Journal of Manufacturing Systems最新文献

Machining feature recognition serves as a foundational step in process planning, crucial for translating design information into manufacturing information. Traditional rule-based methods require extensive manual rule definition, prompting researchers to develop learning-based methods using data-driven algorithms. However, existing learning-based methods typically demand substantial data annotation and show limitations in machining feature segmentation. To address these issues, this paper introduces a novel learning-based machining feature recognition method. The proposed method leverages self-supervised learning to autonomously extract valuable intrinsic information from unlabeled data and incorporates a discriminative loss function to improve feature segmentation performance, thereby enhancing feature recognition results under conditions of limited labeled data. Specifically, the self-supervised learning network is first pre-trained on a large amount of unlabeled point cloud data representing CAD models and then fine-tuned with labeled data using the discriminative loss function. The fine-tuned network can be employed for recognizing machining features. Experimental results demonstrate that the proposed approach is effective during pre-training and improves feature recognition performance with limited amounts of labeled data, potentially reducing annotation efforts and associated costs.

Metal Additive Manufacturing (MAM) has seen significant growth in recent years, with sub-processes like Metal Material Extrusion (MEX) reaching industrial readiness. MEX, known for its cost-effectiveness and ease of integration, targets a distinct market segment compared to established high-end MAM processes. However, despite technological improvements, its overall integration into the industry as a viable manufacturing technology remains incomplete. This paper investigates the competitiveness of MEX, specifically its integration into the supply chain and the implications on cost and carbon emissions. Utilizing real-world data, the research develops a multi-objective optimization (MOO) model for a four-echelon supply chain including suppliers, airports, production facilities, and customers. The optimization model is combined with a previously developed cost model for MEX to optimize facility location in Norway using the NSGA-II algorithm. Employing a case study approach, the paper examines the production of an industrial part using stainless steel 17-4PH, detailing concrete process costs and system-level costs across four different production scenarios: 10, 100, 1,000, and 10,000 parts. The findings indicate MEX’s potential for cost-effective production at low and diversified volumes, supporting the trend towards customization and manufacturing flexibility. However, the study also identifies significant challenges in maintaining competitiveness at higher production volumes. These challenges underline the necessity for further advancements in MEX technology and process optimization to enhance its applicability and efficiency in larger-scale production settings.

Tool Wear Monitoring (TWM) plays a vital role in safeguarding product quality and enhancing machining efficiency. TWM technology mainly includes physics-based models and data-driven methods. However, physical models established under simplified or idealized conditions struggle to capture the complexity of machining processes. Moreover, the predictive efficacy of data-driven methods is heavily contingent upon the quantity of labeled data available. Addressing these issues, a hybrid-driven physics-informed Gaussian process regression model (PIGPR) is proposed. First, a health indicator construction strategy based on feature fitness analysis and Gaussian weighted moving average filtering is proposed to eliminate interference and redundancy in the measurement signal and improve monitoring efficiency. Second, a novel explicit physical model of tool wear was developed, with a determination coefficient of at least 0.98. On this basis, health indicator and proposed priori physical models are employed to constrain the mean function of the Gaussian process regression (GPR), combining data mining and physical models to provide prediction guidance for key physical domain knowledge for the hybrid model. Third, grid search algorithm is used to optimize the model parameters, adaptively identify tool wear conditions, and 95 % prediction confidence interval is given to provide more reliability. Finally, nine sets of experiments with varying cutting settings confirmed the PIGPR model's dependability. The findings demonstrate that the suggested hybrid approach significantly enhances the prediction precision of tool wear, achieving an accuracy of 0.997. Compared to the solely data-driven GPR model, the width and variance of the 95 % confidence interval decreased by 46.44 % and 60.80 %, respectively, which demonstrates that incorporating prior physical knowledge significantly enhances the smoothness and reliability of predictions.

Accurate anomaly detection (AD) of machine tools is crucial to ensure the quality and efficiency of the manufacturing processes. Due to the lack of tool anomaly information, it is difficult for AD model to precisely capture the distribution of health states and then obtain a discriminative decision boundary. Current methods try to reconstruct the normal data distribution without restricting the abnormal, resulting in the unacceptable overlap between normal and abnormal regions and finally leading to high false alarm rate. To tackle these issues, a hierarchical augmented autoencoder is proposed for AD of machine tools during manufacturing. First, a skip-connected autoencoder is built to basically learn the normal representations of multi-sensor data in an unsupervised manner. Then, to improve further emphasis the reconstruction on normality and suppress that on anomalies, we propose hierarchical memory modules to store multi-scale normal prototypical patterns, using them as a prior to guide the reconstruction with preference. Finally, A compound metric loss function is designed to measure data similarity considering both distance and angle perspectives, which can restrain noise interference and enhance model robustness. Extensive experiments are conducted on real-world CNC machine tool datasets, the proposed method achieves better performance for unsupervised AD compared with other typical methods.

Metadata are essential to the manufacturing sector. Encompassing a broad spectrum of information related to manufacturing processes, voluminous metadata are extensively obtained from sensing networks and experimentally validated finite element (FE) models. In the era of digital manufacturing, where metadata holds immense potential, its full value remains largely untapped without thorough analysis and characterisation. Yet, an overwhelming majority of the manufacturing metadata obtained during production lacks crucial information and is categorised as ‘fragmental data’, containing only a few (e.g., 1–2) essential pieces of information. Extremely sparse information within the fragmental data hinders the further analysis and characterisation of underlying scientific patterns. To address this challenge, two physics-informed filters, the probability density function filter (PDFF) and feature-driven neighbour filter (FDNF), were developed and embedded within the Evolutionary Binary (EB) algorithm. These filters enabled the alignment by identifying the origins of a set of naturally unattributed fragmental data, taking the digital characteristics space (DCS) of manufacturing processes as an alignment reference. This was realised by comparing the thermo-mechanical digital characteristics (DC), such as the temperature DC, to the counterparts stored in the DCS. An overall accuracy of 90 % was achieved when identifying the origins of unattributed fragmental hot stamping data using PDFF with a minimum length of 10 and FDNF with minimum length of 25. Results demonstrate a novel methodology to unlock the inherent values from unattributed fragmental data that contains extremely sparse information, thereby revolutionising insights into advanced manufacturing sciences.

Prognostic maintenance (PM) aims to monitor the running status and promptly detect potential failures to improve the availability and productivity of the equipment. The dimensional accuracy and surface integrity of the machined parts are directly influenced by the cutting tools. Thus, tool health monitoring (THM) is crucial to ensure the optimal in-service performance of the parts. Nevertheless, the variability of operating conditions, including milling parameters, workpiece materials, etc., typically results in insufficient fault data to train the model for new conditions, thus presenting a challenge in predicting the remaining useful life (RUL) of cutting tools. To address the above issue, this study proposes a multi-source online transfer learning framework for predicting the RUL of cutting tools cross various operating conditions. A source selection strategy is initially proposed to filter the source conditions that contribute to the target modeling from the numerous candidate operating conditions. Then, online transfer learning is employed to transfer valuable knowledge from source domains to target domains while updating the target data online to reflect the actual machining scene. In contrast to the traditional transfer learning approaches, this study utilizes a hybrid physics-data model as the base learner to improve the predictive precision of the RUL in the future scenarios. The results demonstrate its generalizability and flexibility in accurately tracking tool degradation status, and the prediction accuracy of the RUL reaches more than 93 % in various target operating conditions. This study provides reliable technical support for THM in machining actual complex components.

Inkjet printing is a new display manufacturing technology with high efficiency and low cost. Accurate control of the pixel ink volume uniformity is key to realizing large-scale printing display manufacturing. In the continuous printing process on multiple substrates, a major challenge in volume uniformity control is solving the problem of overall droplet volume variation of the printhead caused by uncertain factors such as the changes in printhead temperature, ink pressure and nozzle plate wettability. In this paper, a multi-objective stochastic planning system based on nozzle-combined printing for accurate control of the pixel ink volume uniformity under overall droplet volume variation is proposed, which can improve the flexibility and adaptability of the display printing system. Firstly, a nozzle-combined printing planning system based on a droplet volume uncertainty set and a rolling update strategy is proposed. The overall droplet volume variation is added to the printing planning system as an uncertainty set, and the set parameters are iteratively updated through multilayer closed-loop feedback during the printing process. Secondly, a new multi-objective printing stochastic planning model is established to realize comprehensive optimization of the pixel ink volume uniformity and printing efficiency. Finally, the proposed system was verified by pixel printing experiments on self-developed display printing equipment. The experimental results showed that the system could achieve a pixel film thickness uniformity of 1.54 % under droplet volume variation, which was 80 % lower than that obtained with the traditional method.

The Job-Shop Scheduling Problem (JSSP) is a well-established and classic NP-hard combinatorial optimization issue. The quality of its scheduling scheme directly affects the operational efficiency of manufacturing systems. Priority Dispatching Rules (PDRs) are often utilized to address JSSP in real-world contexts, but the process of creating effective PDRs can be daunting and time-consuming. It also necessitates comprehensive domain knowledge, typically resulting in mediocre performance. In this paper, we introduce a novel reinforcement learning (RL) model called Disjunctive Graph Embedding with Autoencoder Mechanism for Job Shop Scheduling Problems (DGEAM-JSSP), designed to automate PDRs learning. Our proposed model confronts the issue using a Graph Neural Network (GNN) to learn node features that encapsulate the spatial structure of the JSSP graph representation. The ensuing policy network is size-agnostic, enabling effective generalization on larger-scale instances. Additionally, we employ a transformer encoder, incorporating parallel encoding and a self-attention mechanism, to successfully recognize long-term dependencies among operations in large-scale scheduling problems. We also implemented an end-to-end training approach using the Soft Actor–Critic (SAC) algorithm to instruct the two modules. Computational experiment results reveal that, with a single training, our agent successfully learns a superior dispatching policy, surpassing PDRs and state-of-the-art RL frameworks specifically tailored for each JSSP instance size in solution quality, as well as OR-Tools in execution speed. Moreover, results from random and benchmark instances illustrate that the uniquely-modeled learned policies have impressive generalization performance on real-world instances and significantly larger-scale scenarios involving up to 2000 operations.

Controlling the machine’s state of health (SoH) increases the accuracy of the remaining useful life estimation and enables the control of the failure time by keeping the system operational until the desired maintenance time is reached. To achieve system reliability through SoH control, the system controller must consider the impact of its actions on other parameters, such as degradation. This article proposes a structure for designing degradation-aware controllers for systems with available physical models. A system using this approach can learn autonomously, irrespective of the system’s physical structure and degradation model, and opt for control actions that enhance the system’s reliability and availability. To this end, first, a method is proposed to compute the cost associated with the actions taken by the controller. Second, a new cost function is introduced that incorporates the costs associated with degradation into the cost function utilized in model predictive control. In the third step, dynamic programming and deterministic scheduling are used to calculate the optimal action based on the defined cost function. Finally, the proposed control method is validated through simulation, demonstrating its ability to effectively manage machine degradation and achieve optimal performance according to production and maintenance plans.

Remaining useful life (RUL) prediction is crucial for achieving intelligent and predictive maintenance of aircraft engines. In practical applications, advance prediction values smaller than the true values can prevent serious deferred maintenance accidents. Using asymmetric loss functions directly leads to noticeable accuracy degradation, and existing methods often fail to satisfy accuracy and advance prediction requirements. To address this problem, this paper proposes a novel RUL prediction method based on the Prediction Vector Angle (PVA) minimization and Feature Fusion Gate (FFG) improved Transformer network. Specifically, the FFG is proposed to enhance Transformer prediction accuracy by dynamically fusing global and local features. The concept of PVA is first introduced based on the tilting properties of the RUL descent process. The target of the prediction model is cleverly transformed from error minimization to PVA minimization through the cosine similarity loss function. Various experiments on the CMAPSS dataset demonstrate the effectiveness of the proposed method in achieving high accuracy and advanced prediction. Compared to the state-of-the-art method, RMSE is reduced by at least 2.94 % and Score by 7.00 %. Finally, the PVA minimization mechanism significantly improves long short-term memory and convolutional neural network performance. The proposed method is noteworthy for its superiority and applicability.