Journal of Intelligent Manufacturing最新文献

Manufacturers face two opposing challenges: the escalating demand for customized products and the pressure to reduce delivery lead times. To address these expectations, manufacturers must refine their processes, to achieve highly efficient and autonomous operations. Current manufacturing equipment deployed in several facilities, while reliable and produces quality products, often lacks the ability to utilize advancements from newer technologies. Since replacing legacy equipment may be financially infeasible for many manufacturers, implementing digital transformation practices and technologies can overcome the stated deficiencies and offer cost-affordable initiatives to improve operations, increase productivity, and reduce costs. This paper explores the implementation of computer vision, as a cutting-edge, cost-effective, open-source digital transformation technology in manufacturing facilities. As a rapidly advancing technology, computer vision has the potential to transform manufacturing operations in general, and quality control in particular. The study integrates a digital twin application at the endpoint of an assembly line, effectively performing the role of a quality officer by utilizing state-of-the-art computer vision algorithms to validate end-product assembly orientation. The proposed digital twin, featuring a novel object recognition approach, efficiently classifies objects, identifies and segments errors in assembly, and schedules the paths through the data pipeline to the corresponding robot for autonomous correction. This minimizes the need for human interaction and reduces disruptions to manufacturing operations.

This research employs transfer learning to explore and compare pre-trained deep learning models for defect detection in injection molding processes. It introduces advanced neural network architectures, specifically Inception and ResNet50, which have not been extensively studied in this context. Through systematic evaluation using techniques such as data augmentation, architecture modification, and hyperparameter tuning, the study aims to enhance detection precision. The methodology addresses deployment challenges inherent in defect detection systems and emphasizes the importance of model selection for achieving desired goals. Comparative assessments with contemporary models highlight the effectiveness of the proposed approach in real-world production settings. Improved results obtained with the Inception model demonstrate a precision of 92.3%, recall of 100%, and F1 score of 96%, surpassing ResNet50 as well as previous studies utilizing VGG16 and Yolo v5. This underscores the reliability of the Inception model for defects detection in practical scenarios. Furthermore, beyond accuracy enhancement, the study aligns with the broader goal of advancing sustainable manufacturing by integrating smarter defect detection mechanisms. The findings not only offer a robust framework for selecting optimal detection models but also lay the groundwork for future research endeavors aimed at improving adaptability and efficiency in defect detection systems across various industrial applications. This contributes to the evolution of intelligent manufacturing processes, balancing quality and profitability objectives.

Absolute positioning accuracy is a crucial index for evaluating industrial robot performance and the foundation for motion trajectory and machining accuracy. Current positioning error compensation methods focus on achieving unified compensation within a robot’s workspace. These methods rely heavily on expert knowledge and require a significant amount of manual intervention. To realize refined error compensation and improve the autonomy and intelligence degree of a robot, an intelligent hierarchical positioning error compensation method based on a master–slave controller is proposed in this paper. Specifically, positioning error compensation is addressed through two research questions related to positioning error level diagnosis and compensated pose prediction, and the approach consists of two major processes: automatic creation of a compound branch compensation network and hierarchical positioning error compensation. For the first process, the master controller independently grades the positioning error levels and directs the diagnosis slave controller to create a positioning error level diagnosis model in terms of the robot pose error data. Then, it directs the prediction slave controller to create several compensated pose prediction models based on the pose data of different error levels. Subsequently, the diagnosis and prediction models are integrated to form a compound branch compensation network. For the second process, the master controller first activates the diagnosis branch of the compound branch compensation network to determine the positioning error level of the current robot pose. Then, it activates the prediction branch corresponding to the determined error level to generate the compensated pose. Finally, it uses the diagnosed error level to filter the compensated pose. Experimental cases of a Stäubli robot and a UR robot are applied to verify the feasibility and effectiveness of the proposed method. The experimental results show that the proposed method reduces the positioning error of the Stäubli robot from 0.848 to 0.135 mm and the UR robot from 2.11 to 0.158 mm, outperforming relevant current methods.

Online fabric defect detection plays a critical role in the quality management of textile production. However, the high-impact and low-probability characteristics of defective samples lead to redundant design of network and hinder its real-time performance. To improve the time efficiency, this paper proposes a dynamic inference network (DI-Net) which can dynamically allocate computation resources as the complexity of image. Firstly, “AND” Gates are incorporated into the backbone to control activation of network’s function modules, allowing for dynamic adjustment of network depth. Additionally, the dynamic inference module which contains several exits with inference unit is proposed to collaborate with “AND” Gates. When sample’s confidence at specific exit satisfies the early-exit policy, the inference unit will allow it to early-exit from network and output a negative value to corresponding “AND” Gate. As a result, the output of “AND” Gate will also be negative and subsequent network will not be activated. Finally, the two-stage training strategy and exit-weighted loss function are proposed to avoid crosstalk and facilitate different exits to focus on adequate samples, enabling the efficient training of DI-Net. The experiments on the fabric dataset demonstrate that the proposed DI-Net can achieve detection precision and recall over 99% for normal samples, and approximately 95% for defective samples. Besides, its detection speed has been improved by 20%, reaching 30.1 frames per second and 20.96 m/min. This indicates that the proposed DI-Net can meet the requirements of online fabric defect detection.

Yield is key to profitability in semiconductor manufacturing and controlling the fabrication process is therefore a key duty for engineers in silicon foundries. Analyzing the distribution of the defective dies on a wafer is a necessary step to identify process shifts, and a major step in this analysis takes the form of a classification of these distributions on wafer bitmaps called wafer maps. Current approaches use large to huge state-of-the-art neural networks to perform this classification. We claim that given the task at hand, the use of much smaller, purpose defined neural networks is possible without much accuracy loss, while requiring two orders of magnitude less power than the current solutions. Our work uses actual foundry data from STMicroelectronics 28 nm fabrication facilities that it aims at classifying in 58 categories. We performed experiments using different low power boards for which we report accuracy, power consumption and power efficiency. As a result, we show that to classify 224(times )224 wafer maps at foundry-throughput with an accuracy above 97% using a bit more than 1 W, is feasible.

In recent years, data-driven machine learning models have achieved good results in fault diagnosis of rotating machinery under different working conditions. However, in practical applications, the lack of fault samples under various working conditions makes the training of models difficult. In this paper, a multi scale meta-learning network (MS-MLN) that can be applied to few-shot cross-domain diagnosis of rotating machinery is proposed to address this issue. MS-MLN consists of a multi scale feature encoder, a metric embedding process and a classifier. The model is trained by an episodic metric meta-learning strategy under few-shot and domain shift scenarios. Extensive experiments are carried out to verify the effectiveness of MS-MLN, results show that MS-MLN outperforms most benchmark models in bearing and wind turbine gearbox fault diagnosis. Visualization is applied to the model to study its effectiveness. Ablation study is also conducted to discuss the impact of different parts of the model’s feature encoder on its performance in detail.

Aiming at the problems of complex and time-consuming process of manual adjustment of eccentricity and tilt in the evaluation of machining error of aero-engine saddle rotor, and inaccurate measurement of unbalance after multi-stage rotor assembly, this paper proposes an unbalance prediction method based on Genetic Algorithm Back Propagation (GA-BP) neural network and deep belief networks (DBN). Firstly, according to the definition of single-stage rotor machining error, the influence source of saddle rotor machining error and the evaluation of machining error are analyzed. Secondly, GA-BP neural network is established to obtain the concentricity and flatness of saddle rotors at all stages as the error source of unbalance. Then, the output of the GA-BP neural network is used as the input of the DBN to establish the unbalance prediction network model. Finally, the experimental verification is carried out based on the experimental measurement data of an engine rotor unbalance. The results show that the mean value and root mean square error (RMSE) of the unbalance are 16.72 g·mm and 32.71 g·mm respectively, and R-squared (R²) determination coefficient is 0.96 when the 80 groups of samples are tested by the prediction method of DBN. Compared with the method based on the traditional error transfer model, the proposed method based on DBN and GA-BP reduces the average error and mean square error by 86.08% and 75.97% respectively, which greatly reduces the measurement error of rotor unbalance. Therefore, this method can provide technical guidance for the optimal assembly of multi-stage rotors, thereby improving the assembly quality of multi-stage rotors.

Recently, there is a significant growth in the use of the Cyber-Physical System (CPS). New technologies such as Internet of Things (IoT), Industry 4.0, and Analytics have become enablers of CPS implementation. Study of the development and application of CPS in the supply chain context is valuable to operations management and information systems research and practice; especially, a focus on IoT-enabled CPS in production/manufacturing is highly relevant. Knowledge about the research trends of the development and use of CPS for supply chain management supported by new innovations in IT is very limited in the extant literature. The aim of this research is to investigate the research trends of applying CPS in manufacturing. The study encompasses a scientometric analysis of research on deploying the CPS in production systems. Based on a systematic selection process, we collect a total of 245 articles from the Web of Science (WoS) database as the sample for analysis. Using appropriate software, we conduct bibliometric analyses of the sample articles that include citation, cocitation analysis, centrality co-occurrence analysis, and co-authorship analysis. From the bibliometric analysis, we discover major themes of CPS in manufacturing and their evolutions in the extant literature.

The surface error induced by low-rigid deformation and intermittent cutting is common in the milling process of thin-walled workpieces. Machining errors have a direct impact on the surface accuracy of the machined workpiece, making it crucial to monitor the milling error throughout the thin-walled workpiece machining process. This article provides a strategy for forecasting machining errors in thin-walled workpieces. The prediction strategy faces two difficulties: the flexibility variations in the different machining positions of the thin-walled workpieces and the processing information shifting with the varied machining conditions. To tackle these challenges, the knowledge-embedded parameter construction of the strategy establishes a correlation between error and process information by integrating physical constraints and data information. Transfer learning combines a small amount of real-time data with a large amount of historical data, enabling effective practical data application and reutilization. The experimental evaluations and comparisons have demonstrated the predictive performance and applicability of the machining error prediction strategy.

Precision steel balls are among the most crucial components in the industry, widely used in various equipment related to bearings, such as CNC, automotive, medical, and machinery industries. Due to the reflective surface of steel balls, flaw inspection becomes a challenging task. This paper introduces an automatic optical inspection system that employs a modified DRAEM, a reconstruction-based anomaly detection network, for examining the surface of precision steel balls. We made three modifications to the DRAEM network (Zavrtanik, V., Kristan, M., & Skoca, D. (2021). DRAEM—a discriminatively trained reconstruction embedding for surface anomaly detection. http://arXiv.org/arXiv:2108.07610[cs.CV]), including adjusting the generation process of synthesized anomalies, adding a few skip connections from the encoder to the decoder, and incorporating an attention module to enhance the quality of reconstructed images and reduce misjudgments. Experimental results demonstrate a reduction in the model's underkill rate from 8.8% to 4.6% and the overkill rate from 1.5% to 0.4%. This indicates that the proposed methods addressed the issues of reconstruction distortion and the inability to detect small and inconspicuous defects. The proposed system has been successfully implemented in a case study company, showcasing significant advantages, particularly in scenarios involving new production lines or a lack of sufficient defective samples for collection.