Machine fault diagnosis (FD) is a critical task for predictive maintenance, enabling early fault detection and preventing unexpected failures. Despite its importance, existing FD models are operation-specific with limited generalization across diverse datasets. Foundation models (FM) have demonstrated remarkable potential in both visual and language domains, achieving impressive generalization capabilities even with minimal data through few-shot or zero-shot learning. However, translating these advances to FD presents unique hurdles. Unlike the large-scale, cohesive datasets available for images and text, FD datasets are typically smaller and more heterogeneous, with significant variations in sampling frequencies and the number of channels across different systems and applications. This heterogeneity complicates the design of a universal architecture capable of effectively processing such diverse data while maintaining robust feature extraction and learning capabilities. In this paper, we introduce UniFault, a foundation model for fault diagnosis that systematically addresses these issues. Specifically, the model incorporates a comprehensive data harmonization pipeline featuring two key innovations. First, a unification scheme transforms multivariate inputs into standardized univariate sequences. Second, a novel cross-domain temporal fusion strategy mitigates distribution shifts and enriches sample diversity and count, improving the model generalization across varying conditions. UniFault is pretrained on over 6.9 million samples spanning diverse FD datasets, enabling superior few-shot performance. Extensive experiments on real-world FD datasets demonstrate that UniFault achieves state-of-the-art performance, setting a new benchmark for fault diagnosis models and paving the way for more scalable and robust predictive maintenance solutions.

Product-type changes in connector manufacturing (e.g., Type-1 to Type-2) often cause a severe domain shift that degrades pixel-level inspection models trained on previous products due to differences in conductor count, spacing, scale, and appearance, even though the underlying conductor structures remain partially shared. Collecting dense segmentation labels for each new product generation is expensive, yet unsupervised domain adaptation (UDA) alone may not meet the high accuracy required for deployment. Therefore, in this paper, we propose a two-stage, label-efficient semi-supervised domain adaptation framework for conductor segmentation under product-type shift. In Stage-1, we initialize the target model from a source-trained segmentation network and perform UDA to reduce the feature gap using unlabeled target data. In Stage-2, we further adapt with a small, labeled target subset and the remaining unlabeled target samples via a teacher–student scheme with confidence-aware pseudo labels and conductor-aware structural priors that suppress noisy regions. Experiments on real connector inspection data show that our method achieves 92.7

Accurate gearbox fault diagnosis under varying operational speeds is critical for industrial predictive maintenance. A significant challenge is domain shift, where models trained under one condition fail to generalize to another, especially when only healthy data from the target domain is available for training. This study proposes a novel domain adaptation framework, CDANet, that directly leverages raw sensor data to perform multi-class fault classification without manual feature engineering. The model combines a lightweight CNNbased temporal feature extractor with a frozen DistilBERT encoder to capture transferable, domain-invariant representations, combined with a maximum mean discrepancy loss to align the feature distributions between the source and target domains using only healthy samples. Experimental results demonstrate that our proposed model significantly outperforms conventional deep learning approaches, achieving high classification accuracy across six domain adaptation tasks. This work validates the effectiveness of applying pre-trained models in domain adaptation for gearbox fault diagnosis under real-world domain shift constraints.

Machine learning (ML) has driven major advances across engineering domains, from early data-driven approaches to deep learning and, more recently, foundation models. Despite rapidly increasing model capabilities, the practical impact of ML in engineering systems remains limited. We argue that this gap arises from a fundamental misalignment between prevailing ML paradigms and the realities of engineering environments. Modern ML methods, largely developed for data-rich and statistically stable domains, are fundamentally ill-suited for engineering systems characterized by sparse, noisy, and context-dependent data, uncertain ground truth, non-stationary dynamics, strict deployment constraints, and strong physical and domain knowledge requirements. To address this mismatch, we first identify four core challenges: data and benchmark limitations, distribution shift and generalization, resource and deployment constraints, and insufficient integration of domain knowledge. Then we propose a corresponding four-shift framework: (i) from data quantity to engineered data foundations, (ii) from static learning to adaptive and robust learning, (iii) from model scaling to resource-efficient and deployment-aware intelligence, and (iv) from general-purpose models to domain-specific, knowledge-informed foundation models. Building on this framework, we outline four research directions toward Engineering Intelligence: data foundation intelligence, reliable and generalizable intelligence, sustainable and deployable intelligence, and knowledge-centric intelligence. Finally, we introduce a data maturity-deployment scale framework, arguing that engineering AI should be understood not as a search for a single best model, but as an evolving intelligence strategy that co-adapts with system maturity, operational context, and deployment scale.

Developing an effective prognostics and health management (PHM) prediction pipeline requires iterative decisions on data structure, degradation behavior, feature representation, evaluation metrics, and failure cases. This paper reports a human-supervised Agentic AI workflow for PHM pipeline development through a case study in the PHME 2026 Data Challenge, which focused on remaining useful life (RUL) prediction for a subway door servomotor system. In the proposed workflow, a human researcher supervised the overall exploration, while AI-assisted roles supported planning, experiment specification, implementation, review, case-wise error analysis, and knowledge accumulation. Through this process, we developed a case-aware RUL prediction pipeline combining health-indicator-based estimation, operating-condition-aware correction, training-data-based calibration, temporal consistency checks, and submission-format verification. The final submission achieved an official challenge score of 0.9983. This case study suggests that Agentic AI can help structure, accelerate, and document iterative machine learning research, while human supervision remains essential for selecting directions, managing risks, and interpreting results.

The evolution of artificial intelligence (AI) and machine learning (ML) is reshaping smart manufacturing by providing new capabilities for efficiency, adaptability, and autonomy across industrial value chains. However, the deployment of AI and ML in industrial settings still faces critical challenges, including the complexity of industrial big data, effective data management, integration with heterogeneous sensing and control systems, and the demand for trustworthy, explainable, and reliable operation in high-stakes industrial environments. In this roadmap, we present a comprehensive perspective on the foundations, applications, and emerging directions of AI and ML in smart manufacturing. It is structured in three parts. The first highlights the foundations and trends that frame the evolution of AI in smart manufacturing. The second focuses on key topics where AI is already enabling advances, including industrial big data analytics, advanced sensing and perception, autonomous systems, additive and laser-based manufacturing, digital twins, robotics, supply chain and logistics optimization, and sustainable manufacturing. The third section explores non-traditional ML approaches that are opening new frontiers, such as physics-informed AI, generative AI, semantic AI, advanced digital twins, explainable AI, RAMS, data-centric metrology, LLMs, and foundation models for highly connected and complex manufacturing systems. By identifying both opportunities and remaining barriers across these areas, this roadmap outlines the advances needed in methods, integration strategies, and industrial adoption. We hope this roadmap will serve as a guide for researchers, engineers, and practitioners to accelerate innovation, align academic and industrial priorities, and ensure that AI-driven smart manufacturing delivers reliable, sustainable, and scalable impact for the future of manufacturing ecosystems.

Cross-domain fault diagnosis has become increasingly important for prognostics and health management (PHM) because machine learning models often experience performance degradation when deployed under different operating conditions. Existing domain adaptation methods typically assume that the target domain has all classes available during training, an assumption that is often impractical in industrial applications where only limited healthy target-domain data can be collected from newly deployed machines. To address this challenge, this paper proposes PLM4TSDA, a framework for healthy-only target domain adaptation that integrates effective time-series token construction, lightweight adaptation of pre-trained language models, and simple domain adaptation objectives. Experiments on an industrial Mitsubishi Electric gearbox dataset and the Jiangnan University bearing dataset demonstrate that PLM4TSDA outperforms representative deep learning models trained from scratch, achieving performance improvements of 2%–16% across most transfer tasks. These results demonstrate the effectiveness of pretrained transformer representations for improving cross-domain fault diagnosis under realistic target-scarce industrial scenarios.

Large language model-based question answering has the potential to improve access to technical knowledge, but hallucinated responses remain a severe reliability risk when answers are generated without sufficient evidential support. In reliability-critical domains, such as prognostics and health management (PHM), this risk is particularly concerning, as unsupported information can lead to incorrect maintenance decisions or unsafe actions. However, existing retrieval-augmented generation pipelines often lack explicit control over evidence sufficiency and could produce wrong answers, thereby raising concerns about artificial intelligence reliability issues. In this article, we propose RC-RAG, a reliability-centered retrieval-augmented generation (RAG) framework for PHM question answering that explicitly incorporates evidence evaluation and verification into the retrieval and generation pipeline. RC-RAG augments PHM literature with document-level meta-knowledge, performs hierarchical document-content retrieval, and applies relevance and evidence assessment with verification-aware answer generation to ensure that answers are produced only when sufficient and consistent evidence is available. To evaluate the framework, we construct a curated PHM data challenge question-answering benchmark derived from a series of PHM conferences and data challenge competitions, including those organized by the IEEE Reliability Society and the PHM research community. Experimental results show that RC-RAG achieves significantly higher accuracy and precision compared to other state-of-the-art methods.

In industrial Prognostics and Health Management (PHM), the scarcity of sufficiently large, high-quality datasets remains a persistent challenge, which limits the practical deployment of machine learning-based approaches. Recent efforts to address this include few-shot learning, domain adaptation, and other data-efficient learning paradigms. However, while the use of pre-trained models has shown great promise in other fields such as natural language processing and computer vision, its application in PHM remains relatively underexplored. Although initial studies have begun to introduce foundation-style models for specific components—such as recent efforts on bearing health diagnostics using transformer-based architectures—their development is still in an early stage compared to the maturity and versatility of large language models (LLMs) in the NLP domain. LLMs continue to advance rapidly, offering generalization capabilities that could be highly beneficial in data-constrained PHM settings. While some preliminary research has explored the use of LLMs as intelligent agents for decision support in PHM workflows, their application as direct learners for time-series sensor data remains rare. In this work, we propose a novel framework that adapts pre-trained LLMs for time-series-based PHM tasks. Our approach involves mapping temporal sensor signals to a tokenized format compatible with transformer-based language models, enabling the application of LLMs as generic sequence learners. Building on recent pioneering concepts such as multimodal LLM-based health management systems and prompt-driven signal encoding, our framework is benchmarked on publicly available industrial datasets under low-data conditions. The results demonstrate that our LLM-based approach not only maintains robust performance in scenarios with limited labeled data but also outperforms traditional models in fault classification accuracy. This study contributes a new perspective to the PHM community by highlighting the untapped potential of LLMs as general-purpose, pre-trained models in industrial health monitoring applications. Our findings suggest that incorporating LLMs into PHM workflows can be a powerful and forward-looking strategy to overcome data scarcity and improve adaptability across diverse operational domains.

Over the past ten years, the application of artificial intelligence (AI) and machine learning (ML) in engineering domains has gained significant popularity, showcasing their potential in data-driven contexts. However, the complexity and diversity of engineering problems often require the development of domain-specific AI approaches, which are frequently hindered by a lack of systematic methodologies, scalability, and robustness during the development process. To address this gap, this paper introduces the "ABCDE" as the key elements of Engineering AI and proposes a unified, systematic engineering AI ecosystem framework, including eight essential layers, along with attributes, goals, and applications, to guide the development and deployment of AI solutions for specific engineering needs. Additionally, key challenges are examined, and eight future research directions are highlighted. By providing a comprehensive perspective, this paper aims to advance the strategic implementation of AI, fostering the development of next-generation engineering AI solutions.

In semiconductor manufacturing, discrepancies between the golden and target machines can lead to inconsistencies in product quality and reduced yield, highlighting the need for more effective calibration methods. Traditionally, machine calibration relies on using basic statistical features during the feature extraction step, which limits the ability to fully capture the complex characteristics of machine dynamics. This study aims to expand the spectrum of feature types that contribute to enhancing machine calibration performance. Experiments using a Hitachi Plasma Etching System demonstrate that the new method significantly reduces overall calibration errors, representing more than a twofold improvement over the traditional approach. This enhanced calibration accuracy ensures greater consistency between machines and manufacturing processes, thereby enhancing operational efficiency and minimizing downtime.

Accurate gearbox fault diagnosis across different operating conditions plays an important role in prognostics and health management. In real industrial scenarios, a common challenge arises when the source domain contains multiple fault classes, while the target domain includes only healthy samples during training. To address this issue, this study proposes a unified industrial fault diagnosis framework designed to handle the partial domain adaptation problem. Specifically, the overall framework involves: a unified data processing pipeline, a robust deep learning architecture for accurate fault classification, and integration of maximum mean discrepancy loss to align feature distributions between source and target domains. Experimental results demonstrate that our proposed partial domain adaptation-based deep learning model significantly outperforms benchmark models, achieving accuracy improvements exceeding 20% across multiple domain adaptation tasks. This study provides a practical solution for intelligent gearbox diagnosis under domain shift constraints.

The rise of data-rich manufacturing environments has created demand for artificial intelligence (AI) systems capable of autonomous, adaptive, and goal-oriented operations. Traditional AI methods, being largely task-specific, often lack the flexibility to perform effectively in dynamic, complex industrial settings. Recent advances in large language models (LLMs) have led to the emergence of Agentic AI, which extends AI capabilities through advanced reasoning, planning, tool integration, and multi-agent collaboration. While Agentic AI has been explored in domains such as computer science, healthcare, education, and finance, its adoption in manufacturing remains limited. This paper defines Agentic AI in the manufacturing context, differentiates it from traditional AI agents, and presents a novel framework designed for smart manufacturing. The proposed framework integrates multiple LLM-based agents, a unified Data-Model-Knowledge (DMK) lake, and human expertise to enable advanced perception, reasoning, planning, orchestration, evaluation, optimization, and iterative improvement. A case study of a retrieval-augmented generation (RAG)-based LLM QA system demonstrates the framework's feasibility. Key technical challenges are also discussed. The work aims to provide strategic guidance for the development and deployment of efficient, trustworthy Agentic AI systems in smart manufacturing. (c) 2025 Society of Manufacturing Engineers (SME). Published by Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Prognostics and Health Management (PHM) is vital for complex engineering systems, yet its data-driven solutions are often hampered by the "small data problem"—a scarcity of labeled fault data in industrial settings. This limitation restricts the training and generalization of machine learning models and is compounded by varying operational conditions that reduce the relevance of historical data and pre-trained models. This research introduces a research framework to tackle these small data challenges in PHM. The primary objective is to develop a robust and adaptable PHM methodology by enhancing and synergistically integrating similarity-based Few-Shot Learning (FSL) with large-scale pre-trained time-series models. The research will focus on two main thrusts. First, it aims to improve the generalization capabilities of FSL frameworks by addressing limitations such as noise robustness, domain shift adaptability, and generalization to novel faults across diverse PHM domains. This involves developing noise-robust feature extraction, integrating domain adaptation techniques, and exploring expressive similarity metrics. Second, the study will investigate the effective adaptation of state-of-the-art pre-trained time-series models (e.g., TimesNet) for PHM tasks under data scarcity, focusing on efficient fine-tuning and synergistic integration with the enhanced FSL approaches. The author's prior success in a PHM data challenge using a similarity-based method for spacecraft systems provides preliminary validation. This research is expected to deliver an enhanced PHM framework for high-accuracy diagnostics with limited data, contributing generalized FSL models, systematic methods for leveraging pre-trained models in PHM, and advancing the practical deployment of intelligent PHM solutions.

In the era of Industry 4.0, artificial intelligence (AI) is assuming an increasingly pivotal role within industrial systems. Despite the recent trend within various industries to adopt AI, the actual adoption of AI is not as developed as perceived. A significant factor contributing to this lag is the data issues in AI implementation. How to address these data issues stands as a significant concern confronting both industry and academia. To address data issues, the first step involves mapping out these issues. Therefore, this study conducts a meta-review to explore data issues and methods within the implementation of industrial AI. Seventy-two data issues are identified and categorized into various stages of the data lifecycle, including data source and collection, data access and storage, data integration and interoperation, data pre-processing, data processing, data security and privacy, and AI technology adoption. Subsequently, the study analyzes the data requirements of various AI algorithms. Building on the aforementioned analyses, it proposes a data management framework, addressing how data issues can be systematically resolved at every stage of the data lifecycle. Finally, the study highlights future research directions. In doing so, this study enriches the existing body of knowledge and provides guidelines for professionals navigating the complex landscape of achieving data usability and usefulness in industrial AI.

Recent advancements in industrial artificial intelligence (AI) are reshaping the industry by driving smarter manufacturing, predictive maintenance, and intelligent decision-making. However, existing approaches often focus primarily on algorithms and models while overlooking the importance of systematically integrating domain knowledge, data, and models to develop more comprehensive and effective AI solutions. Therefore, the effective development and deployment of industrial AI require a more comprehensive and systematic approach. To address this gap, this paper reviews previous research, rethinks the role of industrial AI, and proposes a unified industrial AI foundation framework comprising three core modules: the knowledge module, data module, and model module. These modules help to extend and enhance the industrial AI methodology platform, supporting various industrial applications. In addition, a case study on rotating machinery diagnosis is presented to demonstrate the effectiveness of the proposed framework, and several future directions are highlighted for the development of the industrial AI foundation framework.

Planetary gearboxes are critical components in various industrial applications, but their complex structure makes fault diagnosis challenging. This study introduces an innovative approach for fault diagnosis in planetary gearboxes using a hybrid CNN-Transformer model with skip connections. Our method combines the strengths of Convolutional Neural Networks (CNNs) for local feature extraction and Transformer encoders for global sequence modeling, while utilizing skip connections to preserve hierarchical information. The dataset used in this study is the ICPHM data challenge 2023 public dataset, which contains five classes (i.e., normal, surface wear, crack, chipped, tooth missing) and provides vibration signals from the x, y, and z directions on 200 sample long frames (i.e., short time series signals). We trained our neural network on two operational conditions of the planetary gearboxes. Experimental results demonstrate that our approach effectively diagnoses faults with high accuracy and reliability.