Energy, environment, and economy are driving forces behind adopting circular economy principles in industrial production, particularly in the sheet metal industry, where resource optimization and waste reduction are prioritized. Remanufacturing, especially in automotive manufacturing, offers significant material reuse and conservation opportunities. However, challenges persist in developing effective product designs and remanufacturing methods for material recovery. Artificial intelligence (AI), machine learning, and big data analytics present promising solutions to these obstacles, facilitating sustainable practices in metal product re-manufacturing. Predictive models, such as those for incremental sheet forming (ISF), streamline the assessment of remanufacturing feasibility, reducing costs and time associated with technical complexities. Integration of AI enables forecasting potential failures and optimizing remanufacturing processes, resulting in considerable savings in effort, energy, and time while advancing circular economy initiatives. In this study, a series of ISF experiments was conducted using a computer numerically controlled machine to redeform a contoured part. Subsequently, using an experimental dataset, an experimental dataset, three AI classifier models-artificial neural network, random forest, and support vector machine (SVM)-were developed to assess the feasibility of remanufacturing a predeformed sheet metal part. SVM emerged as the top-performing algorithm, demonstrating robust classification capabilities. The present study focuses on a limited parameter set that overlooks other influential factors such as the previous manufacturing processes, material composition, and sheet thickness. Expanding the scope to incorporate these factors will enable more accurate assessments and optimized manufacturing processes, ultimately benefiting the remanufacturing industry.

Sheet metal forming is crucial for producing lightweight, durable components. Single-point incremental forming (SPIF) presents an alternative to deep drawing (DD), but its use remains limited due to insufficient analysis of its pre- and post-forming stages. This study compares the cost of goods sold (COGS) and environmental impacts of SPIF and DD for an asymmetric battery cell casing at various production volumes. It utilizes the manufacturing process design (MPD) and life cycle assessment (LCA) methods to assess all stages needed to produce a part. SPIF shows higher costs and impacts at 1000 parts/year due to longer forming times. Global warming potential (GWP) is similar at 1000-2000 parts/year, after which the impacts of SPIF outweigh those of DD. Sensitivity analysis highlights blank handling as a significant cost driver above 500 parts per year. Process improvements would reduce cycle times, equipment use, and environmental impacts, increasing the competitiveness of SPIF.

Biomaterials offer a promising alternative to nonrenewable engineering materials, and industrial hemp has garnered increasing interest due to its eco-friendly cultivation, short growth cycle, phytoremediation capabilities, and diverse industrial applications. While prior research has explored the use of hemp fiber and hurd in additive manufacturing, the use of low-cost hemp straw remains largely unexamined. In parallel, the brittleness of polylactic acid (PLA) limits its broader application in additive manufacturing, motivating the exploration of strategies to improve its ductility. This study investigates whether incorporating hemp straw powder into PLA can improve deformation capacity while reducing the environmental impacts of parts produced via fused filament fabrication. Hemp straw-PLA composite filaments and pure PLA filaments were extruded and used to produce tensile test specimens. Environmental impacts were evaluated using a gate-to-gate process-based life cycle assessment conducted in accordance with ISO 14040:2006. Modeling was performed in SimaPro using the ecoinvent database, with impact assessment carried out using ReCiPe 2016, Cumulative Energy Demand, and IPCC 2021 methodologies. Sensitivity analyses examined transportation distances, impact assessment methods, specimen thickness, alternative proxies for filler material, and electricity mix scenarios. Results from ReCiPe 2016 indicate that the environmental impact of a pure PLA tensile test specimen (6.83 mPt) is approximately four times higher than that of a hemp straw-PLA specimen (1.63 mPt), with similar trends observed across other impact methods. These differences are primarily attributable to variations in material use, energy consumption, and emissions during filament production and fused filament fabrication. Exploratory tensile testing indicated that increasing specimen thickness can moderately increase elongation at break for pure PLA, although the improvement remained substantially lower than that observed for hemp straw-PLA composites. Overall, the study demonstrates that incorporating hemp straw filler into PLA can improve deformation capacity while reducing environmental impacts under the conditions examined.

The global pandemic has exposed deficiencies in supply chains in rapidly responding to urgent medical needs and demonstrated the challenge of promptly acquiring medical supplies during emergencies. Even under policy tools and using existing e-commerce alternatives, being able to promptly acquire medical supplies, including personal protective equipment and medical devices, is still a significant concern for future emergency situations. Hence, the details for an open, extensible, service-oriented product marketplace platform that enables rapid matchmaking between consumers and manufacturers in such emergency situations are presented in this work. This marketplace is especially designed based on a microservice architecture, where each service operates independently and communicates through message exchanges, optimizing the supply chain for items according to customers' preferences and constraints regarding budget and lead time. In addition, the results of implementing a simplified prototype of the platform, where a healthcare worker is using the online product marketplace to order a replacement part of a respirator mask, are also demonstrated. By using this online product marketplace, manufacturers will benefit from improved accessibility to their products, services, and manufacturing resources (also under normal operations), while consumers will gain unprecedented access to existing products and manufacturing resources.

Abstract Sustainability assessment in biobased manufacturing is commonly performed as a retrospective, static analysis, limiting its usefulness for operational decision-making and stakeholder engagement. This study advances cyber-physical sustainability assessment as an emerging paradigm in which life cycle-based sustainability evaluation is embedded within digitally connected manufacturing systems. A cloud-enabled sustainability assessment tool is developed that integrates automated life cycle inventory importation, life cycle impact assessment using Brightway2-based module, Industry 4.0-informed scenario modeling, and interactive decision support through a web-based dashboard. The approach enables multi-pillar evaluation across environmental, economic, and social dimensions while supporting near-real-time updates, transparent computation, and stakeholder-accessible visualization. The proposed system is demonstrated for the industrial hemp straw production and decortication process, assessing fiber and hurd production under cradle-to-gate system boundaries. Environmental indicators include global warming potential, acidification, cumulative energy demand, land use, and water use; economic performance is evaluated through energy, labor, logistics, and total production costs; and social performance is represented using operational metrics such as labor hours, overtime, training, and safety incidents. By coupling open-source life cycle modeling with cloud-based data infrastructure and actionable visualization, this work demonstrates how sustainability assessment can transition from post hoc reporting to a cyber-physical decision-support capability for biobased manufacturing. The proposed approach improves transparency, responsiveness, and stakeholder engagement, offering a scalable foundation for data-driven sustainability management in emerging biobased value chains.

Global supply chain fragility demands transparent, dynamic risk management tools, particularly for resource-constrained Small and Medium-sized Enterprises (SMEs). Current opaque “black box” methodologies employed by commercial rating agencies for Environmental, Social, and Governance (ESG) scoring are often costly, yielding non-prescriptive results that fail to define specific, evidence-based mitigation strategies. This paper introduces an SLM- and LLM-based “Glass Box” Facility Risk Score framework, leveraging Artificial Intelligence to provide a holistic assessment of a supplier’s underlying structural health, operational reliability, and ethical posture. The framework employs an efficient “scrape-then-verify” methodology for automated, scalable data aggregation from diverse public sources, ensuring full methodological transparency in the resulting score. This comprehensive structural assessment is computed from three weighted pillars: ESG Performance, Operational Resilience (OR), and Quality Performance (QP). The Environmental and Social components of ESG are evaluated using a contextual exposure vs. management model, which utilizes peer-based normalization to distinguish a supplier’s inherent risk exposure from its mitigation capabilities. A prototype application has been developed to implement the “scrape-then-verify” model across the pillars. This framework has been used to demonstrate the technical viability of generating trustworthy, multi-pillar risk profiles at scale, offering evidence-based data for resilient sourcing and targeted supplier improvement within the manufacturing ecosystem.

Impacts of natural hazards on supply chains can be devastating, especially given the increase in their frequency and intensity. For larger and geographically dispersed supply chains, it is critical to monitor supply chain nodes for natural hazard risk and take early actions to mitigate their impact. However, this requires an automated system that is capable of retrieving and analyzing dynamic information specific to these locations for rapid risk assessment. This paper proposes an automated risk identification and assessment system for natural hazards using Large Language Models (LLMs) and news data. First, critical risk features for natural hazard risk assessment are identified from the literature. Then, news articles about natural hazards that occurred in supply nodes are retrieved, and risk feature information is extracted by four LLMs (Llama3.1-8B, Gemma3-12B, DeepSeek-R1-14B, Phi4-14B) using three prompting strategies (zero-shot, few-shot, and chain-of-thought). A bicycle case study that involves a global supply chain network is applied to demonstrate the effectiveness of the proposed system. The performance of LLMs is evaluated for i) correctness by human evaluators, and ii) contextual similarity by using Bidirectional Encoder Representations from Transformers Score (BERTScore). Results show that Llama3.1-8B and Phi4-14B have strong potential for automated risk assessment, achieving higher scores in both human evaluation and BERTScore. However, DeepSeek-R1-14B resulted in the lowest performance among the tested LLMs. Furthermore, hazards impacting a wider region reduce LLM performance due to complex links between location and risk features.

Impacts of natural hazards on supply chains can be devastating, especially given the increase in their frequency and intensity. For larger and geographically dispersed supply chains, it is critical to monitor supply chain nodes for natural hazard risk and take early actions to mitigate their impact. However, this requires an automated system that is capable of retrieving and analyzing dynamic information specific to these locations for rapid risk assessment. This study proposes an automated risk identification and assessment system for natural hazards using large language models (LLMs) to analyze news data. First, critical risk features for natural hazard risk assessment are identified from the literature. Then, news articles about natural hazards that occurred in supply nodes are retrieved, and risk feature information is extracted by four LLMs (Llama3.1-8B, Gemma3-12B, DeepSeek-R1-14B, and Phi4-14B) using three prompting strategies (zero-shot, few-shot, and chain of thought). A bicycle case study that involves a global supply chain network is applied to demonstrate the effectiveness of the proposed system. The performance of the proposed system is evaluated for (i) correctness by human evaluators and (ii) contextual similarity by using Bidirectional Encoder Representations from Transformers Score (BERTScore). Results show, on one hand, that Llama3.1-8B and Phi4-14B have strong potential for automated risk assessment, achieving higher scores in both human evaluation and BERTScore. On the other hand, DeepSeek-R1-14B resulted in the lowest performance among the tested LLMs. Furthermore, hazards impacting a wider region reduce LLM performance due to complex links between locations and risk features.

Life cycle assessment (LCA) is increasingly applied to evaluate emerging biobased materials; however, its use is often constrained by fragmented, inconsistent, and incomplete life cycle inventory (LCI) data. Industrial hemp is representative of such materials, as historical regulatory barriers and heterogeneous study assumptions have resulted in sparse and poorly harmonized product and process data, limiting the reuse and comparability of existing LCAs. This study addresses these challenges by developing harmonized datasets and LCIs. A structured literature synthesis was conducted to systematically screen and extract quantitative parameters from published studies describing upstream and midstream stages of industrial hemp production. Data were compiled for pre-harvest operations, including fertilizer use (nitrogen, phosphorus, and potassium), seeding density, harvest yield, hemp type, and geographic context. Additional data cover emissions from fertilizers to air, soil, and water, as well as water use (irrigation), agricultural machinery, diesel use, and electricity use. Post-harvest data included decortication process energy use, extraction yields for fiber and hurd, and carbon storage potential. To address data sparsity and missing parameter combinations commonly encountered during LCI development, machine learning (ML) is demonstrated as a supporting data-augmentation and gap-filling approach, using hemp yield as an illustrative case. The datasets compiled for the upstream and midstream stages consolidate the key material and energy flows required to build screening-level LCIs for industrial hemp. Reported cultivation inputs and yields exhibited substantial variability across studies, with nitrogen application ranging from 37 to 300 kg/ha and biomass yields from 2,200 to 31,300 kg/ha. Decortication energy demand ranged from 1.21 to 9.63 MJ/kg of fiber, reflecting differences in agronomic practices and processing configurations. Extraction yields ranged from 20

This study examines the thinning behavior of AA3003-H14 aluminum alloy during single point incremental sheet forming (SPIF) through a combination of experimental trials and finite element analysis (FEA) using LS-DYNA. A full factorial experimental design was implemented to assess the effects of wall angle (45°, 55°, 65°) and step size (0.25 mm, 0.50 mm, 0.75 mm) on sheet thinning at various forming depths. Thickness measurements were analyzed using a two-way analysis of variance to determine the significance of process parameters and their interactions. Numerical simulations predicted thickness reduction, effective plastic strain, and von Mises stress distributions, with deviations from experimental results generally remaining below 10%. The findings indicate that wall angle has a dominant influence on thinning, while step size exhibits a moderate effect. The validated FEA model accurately captures localized deformation behavior, offering a predictive tool for optimizing SPIF parameters. This work enhances the understanding of AA3003 thinning mechanisms and supports process improvements for broader industrial adoption of SPIF.

Timely evaluation of risk factors is essential for building resilient supply chains and requires continuous monitoring and updating. This paper proposes a conceptual model for supply chain risk management that uses large language models (LLMs) to detect a disruption trigger event from news data, determines its impact, and develops a proper risk mitigation strategy. Even though the conceptual model can accommodate many kinds of supply chain risks, this study focuses on natural hazard risks affecting global supply chains. Because natural hazards are related to the locations of the suppliers, the model uses country level natural hazard risk scores to evaluate suppliers. Here, a country’s risk score determined by the Index for Risk Management (INFORM) is used as a baseline risk score for each country, which are then, updated by the disruption trigger event’s features captured by the LLM. Lastly, a supply chain optimization model is used to determine alternative suppliers to mitigate the impact of risk on disruption. The conceptual model is demonstrated through the case study of a bicycle company. Preliminary results indicate that the proposed model can be used to improve the resilience of supply chains against natural hazards.

With rising concerns over global warming, resource scarcity, and sustainable development, the industrial sector is under increasing pressure to reduce its environmental impacts, including those of underpinning manufacturing processes, e.g., additive manufacturing (AM). Nickel (Ni) is among the key critical minerals in the U.S.; thus impacts of producing of Ni-based components using AM processes has received a significant attention. This study aims to assesses the environmental impacts of producing Ni powder using life cycle assessment (LCA). Three production scenarios are examined: one using 100% virgin Ni, another incorporating externally recycled Ni (30% of input Ni), and the third using 100% recycled material sourced from internal revert scrap and local suppliers. SimaPro 9 is applied to capture life cycle inventory data and to conduct the impact assessments using the Cumulative Energy Demand (CED), TRACI 2.1, and ReCiPe 2016 methods. The findings of this study highlight the significant environmental benefits of utilizing recycled materials in Ni powder production, which can assist material procurement decisions. The trends in the impact assessment results for the three scenarios are similar; thus for simplicity, only GWP under the TRACI method is further investigated. The results show that the production phase of virgin Ni contributes to over 96% of cradle-to-gate GWP. In contrast, the two scenarios investigating circularity (Scenarios 2 and 3) demonstrated significant reductions in GWP, achieving 96.3% and 99.7% reduction, respectively, compared to Scenario 1. These results support the exploration of circularity and sustainable manufacturing practices, which are crucial in addressing the challenges of resource depletion and climate change.

Course and program developers must assess students' performance and comprehension at multiple levels to identify and implement necessary curriculum improvements in addition to establishing the ultimate utility of the program. Such evaluation efforts ensure that the curriculum remains aligned with industry standards, and effectively meets learning objectives, thereby ensuring that students develop direct, transferable skills relevant to the field. Thus, the objective of the current research is to identify the necessary steps required to construct quantitative learning assessment (QuLA) in an online classroom setting. To achieve this objective, a four-step QuLA method is developed herein. To demonstrate the application of the method, two courses (MFGE 241 and MFGE 341) are selected from the four mechatronics-focused manufacturing engineering courses developed for the online Undergraduate Certificate in Mechatronics for Manufacturing Engineering at Oregon State University (OSU) under a project entitled Modular Educational Certification for Advancing Training Online through Industry Collaboration (MECHATRONIC) funded by the U.S. National Science Foundation (NSF). The outcomes of this research will be helpful for other education developers to conduct a QuLA for their associated courses/programs and to evaluate student performance and understanding. The developed QuLA method can be used in any engineering area, and is meant to serve as a broad recommendation for developing and implementing assessments in STEM courses. (c) 2025 The Authors. Published by ELSEVIER Ltd. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of the scientific committee of the NAMRI/SME.

There is a growing interest in exploring the potential use of bio-based 3D printing filaments as sustainable substitutes for traditional polymer-based filaments. This study analyzes the environmental impacts of hemp materials (hemp straw, hemp hurd, and hemp fiber) when blended with pure PLA (Polylactic Acid) for 3D printing filaments. A comparative Life Cycle Assessment (LCA) was conducted using the ReCiPe 2016 method to evaluate midpoint and endpoint impact categories, along with two single-issue methods, Cumulative Energy Demand (CED) and IPCC Global Warming Potential (GWP100), to assess energy and carbon-related impacts of such filaments. The results indicate that hemp straw PLA filament has the lowest environmental impacts, followed by hemp hurd PLA, hemp fiber PLA, and pure PLA, with the last exhibiting the highest overall impacts in all endpoint categories (human health, ecosystems, and resources) using the ReCiPe 2016 method. Moreover, pure PLA demonstrates the highest global warming potential (GWP) across all filaments, while hemp fiber PLA displays higher impacts in energy consumption due to its higher density. Sensitivity analyses exploring the effects of varying the PLA blend ratio on environmental outcomes reveal that increased PLA content raises energy demand and carbon emissions across all filaments. Hemp-based PLA filaments, particularly hemp straw and hurd, hold promise as sustainable alternatives to pure PLA, especially for applications not requiring high-performance mechanical properties. Future research will focus on evaluating the use of treated biomass in filament production, specifically emphasizing hemp straw to enhance the sustainability performance of 3D printing filament materials.

With recent advancements in computer and network technologies, Additive Manufacturing (AM), together with other manufacturing processes, is subject to cyberattacks. In AM, feedstock materials in powder, fluid, or wire forms are deposited on the build plates layer by layer to build structures along the 3D digital models. One challenge with AM is protecting intellectual property due to the way design information is concealed within the object’s geometry. Furthermore, even with well-placed security measures, adversaries can exploit key parameters that influence part geometry during the printing process through side-channel emissions leakage. This leakage can be used to reconstruct parts through advanced analysis methods for signals from indirect sources, enabling attackers to reverse-engineer a part’s 3D geometry. The goal of this work is to assess the prevalence of research efforts focusing on reconstructing manufactured products using side-channels (auxiliary process-related) information, which will be achieved through a three-phase systematic literature review within both the Compendex and Scopus databases. The finally selected articles were analyzed to understand 1) the type of used side-channels with different manufacturing processes, 2) state-of-the-art techniques in reconstructing manufactured parts, and 3) existing research opportunities in this area. More specifically, the papers were analyzed from different perspectives, and several insights were obtained pertaining to: 1) the type of side-channels used, 2) targeted manufacturing processes, and 3) the implementation of reconstruction techniques. For instance, it was found that a variety of side-channel types have been used, with acoustics being the most common. In addition, it was noticed that most of the articles used Fused Deposition Modeling (FDM) desktop printers, perhaps due to their ability to rapidly and cost-effectively prototype complex, lightweight, and free-form 3D objects. However, one study did attempt to reconstruct an object using both additive and subtractive processes, suggesting that part reconstruction could be applied across various manufacturing processes and may not be limited to AM. Another insight is that reconstruction has utilized different machine learning and deep learning approaches to enhance the accuracy of reconstructed objects, including random forests and neural networks. Based on this literature review, several recommendations are made to encourage more research in this area. A recommended avenue is to consider manufacturing processes beyond AM, including subtractive processes. Additionally, reducing signal leakages should be pursued, especially as new advances in cyber-physical technologies emerge. Despite the useful insights obtained from this review, this work is not without limitations. The results of this analysis may not necessarily be generalizable; thus, examining more publications would be beneficial. Hence, future work will apply forward and backward tracking techniques to uncover additional studies.

Battery lifetime and reliability depend on accurate state-of-health (SOH) estimation, while complex degradation mechanisms and varying operating conditions strengthen this challenge. This study presents two physics-informed neural network (PINN) configurations, PINN-parallel and PINN-series, designed to improve SOH prediction by combining an equivalent circuit model (ECM) with a long short-term memory (LSTM) network. PINN-parallel process inputs data through parallel ECM and LSTM modules and combines their outputs for SOH estimation. On the other hand, the PINN-series uses a sequential approach that feeds ECM-derived parameters into the LSTM network to supplement temporal data analysis with physics information. Both models utilize easily accessible voltage, current, and temperature data that match realistic battery monitoring constraints. Experimental evaluations show that the PINN-series outperforms the PINN-parallel and the baseline LSTM model in accuracy and robustness. It also adapts well to different input conditions. This demonstrates that the simulated battery dynamic states from ECM increase the LSTM's ability to capture degradation patterns and improve the model's ability to explain complex battery behavior. However, a trade-off between the robustness and training efficiency of PINNs is identified. The research outcomes show the potential of PINN models (particularly the PINN-series) in advancing battery management systems, although they require considerable computational resources.

The manufacture of biobased products, such as those derived from industrial hemp, has been heralded with great potential for improving industrial sustainability performance. However, due to their renewable nature, complexities in their value chains, and data variability, timely, accurate, and reliable sustainability assessments of biobased products have been impeded. These barriers have been especially problematic for the nascent US hemp industry. This study uses a narrative literature review and sustainability assessment methodologies to address the gap in the sustainability assessment of hemp-based product life cycles by leveraging Industry 4.0 technologies – termed Industrial Hemp 4.0 (IH4.0). IH4.0 encompasses the entire hemp life cycle, involving smart farming, smart manufacturing, smart supply chains, and smart products, with the potential to integrate circularity. An application of this approach is discussed for decortication, an industrial process for separating hemp hurd and fiber for use in textiles, bio-composites, packaging, and construction materials. Conceptually, IH4.0 ensures effective and transparent sustainability assessment, communication, and collaboration in defining, developing, and implementing viable solutions for the emerging hemp industry. Integrating data and information for industrial hemp across the product life cycle can assist designers, manufacturing engineers, and other stakeholders in creating more sustainable products and making more environmentally conscious decisions.

This research is motivated by the increasing demand for bio-based materials, the recent growth of the U.S. hemp industry, and the broader trend of using sustainable filaments in additive manufacturing. This study presents a comparative life cycle assessment (LCA) of untreated hemp straw and kraft lignin as fillers for polylactic acid (PLA) 3D-printed tensile specimens to evaluate their environmental impacts. Both materials, being low-cost renewable bio-based fillers, can enhance elongation of PLA while reducing the environmental footprint of 3D-printed components. The environmental impacts of hemp straw-PLA and kraft lignin-PLA were assessed using several life cycle impact assessment (LCIA) methods, including ReCiPe 2016 Endpoint (H), Cumulative Energy Demand (CED), and IPCC GWP100. Hemp straw showed lower environmental impacts than kraft lignin across most categories, making it a more favorable option for eco-conscious prosumers. The ReCiPe 2016 results indicated that major impact categories for kraft lignin-PLA were fine particulate matter formation, global warming potential, and human toxicity, with filament production being the major contributor. For hemp straw-PLA, hemp straw pre-processing was the major contributor. The CED method revealed that nonrenewable fossil resources had the highest impact on both materials. IPCC GWP100 results aligned with CED, showing higher greenhouse gas emissions for kraft lignin-PLA, mainly due to fossil fuel use. Sensitivity analysis of transportation distances showed no significant differences in impact results, while alternative LCIA methods (TRACI and IMPACT World+) confirmed the consistency of the findings. To build upon this study, future work will explore the environmental performance of treated hemp materials as alternative fillers for 3D-printed components.

The transition to biobased materials from reliance on fossil resources is pivotal in establishing a sustainable biobased economy. However, sustainability assessment methods and tools face limitations in adequately evaluating biobased products due to their renewable nature and complex value chains. This study uses a narrative review to explore the potential of Industry 4.0 (I4.0) technologies to address these limitations in the sustainability assessment of biobased product manufacturing. Additionally, published articles on Industry 5.0 (I5.0) technologies were reviewed to highlight the integration of human creativity with intelligent technologies and opportunities for tailoring sustainability assessment methods and tools to address the unique challenges of biobased product manufacturing. By harnessing the capabilities of I4.0, real-time data exchange and predictive modeling can be used to create and assess the sustainability profile of biobased products. Establishing transparent, decentralized systems powered by Internet of Things (IoT) technologies will be crucial to enhancing sustainability assessment. Robust frameworks can be developed leveraging cognitive computing capabilities and collaborative approaches of I5.0, rendering sustainability assessment methods and tools more adaptive, accurate, reliable, and holistic in supporting the sustainable growth of the biobased industry, ensuring its economic competitiveness against alternative fossil-based product industries.

Accurate prediction of repair duration is an important challenge in product maintenance due to its implications for resource allocation, customer satisfaction, and operational performance. This study aims to develop a deep learning framework to help fleet repair shops accurately categorize repair time given product historical data. The study uses an automobile repair and maintenance dataset and creates an end-to-end predictive framework by employing a multi-head attention network designed for tabular data. The developed framework combines categorical information, transformed through embeddings and attention mechanisms, with numerical historical data to facilitate integration and learning from diverse data features. A weighted loss function is introduced to overcome class imbalance issues in large datasets. Moreover, an online learning strategy is used for continuous incremental model updates to maintain predictive accuracy in evolving operational environments. Our empirical findings demonstrate that the multi-head attention mechanism extracts meaningful interactions between vehicle identifiers and repair types compared to a feed-forward neural network and a random forest model. Also, combining historical maintenance data with an online learning strategy facilitates real-time adjustments to changing patterns and increases the model's predictive performance on new data. The model is tested on real-world repair data spanning 2013 to 2020 and achieves an accuracy of 78