Journal of Intelligent Manufacturing最新文献

Recent developments and general penetration of society by relations between robots and human beings generate multiple feelings, opinions, and reactions. Such a situation develops a request to analyze this area; multiple references to facts indicate that the situation differs from public opinion. This paper provides a detailed analysis performed on the wide area of human–robot interaction (HRI). It delivers an original classification of HRI with respect to human emotion, technical means, human reaction prediction, and the general cooperation-collaboration field. Analysis was executed using reference outcome sorting and reasoning into separate groups, provided in separate tables. Finally, the analysis is finished by developing a big picture of the situation with strong points and general tendencies outlined. The paper concludes that HRI still lacks methodology and training techniques for the initial stage of human–robot cooperation. Also, in the paper, instrumentation for HRI is analyzed, and it is inferred that the main bottlenecks remain in the process of being understood, lacking an intuitive interface and HRI rules formulation, which are suggested for future work.

This study proposes a new method to estimate the state of the high-frequency induction brazing by using the ensembled Rotational multi-pyramid-transformer tiny (RoMP-T²). The proposed method aims to identify the exact state of an induction brazing process because this information is effective to develop an automatic control system of an induction brazing machine. The proposed state estimation method features three characteristics. First, the method addresses a novel neural network for object detection titled the RoMP-T². This neural network includes a rotational bounding box, multilevel and multiscale feature extraction module, and pyramid vision transformer, which effectively extract features highly correlated to an inducing brazing process from images. Second, the ensembled architecture of the RoMP-T² is addressed to extract features from both optical and thermal images. Bayesian optimization was also addressed to optimize hyperparameters in the ensembled architecture of the RoMP-T². Hence, the ensembled RoMP-T² compensates features extracted from each optical and thermal images, accurately detecting an exact state and location of the filler material during an induction brazing process. Third, the proposed method addresses a cumulative alarm (CA) for determining the completion of the brazing process. The CA significantly reduces a false alarm rate, securing high safety and reliability when the proposed method is implemented to an automation process of the high-frequency induction brazing. An analysis on experiments with optical and thermal images reveals that the ensembled architecture secures the highest accuracy by compensating a limit of feature extraction from each optical and thermal image. The quantitative comparison of the RoMP-T² with other base-line neural networks confirms that the proposed neural network outperforms other neutral networks in both accuracy and robustness perspectives. Furthermore, systematic analysis on experiments reveals that the CA significantly decreases a false alarm rate and thereby increases productivity. These experimental evidences confirm that the proposed framework would be effective to develop an active management system of an induction brazing process, which would be indispensable for manufacturing process automation in a smart factory.

Manufacturing industries are eager to replace traditional robot manipulators with collaborative robots due to their cost-effectiveness, safety, smaller footprint and intuitive user interfaces. With industrial advancement, cobots are required to be more independent and intelligent to do more complex tasks in collaboration with humans. Therefore, to effectively detect the presence of humans/obstacles in the surroundings, cobots must use different sensing modalities, both internal and external. This paper presents a detailed review of sensor technologies used for detecting a human operator in the robotic manipulator environment. An overview of different sensors installed locations, the manipulator details and the main algorithms used to detect the human in the cobot workspace are presented. We summarize existing literature in three categories related to the environment for evaluating sensor performance: entirely simulated, partially simulated and hardware implementation focusing on the ‘hardware implementation’ category where the data and experimental environment are physical rather than virtual. We present how the sensor systems have been used in various use cases and scenarios to aid human–robot collaboration and discuss challenges for future work.

Abstract

Prognostics and Health Management (PHM) approaches aim to intervene in the equipment of production systems before faults occur. To properly implement a PHM system, data-centric steps must be taken, including data acquisition and manipulation, detection of machine states, health assessment, prognosis of future failures, and advisory generation. The data generated by different data sources, such as maintenance management systems, equipment manufacturer manuals, design documentation, and process monitoring and control systems, are fundamental for PHM steps. Discovering and using the knowledge embedded in this data is relevant because, for example, data-driven techniques require knowledge, maintenance data often contain tacit knowledge that can facilitate knowledge transfer and collaboration between maintenance personnel with different levels of experience and expertise, and the knowledge related to the same types of systems could be context-dependent. However, the heterogeneity of data sources, the variety of data types, and the possibility of context-dependent data pose challenges in revealing the real value of data and discovering the useful, yet hidden, patterns embedded in maintenance data that can lead to explicit knowledge. Ontologies can effectively contribute to this issue through the organization of data, semantic annotation, integration, and checking of consistency. Several ontologies contributing to the PHM process have been proposed in the scientific literature. However, to the best of our knowledge, no overview of the available ontologies contributing to the PHM steps of production systems is present in the literature. Therefore, this paper aimed to investigate the ontologies and knowledge graphs proposed in the literature for the PHM of production systems. A systematic analysis and mapping of the literature was performed, and the main information was extracted and discussed according to (i) the type and year of the publication, (ii) the ontological and non-ontological resources adopted for designing the ontology/knowledge graph, (iii) the method adopted for implementing the approach, (iv) the type of application, (v) the step(s) of the PHM process on which the article is focused, and (vi) the type of decisions (strategical, tactical, or operational) to which the ontology/knowledge graph is adopted. Subsequently, the conducted analysis led to the definition of a research agenda in the domain, including the following challenges to address: (1) alignment of the ontologies in the maintenance field with respect to top-level ontologies, (2) connection among the different PHM steps at the operational level, (3) major exploitation of the combination of data-driven AI, ontologies, and reasoning for predictive maintenance, and (4) supporting sustainability-related challenges through the connection between the production system, maintenance system, and product.

Pre-trained large language models (LLMs) have gained significant attention in the field of natural language processing (NLP), especially for the task of text summarization, generation, and question answering. The success of LMs can be attributed to the attention mechanism introduced in Transformer models, which have outperformed traditional recurrent neural network models (e.g., LSTM) in modeling sequential data. In this paper, we leverage pre-trained causal language models for the downstream task of failure analysis triplet generation (FATG), which involves generating a sequence of failure analysis decision steps for identifying failure root causes in the semiconductor industry. In particular, we conduct extensive comparative analysis of various transformer models for the FATG task and find that the BERT-GPT-2 Transformer (Big GCVAE), fine-tuned on a proposed Generalized-Controllable Variational AutoEncoder loss (GCVAE), exhibits superior performance in generating informative latent space by promoting disentanglement of latent factors. Specifically, we observe that fine-tuning the Transformer style BERT-GPT2 on the GCVAE loss yields optimal representation by reducing the trade-off between reconstruction loss and KL-divergence, promoting meaningful, diverse and coherent FATs similar to expert expectations.

The heat generated due to internal and external rotating components, electrical parts, and varying ambient temperatures can cause thermal deformations and significantly impact the precision of machine tools (MTs). Thermal error is crucial in industrial processes, corresponding to approximately 60–70% of MT errors. Accordingly, developing an accurate thermal error prediction model for MTs is essential for their high precision. Therefore, this study proposes an artificial neural network (ANN) model to predict the thermal deformation of a high-speed spindle. However, an important feature for the development of a reliable prediction model is the optimization of the input parameters such that the model generates accurate predictions. Hence, the development of an algorithm to determine the optimal input parameters is essential. Therefore, a genetic algorithm (GA)-based optimization model is also developed in this study to select the optimal input combinations (supply coolant temperature, coolant temperature difference between the inlet and outlet of the spindle, and supply coolant flow rate) for different spindle speeds ranging from 10,000 to 24,000 rpm in increments of 2000 rpm. The R² values of the ANN prediction model are in the range of 0.94 to 0.98 for different spindle speeds. Furthermore, the optimized input parameters are used in single- and dual-spindle systems to verify the accuracy of the developed model as per ISO 230-3. For a single-spindle system, the thermal deformation prediction accuracy of the developed model is in the range of 96.26 to 98.82% and within 1.04 μm compared with the experimental findings. Moreover, when applied to a dual-spindle system, the model’s accuracy is improved by 7.31% compared with that of the variable coolant volume (VCV) method. The maximum deviation of the dual-spindle system can be controlled to within 2.52 μm using the optimized input parameters for a single-spindle system without further optimizing the parameters. The results show that the proposed input attribute optimization (IAO) model can also be adopted for dual-spindle systems to achieve greater prediction accuracy and precision of the machining process, and one industrial cooler can be used for multiple spindles of the same type. In dual-spindle systems operating at different spindle speeds, the power consumption could be reduced by 11% to 34%, and the total lifetime CO₂ emissions could be reduced from 72,981 to 52,595.5 kg. These substantial reductions in energy consumption and CO₂ emissions highlight the potential of dual-spindle systems to contribute to sustainable manufacturing.

Graphical abstract

In advanced manufacturing, the incorporation of sensing technology provides an opportunity to achieve efficient in situ process monitoring using machine learning methods. Meanwhile, the advances of information technologies also enable a connected and decentralized environment for manufacturing systems, making different manufacturing units in the system collaborate more closely. In a decentralized manufacturing system, the involved units may fabricate same or similar products and deploy their own machine learning model for online process monitoring. However, due to the possible inconsistency of task progress during the operation, it is also common that some units have more informative data while some have less informative data. Thus, the monitoring performance of machine learning model for each unit may highly vary. Therefore, it is extremely valuable to achieve efficient and secured knowledge sharing among the units in a decentralized manufacturing system for enhancement of poorly performed models. To realize this goal, this paper proposes a novel knowledge distillation-based information sharing (KD-IS) framework, which could distill informative knowledge from well performed models to improve the monitoring performance of poorly performed models. To validate the effectiveness of this method, a real-world case study is conducted in a connected fused filament fabrication (FFF)-based additive manufacturing (AM) platform. The experimental results show that the developed method is very efficient in improving model monitoring performance at poorly performed models, with solid protection on potential data privacy.

With the development of high-end manufacturing, a variety of sophisticated parts with complex curved surfaces have emerged, and curved surface profile monitoring is of great importance for achieving the higher performance of a part. Benefiting from the recent advancements in non-contact measurement systems, millions of high-density point clouds are rapidly collected to represent the entire curved surface, which can reflect the geometrical and spatial features. The traditional discrete key quality characteristics-based monitoring approaches are not capable of handling complex curved surfaces. A novel curved surface profile monitoring approach based on geometrical-spatial joint features is proposed, which consists of point cloud data preprocessing, Laplace–Beltrami spectrum calculation, spatial geodesic clustering degree definition, and multivariate control chart construction. It takes full advantage of the entire wealth information on complex curved surfaces and can detect the small shifts of geometrical shape and spatial distribution information of non-Euclidean surfaces. Two real-world engineering surfaces case studies illustrate the proposed approach is effective and feasible.

Abstract

Melt pool characteristics reflect the formation mechanisms and potential issues of flaws. Long-term, high-precision, and real-time detection of melt pool characteristics is one of the major challenges in the industrial application of additive manufacturing technology. This work proposes, for the first time, the melt pool characteristics detection platform based on multi-information fusion in the plasma arc welding (PAW) process, which fully utilizes real-time photodiode signals and high-precision, information-rich melt pool temperature fields. By optimizing the detection area and wavelength selection of the platform, particularly through the unique photodiode signal acquisition system capable of detecting the high-sensitivity area of the melt pool, we effectively mitigate the influences of intense arc light and welding wire obstruction on the temperature signals and photodiode signals. Through applying machine learning, the trained model integrates photodiode signals with temperature signals from the high-sensitivity area, thereby achieving real-time acquisition of high-precision average temperature. By combining the fused signals collected from the platform and the scanning results from micro-computed tomography (CT), we evaluate and verify the influence of flaws and droplets on the melt pool characteristics, realizing the determination of flaw occurrence based on the abnormal variations of average temperature. The experimental results demonstrated that the platform fully utilized the advantages of long-term and real-time acquisition of the photodiode signal and the high-precision and information-rich of the melt pool temperature field, achieving long-term, high-precision, and real-time detection of melt pool characteristics.

Part defects in additive manufacturing are more frequent compared to machining or molding. Failures can go unnoticed for hours, wasting resources and extending process cycle times. This paper describes a Machine Learning based method for automated sensing of onset failure in additive manufacturing machinery. Investigations are conducted on a Fused Filament Fabrication (FFF) 3D printer, and the same methods are then applied to a digital light processing 3D printer. The investigation focuses on signal-based analysis, specifically passive sensing of stepper motors relating DC current measurements to the torque on a stepper, as opposed to any active acoustic interrogation of the part. Passive methods are used to characterize the loading on a feeder stepper in an FFF machine, forming a model that can identify early signs of filament-based failure with 85.65% 10-fold cross-validation accuracy. Efforts show filament breakage can be detected minutes before material runout would cause a defect, allowing ample time to pause, correct, or control the print. The machine learning pipeline was not naively conceived but optimized through automated machine learning.