Resistance Spot Welding (RSW) is a popular technique for joining sheet metals. Due to the involvement of multiple process parameters, ensuring continuous quality assessment is crucial. While machine learning (ML) methods have demonstrated effectiveness in monitoring welding quality, their application is often limited by the high cost and time required to collect sufficient training data. A promising solution is the integration of prior knowledge into the ML process through transfer learning (TL), enabling the development of more generalized models. This study proposes a TL-based methodology for RSW quality monitoring in the case of limited datasets. An experimental campaign was conducted to generate a target domain dataset comprising welding points produced under varying process conditions. A neural network was trained to predict the nugget diameter, which is typically more difficult and time-consuming to measure. Subsequently, TL techniques were employed to transfer knowledge from a model trained to predict the peak load of tensile shear tests to the nugget diameter prediction model. The source domain dataset used for TL included samples obtained under diverse experimental conditions, encompassing different materials, welding parameters, and electrode types. The results demonstrate that TL enhances model generalization and predictive performance across the full range of nugget diameters, including challenging cases with extremely small or large values that typically hinder accurate prediction. Accordingly, the model performance improved by 25

Additive Manufacturing (AM) offers unprecedented design freedom, yet evaluating the manufacturing resources required for complex geometries remains a significant computational bottleneck during the early design phase. Traditional slicing software, while accurate, is too slow to support high-frequency iterative workflows such as Generative Design. To address this, we propose the Parameter-Aware Geometric Estimator (PAGE-Net), a Graph Neural Network (GNN) framework designed to instantly predict Life Cycle Inventory (LCI) data–specifically Part Mass, Support Mass, and Total Print Time–directly from raw 3D meshes. Unlike existing voxel-based deep learning methods that suffer from discretization errors or static parameter assumptions, PAGE-Net leverages Feature-Steered Graph Convolutions (FeaStConv) to extract topological features from the native mesh while dynamically incorporating user-defined printing parameters (e.g., infill density, layer height). Trained and validated on a comprehensive dataset of approximately 90,000 geometries using a robust 3-fold cross-validation scheme, the model achieves high predictive accuracy, with R^2 scores exceeding 0.96 for material and time estimation. Computational benchmarks demonstrate an average inference time of 77 milliseconds per object–offering a speedup of approximately 30× compared to optimized command-line slicing. By providing near real-time feedback on manufacturing resources, this framework serves as a critical enabler for data-driven Eco-Design and automated topology optimization.

One-of-a-kind production refers to the manufacturing of highly customised products tailored to specific customer requirements. In this context, the ability to effectively reuse knowledge is crucial, as it can significantly reduce production time and costs. However, the feasibility of Knowledge Reuse in One-of-a-kind production remains uncertain, due to the lack of studies examining how conditions in One-of-a-kind production influence the emergence of barriers to Knowledge Reuse. This research aims to address this issue by systematically mapping the characteristics of product development processes in One-of-a-kind production companies against the most common barriers to Knowledge Reuse. The results are shown in matrix form (the BROK matrix), representing the relationships between characteristics and barriers. The findings indicate that extreme customisation under time pressures, the continuous influence of customers, the high uncertainties caused by the Once-successful approach, and the high-level concurrency, combined with the limited resources available, significantly impact the Knowledge Reuse process at multiple levels. Finally, an industrial case study was conducted to validate the BROK matrix. The results of the semi-structured interviews show a strong alignment between the barriers suggested by the BROK matrix and those observed by the production managers.

Additive Manufacturing is a transformative production technology, but its sustainable adoption is hindered by the lack of tools to evaluate the environmental and economic impacts of design decisions during the early conceptual stages. Particularly, the challenge lies in accurately predicting key manufacturing outputs (e.g., print time, material consumption, and support material) early in the design process, translating abstract design choices into tangible sustainability metrics.This paper introduces a Graph Neural Network (GNN) framework that learns directly from the high-fidelity mesh geometry of 3D models and incorporates user-defined printing parameters (e.g., layer height, infill density). The GNN predicts three critical outputs: final part mass, support material mass, and total building time. These predictions form the foundation for a real-time environmental impact assessment based on Life Cycle Assessment principles. By enabling designers to instantly explore the trade-offs between part quality, build speed, and environmental footprint without requiring deep process knowledge, this tool facilitates early-stage optimization, minimizes material waste, and enhances the overall efficiency of the additive manufacturing process.The GNN was trained on a dataset of approximately 60,000 instances derived from the Slice-100K dataset, which includes a wide range of geometries and printing configurations. It demonstrated high predictive accuracy, explaining over 92% of the variance for all three outputs. The model outperformed existing methods in predictive accuracy and flexibility, particularly by enabling dynamic predictions across 81 different printing configurations. Experimental validation on physical 3D-printed parts confirmed the model's practical utility, with acceptable prediction errors for early-stage design.

As the global population is expected to reach 10.3 billion by the mid-2080s, optimizing agricultural production and resource management is crucial. Climate change and environmental degradation further complicate these challenges, impacting crop productivity and food security. Traditional farming methods struggle with efficiently managing nutrients and water while ensuring high-quality products, leading to resource wastage and food safety concerns. This study aims to develop a hybrid model combining machine learning and physics-based techniques to predict fresh weight, leaf area, nitrate levels, and water consumption in lettuce grown in aeroponic systems, thereby enhancing resource management and product quality. We integrated a physics-based model with machine learning algorithms to create a dynamic hybrid framework. The model was validated with real-time data from aeroponic systems, showing good predictive performance, particularly for fresh weight and total leaf area. In contrast, predictions of nitrate content and water consumption were less accurate, due in part to smaller training datasets and limitations of the physics-based component under soilless conditions. Despite these challenges, the hybrid model offers a promising solution for optimizing controlled environment agriculture, addressing critical challenges in modern agriculture by improving efficiency and sustainability.

In the era of Industry 4.0/5.0, collaborative robots (cobots) play a pivotal role in transforming manufacturing processes by enabling human-robot collaboration and adaptability. However, realizing the full potential of cobots necessitates advanced digital technologies. This paper introduces a digital twin framework specifically tailored for collaborative robots within the context of Industry 4.0 and 5.0. The framework leverages digital twin technology to create a virtual representation of physical cobots, enabling real-time monitoring and optimization of their performance. Emphasizing data flow, communication protocols, and visualization tools, the framework facilitates seamless integration of cobots into smart manufacturing environments. The digital twin application is implemented in the FlexSim software and it has been exploited to perform dynamic task allocation among UR3e cobots. Through the use case, the paper demonstrates the effectiveness and versatility of the proposed framework, offering a roadmap for harnessing the potential of digital twin for cobots management.

The increasing focus of the European Union on a sustainable and circular economy in the industrial sector has led to the development of the Digital Product Passport (DPP). It acts as a digital identity assigned to physical products, providing a structured record of essential lifecycle information, including composition, maintenance, environmental impact, and end-of-life management. By facilitating data flow among authorities, supply chain businesses, and consumers, the DPP enhances traceability and regulatory compliance, serving as a key lever for circular economy strategies. Integrating the DPP into a PLM system presents significant challenges, such as ensuring data standardization, achieving interoperability, and maintaining compliance with evolving regulations. This paper explores the theoretical integration of the DPP into a PLM system, analyzing technical requirements, exploring potential architecture, and assessing implications for product data management. The research identifies key functional areas where the DPP can enhance PLM capabilities, such as data exchange, regulatory compliance, and supply chain transparency. By proposing a conceptual framework for DPP-PLM integration, this work aims to lay the foundation for future implementations, thereby contributing to a more sustainable and circular approach to product lifecycle management.

Climate change and population increase are becoming a threat to human feeding. New technologies and practices are under development, and a significant effort is being put into developing indoor farming, which allows for all-year-round production of high-quality food, regardless of the climate. Moreover, indoor farming promises extreme water and chemical usage reduction, specifically when the system is autonomously regulated with an IoT architecture. Despite these attractive characteristics, indoor systems require considerable energy to provide adequate temperature and lighting for cultivated crops. This demand is often high enough to make the production system economically unsustainable. This work aims to develop a cultivation protocol for baby lettuce plants (up to three weeks old plants) that can increase overall productivity while mitigating the issue of high energy demand. To this aim, we performed a Design of Experiment to assess crop responses to different levels of nutrients, temperature, and light intensity with the productivity of the system and the quality of the harvested product. The collected data were used to design a dynamic cultivation protocol, which defines different growing conditions according to the plant development stage. Results demonstrate that the dynamic protocol can enhance system productivity by up to 25 % in biomass accumulation, compared with the productivity obtained with fixed growing conditions, while maintaining the same high quality. Furthermore, the improvement is achieved without increasing the resource use, confirming the potential of this approach to enhance the economic sustainability of indoor soilless farming.

The nesting problem of 2D shapes, which has impactful application in the cutting and packing fields, has been studied for many years. Previous papers are mainly focused on proposing new algorithms and prove their efficiency in terms of packing density or computation time. However, the results are reported only on few datasets and the comparison is done only with respect to few competing algorithms. The aim of the paper is to analyse and compare the results obtained by strip-packing algorithms published in the last 20 years. The results show that the effectiveness of the algorithms varies widely across different datasets, and there is a lack of comprehensive benchmarking that considers both the quality of solution and the computational time required to achieve it. Furthermore, since no algorithm clearly outperforms all the others, further methods to address the nesting problem with reinforcement learning and neural networks could be investigated to improve the generalization ability on the nesting problem.

Due to the emergence of global challenges, the European policy called Industry 5.0 demands technologies to become resilient, sustainable and humancentric; the integration of Artificial Intelligence (AI) and Knowledge Management (KM) (i.e., AI-based KM) represents an opportunity to accomplish this ambitious goal thanks to the possibility of combining humans and computers' unique skills. While the compliance of AI-based KM with human-centric requirements is apparent, its long-term sustainability is still an object of discussion. In response to I5.0's request for metrics of sustainability, this paper presents a lifecycle analysis of AI-based KM's applications retrieved from literature to establish whether they can be deemed socially, environmentally, and economically sustainable. A list of sources was collected by means of a literature search, to be later evaluated through sustainability criteria specific for the design and use phases of life. It was found that AI-based KM applications supporting fractions of the knowledge lifecycle may become fragile in the long-term, because potentially leading to repositories' indefinite growth and information overload in humans and computers; furthermore, it emerged that not enough attention was paid to AI's accountability and data governance. The mentioned pain points, complemented in this article with practical examples of the counter-measures adopted by virtuous sources, provide scholars with practical indications to lead future research efforts, and contribute to the validation of a framework to assess the longterm sustainability of AI solutions.

The job shop scheduling problem, a notable NP-hard problem, requires scheduling jobs with multiple operations on specific machines in a predetermined order. A strong assumption is that all the information of the manufacturing environment is known in advance and there is no modification during the scheduling process. However, the real-world environment is significantly affected by uncertainties. The dynamic job shop scheduling is a variant of the job shop scheduling problem in which the scheduling environment is subject to changes over time including variations in job arrival times, processing times, machine breakdowns, resource availability and job priority. To address this issue, this paper presents a single-agent reinforcement learning algorithm, which implements a proximal policy optimization that uses masking to reduce the search space and improve efficiency. The algorithm was tested in both deterministic and dynamic environments and compared to traditional scheduling methods. The results demonstrate that the proposed approach is comparable to traditional methods in deterministic cases and outperforms them in dynamic environments. These findings emphasize the potential of reinforcement learning in addressing and optimizing complex scheduling challenges.

The quality of resistance spot welding (RSW) joints is strongly affected by the condition of the electrodes. This work develops a machine learning-based tool to automatically assess the influence of electrode wear on the quality of RSW welds. Two different experimental campaigns were performed to evaluate the effect of electrode wear on the mechanical strength of spot welds. The resulting failure load of the joints has been used to define the weld quality classes of the machine learning tool, while data from electrode displacement and electrode force sensors, embedded in the welding machine, have been processed to identify the predictors of the tool. Some machine learning algorithms have been tested. The most performing algorithm, i.e., the neural network, achieved an accuracy of 90%. This work provides important theoretical and practical contributions. First, the decreasing thermal expansion of the weld nugget as the electrode degradation advances results in a strong correlation between the difference of the maximum displacement value and the last value recorded during the welding and the relative failure load. Then, this work offers a practical decision support tool for manufacturers. In fact, the automatic detection of low-quality welds allows to reduce or eliminate unnecessary redundant welds, which are performed to compensate for the uncertainty of electrode wear. This leads to savings in time, energy, and resources for manufacturers. Finally, general recommendations for the timing of redressing or replacing the electrode are provided in the manuscript based on the company willingness to accept some non-compliant welds or not.

[Context] Within the framework of PLM; Circular Economy (CE) is an important concept that seeks to design out waste and pollution, keep products and materials in use and regenerate natural systems. Early Design Stages (EDS) are crucial because they set the foundation for the rest of the production process and can determine the product's overall functionality, usability, and manufacturing processes. [Problem] As companies increasingly recognize the benefits of transitioning to a circular economy, there is a growing need for tools and methodologies to support the design of circular products and services. [Proposal] This paper presents a CE card deck as a novel approach to facilitate the early stages of product development. The deck consists of 10 cards based on Morseletto's "Targets for a circular economy" work that represents mainstream circular economy principles and strategies and can be used by designers, engineers, and other stakeholders to generate ideas, evaluate options, and make informed decisions. The results of a pilot study with design and engineering students (Master level) suggest that the card deck can support the exploration of CE concepts and facilitate the identification of circular solutions, in a Design by AM context (DbAM). The paper concludes with a discussion of the potential benefits and limitations of the card deck approach, and its integration into a PLM framework, and gives suggestions for future research.

Vertical farming has gained increased attention in recent years due to its capacity to reduce the environmental impact of agricultural production in terms of water consumption and soil and fertilizer usage. In the literature, many works describe and evaluate applications of vertical farming. However, no work addresses the issue of classifying the KPIs for vertical farming and highlights both the most assessed aspects and the lack of evaluations. The main contribution of this study is to conduct a literature review to identify and classify the KPIs used in vertical farming. To this aim, we first proposed a methodology to define the KPI categories. Then, we identified the KPIs used in the literature, and we classified them according to the defined categories. Finally, we analyzed the obtained results. As a result, a collection of 78 KPIs were compiled and organized into the proposed categories. The analyses on the frequency of the KPIs allow us to conclude that the KPIs related to productivity are the most used as compared to those related to sustainability and quality. Furthermore, very few papers perform a cross-category evaluation. This study underscores the necessity for a more balanced consideration of productivity, quality, and sustainability in the context of vertical farming.

Resistance spot welding is used to join two or more metal sheets by overlapping them and producing a localized spot weld through the simultaneous application of pressure and heat to the weld area by two electrodes. In case of unsuitable choices taken when setting the welding process, some issues might arise on the assembled workpiece, and even cause some damage to the structure of the welded materials. The presence of expulsion, for instance, is frequently used as a welding quality indicator. The present work aims to address the challenge of automatically detecting the presence of expulsions during the welding process. By implementing an automated expulsion detection system, manufacturers can proactively monitor electrode condition and schedule maintenance or replacement before weld quality begins to deteriorate. To achieve this goal, a deep learning algorithm, specifically a Convolutional Neural Network model, is proposed. The implementation of such an algorithm offers significant advantages in the quality control process of resistance spot welding operations.

Digital management system is widely used in the business activities of enterprises. Practice has proved that the implementation of the manufacturing execution system (MES) can better monitor and manage the production process, improve the production efficiency of enterprises, and effectively realize zero defect management (ZDM). Against this background, the influencing factors of implementing MES in ZDM enterprises in digital transformation were obtained by using the literature extraction-Delphi method, and the relationship between the factors was analyzed by using the system dynamics simulation model in this study. It is found that different from the existing research works on the implementation of MES in enterprises, staff preparation and level of information sharing are the most influential factors and play an important role in the implementation of MES in ZDM enterprises. Equipment preparation and client preparation followed closely, with supplier implementation team and scale infrastructure conditions playing a key role in providing capability support. This finding provides the direction for enterprises to improve the relevant implementation measures in time to ensure the effective implementation of MES in ZDM enterprises, and also provides a breakthrough for relevant researchers to find valuable research fields.

Data management systems represent a powerful tool for collecting and managing real-time data in a company. One-of-a-Kind Production (OKP) firms, which leverage on their knowledge about past products and manufacturing processes to reduce lead time, are the ones which can potentially benefit the most from deploying systems like PLM and MES. However, the benefits arising from data management systems are still limited due to the scarce integration among them. In particular, for OKP companies it is crucial to be able to collect and formalize the knowledge generated by operators during the manufacturing process, in order to constantly improve the design process. In this way, the iterations of trial and error can be reduced, saving time and costs. This paper proposes a framework (i.e., aKnowledge Base System) for the integration of aPLM software used for product design and production cycle definition, and a MES software able to collect operators' feedbacks about manufacturing criticalities. The integration is realized by means of a centralized database, able to receive data from both systems and to relate them to generate useful information. Finally, a case study developed on a car prototyping company is presented to exemplify the advantages of systems integration.

Digital twins are considered the next step in IoT-based cyber–physical systems; they allow for the real-time monitoring of assets, and they provide a comprehensive understanding of a system behavior, allowing for data-driven insights and informed choices. However, no comprehensive framework exists for the development of IoT-based digital twins. Moreover, the existing frameworks do not consider the aspects introduced by the Industry 5.0 paradigm, such as sustainability, human-centricity, and resilience. This paper proposes a framework based on the one defined as the outcome of a project funded by the European Union between 2010 and 2013 called the IoT Architectural Reference Model (IoT-A or IoT-ARM), with the aim of the development and implementation of a standard IoT framework that includes digital twins. This framework establishes and implements a standardized collection of architectural instruments for modeling IoT systems in the 5.0 era, serving as a benchmark for the design and implementation of an IoT architecture focused on digital twins and enabling the sustainability, resilience, and human-centricity of the information system. Furthermore, a proof of concept of a monitoring digital twin for a vertical farming system has been developed to test the validity of the framework, and a discussion of applications in the manufacturing and service sectors is presented.

Digital twins are employed to monitor, analyze, and predict the behavior of a manufacturing system. In literature, the concept of digital twin is mainly associated with simulation models, and only a few works addressed the integration between physics-based and data driven models. This work extends this definition and describes how different kinds of knowledge can be integrated in a hybrid knowledge based system supporting digital twin. Following the Industry 5.0 paradigm, the hybrid system can improve the human-centricity, sustainability, and resilience of a manufacturing system. The hybrid digital twin is structured as an enterprise information system, which can be connected to the other information systems of the company, such as ERP, MES, and PLM, as well as other platforms of total quality management and total productive maintenance. A discussion on the application of the proposed approach in fusion welding, as an example of a very complex process, is also presented.