Computers in Industry最新文献

Digital Twins (DT) encompass virtual models interconnected with a physical system through data links. Although DTs hold significant potential for positive organisational impact, their successful adoption in industrial practice remains limited. Whereas existing research predominantly focuses on technical challenges, more recent studies underscore the importance of addressing organisational and human factors to overcome implementation barriers. One central aspect in this context is stakeholder communication, especially given the ambiguous nature of the term DT in academic and industrial discussions. To expand the limited understanding of the factors causing challenging DT stakeholder communications, this article presents findings from an extensive exploratory study. It involves 27 in-depth interviews and two focus groups with highly experienced DT professionals. By employing grounded theory and the Gioia methodology, a grounded model for DT stakeholder communication challenges is derived. This model reveals the complex communication dynamics within DT projects, emphasising the emergence of novel stakeholder communication patterns that heavily rely on multidisciplinary collaboration. In total, 28 communication challenges were identified, grouped into eight theoretical themes and categorised into two aggregate dimensions: human- and organisation-centric challenges. Additionally, the study identified 15 practices, e.g., defining clear objectives, and starting small and building gradually, that organisations are following to mitigate these challenges. As a result, this article provides the theoretical groundwork for a comprehensive understanding of DT stakeholder communication and its associated challenges by revealing distinctive features and offering practical guidance to overcome critical challenges in DT projects.

Object picking is a fundamental, long-lasting, and yet unsolved problem in industrial applications. To complete it, 6 Degrees-of-Freedom pose estimation can be crucial. This task, easy for humans, is a challenge for machines as it involves multiple intelligent processes (for example object detection, recognition, pose prediction). Pose estimation has recently made huge steps forward, due to the advent of Deep Learning. However, in real-world applications it is not trivial to compute it: each use-case needs an annotated dataset and a model robust enough to face its specific challenges. In this paper, we present a comprehensive investigation focused on a specific use-case: the picking of four industrial objects by a collaborative robot’s arm, addressing challenges related to reflective textures and pose ambiguities of heterogeneous shapes. Thus, Artificial Intelligence is crucial in this process, utilizing Convolutional Neural Networks to discern an object’s pose by extracting hierarchical features from a single image. In detail, we propose a new synthetic dataset of industrial objects and a fine-tuning method to close the sim-to-real domain gap. In addition, we improved an existing pipeline for pose estimation and introduced a new version of an existing method, based on Convolutional Neural Networks. Finally, extensive experiments were conducted with a Universal Robot UR5e. Results show our strategy achieves good performances with an average successful picking rate of 75% on these new objects. Considering the lack of available datasets for pose estimation, coupled with the significant time and labor required for annotating new images, we contribute to the scientific community by providing a comprehensive dataset, and the associated generation and estimation pipelines.¹

Generative Artificial Intelligence (AI) models serve as powerful tools for organizations aiming to integrate advanced data analysis and automation into their applications and services. Citizen data scientists—individuals without formal training but skilled in data analysis—combine domain expertise with analytical skills, making them invaluable assets in the retail sector. Generative AI models can further enhance their performance, offering a cost-effective alternative to hiring professional data scientists. However, it is unclear how AI models can effectively contribute to this development and what challenges may arise. This study explores the impact of generative AI models on citizen data scientists in retail firms. We investigate the strengths, weaknesses, opportunities, and threats of these models. Survey data from 268 retail companies is used to develop and validate a new model. Findings highlight that misinformation, lack of explainability, biased content generation, and data security and privacy concerns in generative AI models are major factors affecting citizen data scientists’ performance. Practical implications suggest that generative AI can empower retail firms by enabling advanced data science techniques and real-time decision-making. However, firms must address drawbacks and threats in generative AI models through robust policies and collaboration between domain experts and AI developers.

After two decades of research and development, process mining techniques are now recognized as essential analysis tools, as they have their own Gartner Magic Quadrant. The development of process mining techniques is rooted in process-related research fields such as Business Process Management and fueled by increasing data availability. To cope with the complexity of business processes, the focus of process mining techniques needs to go beyond workflow-like processes, that represent the life-cycle of a single case and enable multiple object types and events. This can only be accomplished by capitalizing on essential concepts from production and logistics domains, such as Bills-of-Materials (BOMs), and Customer Order Decoupling Points (CODPs). Pioneer researchers, e.g. Hans Wortmann contributed to the development of Enterprise Resource Planning, enterprise modeling, product models, and lean manufacturing. Experiences from these fields help to lift the process mining domain from case-based (i.e. workflow mining) to object-centered process mining. These contributions could be realized by conducting insightful case studies at company sites, one of them being discussed in this paper. The evaluation of process mining techniques is elaborated by proposing an “evaluation ladder”, and its application is shown in the case study under consideration.

Accurate X-ray image defect segmentation is of paramount importance in industrial contexts, as it is the foundation for product quality control and production safety. Deep learning (DL) has demonstrated powerful image scene understanding capabilities and has achieved unprecedented performance in defect segmentation tasks. However, existing DL methods suffer from significant performance degradation when facing low-contrast X-ray images, as the core information of defects is often obscured and the profile details are ambiguous. To address this issue, this paper explicitly decomposes the X-ray defect segmentation task into two subtasks: core feature learning and elasticity profile refinement, allowing task “serial” decomposition and performance “parallel” improvement. On this basis, a novel core-profile decomposition network (CPDNet) is developed to achieve accurate defect segmentation of X-ray images. Specifically, the core feature learning module is designed to construct the effective feature space from two views, discriminative and structural, to extract defect-related core features from X-ray images. Subsequently, the elasticity profile refinement module is developed to further improve the defect segmentation performance, which makes the first attempt to define the profile refinement as an out-of-distribution detection and leverage the elasticity score to refine the profile details at the pixel level. To fully evaluate the presented method, we conduct a series of experiments using two real-world X-ray defect datasets, and the results demonstrate that the CPDNet outperforms state-of-the-art methods.

Efficient firefighting operations are crucial for ensuring the safety of firefighters and preventing direct exposure to high-temperature and high-radiation environments. However, traditional firefighting robots face the challenges of low efficiency, high misjudgment rates, and difficulty in control during firefighting processes, particularly in extremely complex and dynamically changing fire scenes. Therefore, this article proposes a novel convolution-based context-guided dual attention lightweight network (CG-DALNet) model to develop efficient firefighting methods for firefighting robots. To expand the field of fire perception, this study employs monocular vision from drones to assist ground firefighting robots in autonomous firefighting decision-making in an end-to-end manner. By introducing depthwise separable convolutions to construct the feature backbone layer, the number of the parameters in the model is reduced. To better understand target position information in fire scenes, we propose a position attention module guided by contextual features to enhance the model's positional awareness. Additionally, to efficiently integrate feature information at different scales in the fire scene, we adopt a residual-connected convolutional kernel attention module to enhance the model's ability to express complex fire scene features. Numerical experiments show that the proposed CG-DALNet lightweight network model achieves significant performance improvement in autonomous firefighting tasks for robots. This research provides an innovative solution for autonomous firefighting methods for firefighting robots and demonstrates its effectiveness and potential.

The booming Internet economy and generative artificial intelligence have driven the rapid growth of the digital content trading industry, creating an urgent need for the fair protection of the rights of both buyers and sellers. To meet this need, a technique known as buyer–seller watermarking has emerged. Despite its existence, the majority of existing buyer–seller watermarking schemes adopt the owner-side embedding mode, which results in poor scalability. While a handful of schemes adopt the client-side embedding mode to enhance scalability, they either require the deep involvement of a trusted third party or fall short of ensuring complete fairness due to the unresolved unbinding problem. To address these challenges, this paper proposes a fair and scalable watermarking scheme for digital content transactions based on proxy re-encryption and digital signatures. For one thing, this scheme solves the unbinding problem and ensures complete fair protection of the rights of both buyers and sellers. For another, it adopts the client-side embedding mode and has good scalability. Additionally, it eliminates the need for a trusted third party. Finally, theoretical analysis and experiments demonstrate that the proposed scheme achieves the intended design goals and possesses superior efficiency advantages.

Computational fluid dynamics (CFD) has been extensively used as a simulation tool for product development in various industrial fields. Engineers sequentially query the CFD simulator to evaluate their design instances, during which they improve the new designs based on previous evaluations. The high cost of performing CFD simulations for numerous design instances is a practical challenge. To reduce this cost, machine learning (ML) approaches have been employed to approximate CFD simulations. Although ML enables the fast approximation of CFD, it can suffer from low accuracy when making predictions for design instances that significantly deviate from the training dataset. In this study, we propose a CFD-ML combined system based on stream-based active learning to utilize the CFD simulator cost-efficiently. The proposed method has two main objectives: reducing the number of CFD simulations and ensuring high accuracy of the ML approximations. When a design instance is queried, the CFD-ML system interchangeably uses the CFD simulator and the ML model depending on the predictive uncertainty of the ML model. If the uncertainty of the ML model is high, the CFD simulator is used to obtain an evaluation result, which is subsequently used to enhance the ML model. Conversely, if the uncertainty is low, the ML model is used to obtain an approximated evaluation result. The CFD-ML system reduces computational costs compared to exclusive reliance on the CFD simulator and yields more accurate evaluations compared to exclusive reliance on the ML model. We demonstrated the effectiveness of the proposed method through a case study on a centrifugal fan development task.

Trade restrictions, the COVID-19 pandemic, and geopolitical conflicts have significantly exposed vulnerabilities within traditional global supply chains. These events underscore the need for organisations to establish more resilient and flexible supply chains. To address these challenges, the concept of the autonomous supply chain (ASC), characterised by predictive and self-decision-making capabilities, has recently emerged as a promising solution. However, research on ASCs is relatively limited, with no existing studies specifically focusing on their implementations. This paper aims to address this gap by presenting an implementation of ASC using a multi-agent approach. It presents a methodology for the analysis and design of such an agent-based ASC system (A2SC). This paper provides a concrete case study, the autonomous meat supply chain, which showcases the practical implementation of the A2SC system using the proposed methodology. Additionally, a system architecture and a toolkit for developing such A2SC systems are presented. Despite limitations, this work demonstrates a promising approach for implementing an effective ASC system.

The Digital Twin paradigm is a very promising technology that can be applied to various fields and applications. However, it lacks a unifying framework for classifying and defining use cases. The goal of this paper is to address the identified gap. Using a field study and a bottom-up approach, it aims to categorize the various uses of the industrial Digital Twin to help formalize the concept and rationalize its adoption by a range of industrial sectors. The study is based on an iterative process of collecting use cases from a wide variety of verticals, applying grounded theory principles. The usage scenarios were extracted, synthesized, grouped and abstracted to develop an actionable use cases classification framework. This article presents the resulting taxonomy and illustrates it by detailing real industrial use cases, including their value proposition and application areas. This collection, classification and analysis of use cases led to a study of the common aspects proposed in academic and industrial definitions of the Digital Twin. The goal was to combine and generalize these aspects into a pragmatic and unifying definition, on which the Alliance for Industry of the Future (AIF) committee has converged. The main contributions of this work include proposing, from a joint industrial and academic perspective, (i) the first domain-independent and industry-focused systematic collection of Digital Twin use cases, (ii) a comprehensive framework for analyzing and classifying Digital Twin use cases and their requirements, and (iii) a consensual general definition of the industrial Digital Twin to contribute to the structuring and standardization of this very active ecosystem.