed model improves accuracy and generalizability in predicting the 8-layer temperature profile across a wide range of melt pool dimensions. Additionally, the progressively upsampling module of MPTP-Net contributes in predicting the high-fidelity temperature profile with accurate boundaries and smooth temperature gradients of the melt pool. Through extensive validation using datasets derived from both Gaussian and ring beams, our model consistently demonstrates a superior degree of concordance between the predicted and actual melt pool temperature profiles than the state-of-the-art methods.

Textile anomaly detection with high accuracy and fast frame rates are desired in real industrial scenarios. To this end, we propose an efficient memory guided distillation network, which includes encoder, decoder, and segmentation networks. Instead of utilizing a pre-trained large network as the encoder, we utilize a small feature extraction network, whose features are distilled from a teacher network. To improve the reconstruction quality with small networks, we further introduce an efficient memory bank, whose features are extracted by the teacher network with normal reference inputs. Considering the blurry reconstruction may lead to false-positive results, we further introduce a pseudo-normal simulation method by augmenting the inputs with blurry effects. Besides, we construct a Textile Anomaly dataset (Textile AD) for textile anomaly detection with pixel-wise labels for comprehensively evaluation and our method demonstrates superior performance on the Textile AD dataset. Additionally, we performed experiments using the publicly accessible MVTec-AD industrial anomaly dataset and our approach aligns closely with the performance of cutting-edge methodologies, which demonstrates that our method is applicable to other industrial product categories. Our Textile AD is shared in https://github.com/Songziyangtju/Textile-AD-dataset.

In industrial scenarios, weakly supervised pixel-level defect detection methods leverage image-level labels for training, significantly reducing the effort required for manual annotation. However, existing methods suffer from confusion or incompleteness in predicting defect regions since defects usually show weak appearances that are similar to the background. To address this issue, we propose a foreground–background separation transformer (FBSFormer) for weakly supervised pixel-level defect detection. FBSFormer introduces a foreground–background separation (FBS) module, which utilizes the attention map to separate the foreground defect feature and background feature and pushes their distance intrinsically by learning with opposite labels. In addition, we present an attention-map refinement (AMR) module, which aims to generate a more accurate attention map to better guide the separation of defect and background features. During the inference stage, the refined attention map is combined with the class activation map (CAM) corresponding to the defect feature of FBS to generate the final result. Extensive experiments are conducted on three industrial surface defect datasets including DAGM 2007, KolektorSDD2, and Magnetic Tile. The results demonstrate that the proposed approach achieves outstanding performance compared to the state-of-the-art methods.

The rapid replacement of large-scale end-of-life (EOL) heavy machineries like automobiles, aircrafts and industrial robots necessitates efficient resource recovery to promote sustainable and eco-friendly manufacturing. This study therefore focuses on multi-manned disassembly lines in recycling large-scale products, bridging the gap between theory and practice. We introduce complex, safety-sensitive tasks that require collaborative efforts of multiple workers in the Multi-Manned Disassembly Line Balancing Problem (MMDLBP) for the first time. We also consider worker heterogeneity due to varying training and skills, as manual stations are inherently worker-dependent in nature. To address this Multi-Manned Disassembly Line Balancing Problem with Worker Heterogeneity and Collaboration (MMDLBP-HC), we establish a mixed-integer programming model to minimize cycle time and labor cost simultaneously. Given its NP-hard nature, we develop a Multi-Mechanism-Enhanced Bi-Objective African Vultures Optimization Algorithm (MBAVOA). It employs specified encoding with numerical branching, precedence-priority concurrent decoding, and selective opposition-based learning. We also combine trigonometric-based mechanisms with the African vulture optimization algorithm (AVOA) to enhance exploration. Additionally, adaptive neighborhood search mechanisms are tailored for inter-individual information exchange. Numerical experiments compare MBAVOA to four meta-heuristics and an exact algorithm. The results demonstrate the model accuracy and the effectiveness of the encoding and decoding mechanisms, while MBAVOA outperforms benchmark algorithms significantly. Finally, we offer managerial applications to guide practitioners in balancing plan formation and training program design.

In recent years, supervised learning has been the predominant method for wafer map defect pattern classification (WM-DPC), requiring a substantial amount of labeled data to build effective models. Nonetheless, gathering industrial data is challenging and demands significant manual labeling efforts, making it both expensive and time-consuming. To overcome these obstacles, we introduced a contrastive learning framework for WM-DPC based on automatic data augmentation. This innovative augmentation approach takes account of the regional defect density characteristic of various defect types, addressing the limitations of traditional fixed data augmentation and improving the model’s generalization capacity. The framework operates in two phases. At first, a lightweight encoder extracts rich representative features from unlabeled data. Then, the classification network is fine-tuned with a limited labeled data set. Experimental outcomes using the public WM-811K dataset showed that the proposed automatic data augmentation and lightweight encoder effectively captured detailed representative features from unlabeled data, and achieved an average accuracy close to 91% after fine-tuning with minimal labeled data.

A high-precision measurement machine tool faces the challenge of correlating the overall motion accuracy with the components form and positional accuracy. This study presents an innovative method for addressing this issue in ultra-precision measuring machines. A geometric error model based on multibody theory, and a weight model are established to predict measurement results and correlate overall motion accuracy with individual component accuracy. To validate the model, a target overall motion accuracy of 100 nm is set and the all the individual components accuracy is calculated by the geometric error weights derived from the proposed model. By fabricating a critical component, the linear guideway, to meet specific individual accuracies and incorporating it in an ultra-precise measurement machine, the study demonstrates achieving the individual accuracies with the magnetorheological polishing. Finally, the overall motion accuracy is validated by a cross test among the designed machine, DUI profilometer, and Zygo interferometer. By measuring a same optical surface, the measurement results show the surface PV differences better than 100 nm. The results demonstrate the validation of the correlation between overall motion accuracy and component accuracy established by the method described in this paper. In conclusion, this study offers an accurate design solution for determining overall motion and individual accuracies, enabling high accuracy in intelligent manufacturing equipment.

As a non-conventional machining technique, the micro electrical discharge machining (micro-EDM) process primarily involves the removal of material from the workpiece through high-frequency discharges. The machined surface is covered with multiple overlapping craters to form geometric features with specific surface quality and dimensional accuracy. Consequently, there is a significant need to explore the crater morphology induced by the discharge pulses, which contributes to the precise control of component size and shape. This study targets the identification of material removal in relation to pulse-crater matching within micro-EDM. Initially, pertinent parameters of both pulses and craters are characterized and correlated through a single pulse discharge experiment. Subsequently, accompanied by a pulse classification, a continuous pulse discharge experiment is designed to establish a one-to-one correspondence between erosion craters and the discharge pulses associated with normal discharge, effective discharge, and arc phenomena, which all contribute to material removal. The impact of different discharge pulse types on workpiece material removal is further investigated, with explanations based on energy density and the fraction of energy entering the workpiece. Employing machine learning methods, predictive models for crater-related parameters are developed based on the monitored electrical signals. A comparison of the prediction results from different regression models with various inputs confirms the profound nonlinearity and stochastic nature of the EDM process. Ultimately, the artificial neural network model shows to be optimal in predictive performance, yielding relative errors of 7.81%, 12.49%, and 18.82% for crater diameter, depth, and volume, respectively. Notably, the prediction error for cumulative material removal is only 1.64%, affirming the soundness of the proposed material removal identification for different discharge pulses. Other material removal volume calculation approaches often hinge on machining parameters or statistical distributions. Contrarily, the distinctive characteristic of this approach lies in its implementation of precise pulse-crater correlations of various discharge types based on in-process data. This method is further applied to the prediction of the total material removal volume in micro-EDM drilling. The results are promising for enhancing geometric dimension control in EDM, particularly regarding machining depth.

Tungsten Inert Gas (TIG) welding is a manufacturing process that utilizes argon as a shielding gas and tungsten as an electrode to join metals at high temperatures. The weld penetration is the key to determine the quality of the weld. However, the lack of sensing technology makes weld penetration difficult to predict, which imposes a major challenge to process stability and weld quality. To address this challenge, a digital twin-driven method is proposed for characterizing the melt pool morphology and melt penetration prediction. To achieve this, an analytical model of the melt pool with time-varying welding speed under the action of a double ellipsoidal circular heat source is first derived. The analytical model is solved using the numerical integration method. The prediction of melt depth and melt width is achieved by extracting isotherms. Meanwhile, a digital reconstruction of the welding scene was achieved by implementing the Neural Radiance Fields (NeRF) method. The target rendering of the melt pool and welding scene is accomplished by constructing voxels and meshes. Furthermore, VR is utilized as the interface for human–computer interaction, and a digital twin model of the molten pool morphology and welding scene is generated. The prediction model's accuracy is verified through welding experiments using 304L steel on a robotic welding system. The results show that in the 0–4 s stage, the penetration error is controlled within 7%. In the stage of 4–16 s when the speed changes, the maximum error of penetration is 16.59%. In terms of welding scene reconstruction quality, PSNR is 33.98 and SSIM reaches 0.9032. The method allows real-life simulation of different welding conditions and parameter combinations prior to welding, assessing their impact on the welding results, in order to find the optimal configuration of process parameters. It can also be remotely realized to monitor and control the melt penetration in real-time during the welding process. This method provides a new solution and a theoretical guidance system to solve the welding penetration control problems and it plays an important role in promoting welding intelligence.

Cavities in a weld seriously affect the airtightness of the chip, which makes chip inspection a crucial step in intelligent manufacturing. In recent years, deep learning-based defect inspection models have shown significant advantages in reducing human errors. However, due to the scarcity of defective data, deep learning-based models are susceptible to overfitting. Moreover, the multiscale and uneven grayscale distribution of cavities further compound the challenges faced by these models. To address these issues, we develop a chip inspection system based on a multiscale subarea attention network (MSANet) for cavity defect detection. In the system, the segment anything model is embedded to interactively segment the weld. Furthermore, to circumvent the overfitting problem, a large-scale cavity dataset is built by splitting the segmented weld into multiple patches. Notably, a novel MSANet is proposed to precisely segment the varying cavities, and a source-to-destination Dijkstra algorithm is designed to assess the chip quality. The experimental results demonstrate that our chip inspection system achieves a 99.24% F1-score and 99.26% AUC.

Two-photon polymerization (TPP) has emerged as an advanced additive manufacturing technique, allowing for the creation of three-dimensional micro-nano structures with high precision based on two-photon absorption principle. Precisely control light dosage determined by the printing parameters, is crucial for inducing photopolymerization across different photocurable materials and various structures. To address the challenges of parameter optimization, deep learning models were employed to quickly obtained the ideal printing parameters through automated visual inspection during TPP printing process and after post-processing. A dataset was collected from the video recordings during printing process and the images obtained from after post-processing of samples. Data augmentation techniques were applied to enhance the dataset. For the TPP printing process, the mean prediction accuracy increasing from 95.1% to 96.8% for the 3D-CNN model and from 95.4% to 97.8% for the CNN-LSTM model. For the post-processing, the mean prediction accuracy with CNN model increases from 94.5% to 95.2%. Consequently, spatial–temporal DL models were trained based on these datasets, and the results of dual visual inspection method demonstrated a high accuracy of 93.1% and a rapid recognition time of 48 ms. And an analysis of the failure cases of the deep learning models was conducted. Additionally, the optimal printing parameter ranges was determination for various combinations of materials and structures. This system plays a crucial role in accelerating the optimization of TPP process parameters and quality inspection, effectively addressing the challenges in the industrialization process of TPP technology.