Enabling Dependability in Smart Manufacturing: RAMS and AI/ML Integration

Jing (Janet) Lin¹, Liangwei Zhang²

1 Department of Civil, Environmental and Natural Resources Engineering, Luleå University of Technology, Luleå, Sweden 2 Department of Industrial Engineering, Dongguan University of Technology, Dongguan, China

E-mail: janet.lin@ltu.se

Status

Reliability, Availability, Maintainability, and Safety (RAMS) have long provided the foundation for asset design, maintenance, and operational optimization in manufacturing. Traditional RAMS approaches—often supported by Reliability-Centered Maintenance (RCM), Condition-Based Maintenance (CBM), and

Prognostics and Health Management (PHM)—have primarily addressed physical degradation and failure patterns [1]. These methods focus on maximizing uptime and minimizing risk through structured maintenance planning and statistical analysis. However, the nature of manufacturing systems is changing. The emergence of Cyber-Physical Systems (CPS), the Industrial Internet of Things (IIoT), and embedded Artificial Intelligence (AI) is transforming industrial environments into intelligent, interconnected ecosystems [2]. In this context, assets are no longer standalone mechanical components; they are dynamic, software-integrated entities that operate in real time, interact with users and other machines, and continuously adapt to contextual and environmental changes. To meet the demands of this new landscape, we introduce Dependability-Centered Asset Management (DCAM)—a forward-looking framework that extends and evolves the RAMS paradigm [3]. DCAM integrates traditional reliability engineering with system-level dependability science and AI assurance methods to address modern challenges such as digital traceability, cyber-physical resilience, and lifecycle adaptability. DCAM promotes a holistic, lifecycle-oriented approach to dependability. It embeds reliability thinking from the earliest stages of design through to operation, evolution, and renewal. It draws upon digital technologies—including machine learning, digital twins, and edge/cloud architectures—to support predictive diagnostics, adaptive maintenance, and context-aware decision-making. By shifting focus from static reliability metrics such as Mean Time Between Failures (MTBF) to dynamic indicators of system resilience and AI model trustworthiness, DCAM aligns technical performance with broader goals of transparency, sustainability, and operational integrity. It enables manufacturers to respond not only to mechanical failure, but also to the risks and uncertainties introduced by AI-driven automation and distributed intelligence. As manufacturing systems become increasingly complex and autonomous, RAMS must evolve accordingly. DCAM offers a bridge between legacy reliability principles and future-ready dependability strategies— uniting physical, digital, and organizational dimensions into a unified framework for smart manufacturing. While RAMS principles provide the foundation, Table 1 highlights how asset management paradigms have evolved—from reliability-centered and software-driven approaches toward the integrated, AI-enabled perspective embodied in DCAM, which addresses the complexity of cyber-physical systems and lifecycle sustainability in smart manufacturing.

Current and future challenges

The integration of Artificial Intelligence (AI) and Machine Learning (ML) into manufacturing systems has catalyzed a shift from reactive maintenance toward predictive diagnostics and autonomous decision-making [4]. While this transition brings substantial benefits, it also introduces new complexities that challenge traditional approaches to Reliability, Availability, Maintainability, and Safety (RAMS) [5]. A key challenge lies in the fragmentation between traditional reliability engineering and broader system-level dependability. Established RAMS tools such as Failure Modes and Effects Analysis (FMEA) and Reliability-Centered Maintenance (RCM) remain largely hardware-focused, often overlooking the cyber, digital, and contextual elements that now define modern industrial systems. In contrast, dependability frameworks from the software domain emphasize attributes such as fault tolerance, robustness, and integrity—but frequently lack integration with physical degradation models or lifecycle asset management [6]. This disconnect hampers the development of unified strategies for managing the hybrid nature of AI-enabled manufacturing infrastructures. A second challenge stems from the lack of lifecycle assurance and traceability in AI/ML-driven diagnostics and decision systems. As AI models are increasingly embedded into operational processes, their outputs must be explainable, auditable, and resilient to real-world uncertainty. Yet, current practice often lacks mechanisms to assess AI reliability across different deployment scenarios, data distributions, or operational contexts [7]. This creates a twofold concern: AI must contribute to system dependability, while its own

behavior must also be dependable. Addressing this requires continuous validation pipelines, runtime monitoring, fallback mechanisms, and assurance frameworks tailored for AI models operating in safety-critical environments. Third, the growing interdependence of physical and digital components introduces new vulnerabilities. Failures in edge computing devices, corrupted sensor data, or loss of network connectivity can cascade through systems and undermine availability and safety at scale [8]. These cyber-physical risks call for diagnostic models and maintenance strategies that are context-aware, adaptive, and capable of responding dynamically—capabilities that traditional RAMS methods are not well equipped to provide. Organizational and human factors also present significant barriers. Adopting AI-driven RAMS requires not just technology but transformation—including workforce upskilling, changes in operational culture, and trust in data-centric decision-making [9]. Resistance to automation or lack of interdisciplinary collaboration can delay or derail the transition, especially in sectors where legacy systems and practices remain dominant. Finally, sustainability goals introduce a transformative pressure on RAMS thinking. Today’s manufacturing systems must be evaluated not only on technical and economic performance but also on environmental impact, resource efficiency, and long-term societal value. This expands the role of RAMS from failure avoidance to lifecycle stewardship—requiring methods that can integrate environmental and circularity metrics alongside traditional reliability indicators [10]. Together, these challenges signal the need for a new generation of dependability frameworks—ones that unify physical, digital, and organizational dimensions. The Dependability-Centered Asset Management (DCAM) approach responds to this need, offering a holistic foundation that embeds adaptability, explainability, and sustainability at the heart of next-generation manufacturing dependability.

Advances in science and technology to meet challenges

Overcoming the limitations of traditional RAMS frameworks in the era of smart manufacturing requires new scientific and technological approaches—ones capable of addressing the complexity of cyber-physical systems, autonomous operations, and data-driven decision environments [11]. The Dependability-Centered Asset Management (DCAM) framework offers a pathway forward by unifying lifecycle thinking, artificial intelligence (AI), and systems-level resilience into a cohesive strategy for modern manufacturing.

Figure 1. Key pillars of DCAM

Figure 1 illustrates the five foundational pillars of the DCAM framework, which collectively support AI-

driven dependability across the asset lifecycle. A foundational advancement lies in the application of non-traditional AI and machine learning (ML) methods for predictive diagnostics, anomaly detection, and remaining useful life (RUL) estimation. These technologies enable continuous asset monitoring, early detection of performance degradation, and data-driven optimization of maintenance schedules. Advanced learning techniques—such as reinforcement learning and federated learning—facilitate localized adaptation and decentralized intelligence across distributed factory systems, particularly where real-time responsiveness and data privacy are critical [12]. A second technological pillar is the increasing use of digital twins—virtual representations of physical assets and processes. Digital twins enable simulation of failure modes, evaluation of control strategies, and proactive assessment of system resilience under various operating conditions [13]. When synchronized with real-time sensor data, they support predictive analytics, scenario-based intervention planning, and long-term lifecycle optimization. Another key advancement is the convergence of RAMS, Prognostics and Health Management (PHM), and Aging Management. Each of these disciplines offers unique contributions across different stages of the asset lifecycle. RAMS provides statistical indicators of population-level reliability (e.g., MTBF), PHM focuses on real-time monitoring and individualized RUL estimation, and aging management ensures sustainable operation through end-of-life decision-making. AI/ML technologies play a central role in integrating these domains by enabling dynamic risk modelling, context-aware scheduling, and asset-specific decision support [14]. This convergence is further supported by emerging AI paradigms such as hybrid AI (e.g., physics-informed ML) and uncertainty-aware models, which enhance both interpretability and predictive robustness [15]. These approaches are closely aligned with advances in trustworthy AI and digital twin ecosystems, reinforcing the growing need for coordinated RAMS and AI assurance strategies at the system level. To enable safe, transparent, and explainable AI in manufacturing contexts, progress is being made in AI assurance. Techniques such as model verification, runtime monitoring, and explainability frameworks are gaining traction, particularly in safety-critical applications [16]. Additional tools—including fault injection, adversarial robustness testing, and uncertainty quantification—are increasingly used to validate ML models under operational stress and shifting data distributions. The advancement of edge and cloud computing infrastructures has also laid the foundation for scalable, distributed intelligence. Edge devices provide low-latency monitoring and localized decision-making, while cloud platforms enable fleet-level analysis, benchmarking, and optimization [17]. Together, they establish a flexible architecture for responsive and comprehensive RAMS decision support. Emerging capabilities in context-aware and adaptive decision-making offer another leap forward. These approaches allow maintenance and safety strategies to be tailored based on real-time contextual variables— such as asset criticality, usage intensity, environmental conditions, and cybersecurity posture. The result is more flexible, risk-informed asset management across varying operational scenarios. Lastly, the integration of AI with sustainability analytics introduces a multi-dimensional perspective on dependability. Future systems will increasingly depend on composite performance metrics that incorporate environmental and societal factors—such as energy consumption, carbon footprint, and material circularity—alongside traditional reliability indicators [18]. Together, these advances signal a transformative shift in how dependability is conceived and managed in smart manufacturing. When embedded within the DCAM framework, they enable a move away from static, failure-avoidance paradigms toward adaptive, lifecycle-oriented, and resilience-driven strategies. Yet, while these technologies greatly enhance the capabilities of modern RAMS, they also introduce a parallel imperative: ensuring the dependability of AI itself. As AI becomes embedded in decision-making systems, its behavior must remain trustworthy, robust, and explainable under real-world conditions. AI models must be continuously validated, monitored for data drift and degradation, and supported by fallback

strategies to ensure operational reliability. In this dual role—as both enabler and subject of reliability—AI demands rigorous lifecycle assurance. Its trustworthiness is essential not only for technical performance but also for safety, compliance, and user confidence in future manufacturing systems.

Concluding remarks

Smart manufacturing is entering a new phase—one defined by the convergence of physical, digital, and cognitive systems. While RAMS principles continue to serve as the foundation for reliable operations, they must evolve to address the complexity of modern industrial environments shaped by AI-driven automation, cyber-physical interconnectivity, and sustainability imperatives. Traditional reliability tools alone are no longer sufficient to ensure trust, adaptability, and long-term value in these intelligent, dynamic ecosystems. The Dependability-Centered Asset Management (DCAM) framework represents a timely and necessary evolution of RAMS thinking. By embedding lifecycle awareness, system-level dependability, and AI assurance into asset management strategies, DCAM provides a holistic and future-oriented approach for managing manufacturing systems that are increasingly autonomous, data-driven, and software-defined. This chapter has outlined the limitations of existing RAMS practices in the face of emerging technological and organizational challenges. It has also highlighted scientific and technological advances—ranging from hybrid AI models and digital twins to edge-cloud intelligence and context-aware maintenance—that enable a new generation of dependability solutions. At the center of these advances lies AI, which plays a dual role: both as a powerful enabler of predictive and adaptive capabilities, and as a source of new reliability and safety concerns. Managing this duality requires robust mechanisms for AI traceability, contextual awareness, and human-centric integration. As manufacturing systems continue to scale in complexity and autonomy, DCAM offers a strategic framework to ensure that smart factories are not only productive and efficient, but also dependable, transparent, and sustainable across their entire lifecycle. It bridges the gap between traditional reliability engineering and the evolving needs of AI-integrated industrial systems—enabling manufacturers to design for resilience, operate with confidence, and innovate with responsibility. Ultimately, DCAM provides a practical and adaptable roadmap for the future of manufacturing dependability—grounded in engineering rigor, enriched by AI, and guided by the principles of lifecycle stewardship and system trustworthiness.

Acknowledgements

This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant 72471060

References

[1] R. Wang, J. Xu, W. Zhang, J. Gao, Y. Li, and F. Chen, Reliability analysis of complex electromechanical systems: State of the art, challenges, and prospects, vol. 38, no. 7. 2022. [2] F. Qiu et al., “A Review on Integrating IoT, IIoT, and Industry 4.0: A Pathway to Smart Manufacturing and Digital Transformation,” IET Inf. Secur., vol. 2025, no. 1, p. 9275962, 2025. [3] J. Lin, “Dependability-Centered Asset Management(DCAM): Toward Trustworthy and Sustainable Systems in the CPS Era,” IEEE Reliab. Mag., vol. online, pp. 1–7, 2025. [4] R. Siraskar, S. Kumar, S. Patil, A. Bongale, and K. Kotecha, Reinforcement learning for predictive maintenance: a systematic technical review, vol. 56, no. 11. Springer Netherlands, 2023. [5] H. A. Fernandez-Bobadilla and U. Martin, “Modern Tendencies in Vehicle-Based Condition Monitoring of the Railway Track,” IEEE Trans. Instrum. Meas., vol. 72, pp. 1–44, 2023. [6] S. M. Hoseyni, F. Di Maio, and E. Zio, “Condition-based probabilistic safety assessment for maintenance decision making regarding a nuclear power plant steam generator undergoing multiple degradation mechanisms,” Reliab. Eng. Syst. Saf., vol. 191, p. 106583,

[7] [. Glikson and A. W. Woolley, “Human trust in artificial intelligence: Review of empirical research,” Acad. Manag. Ann., vol. 14, no. 2, pp. 627–660, 2020. [8] A. H. Eyeleko and T. Feng, “A Critical Overview of Industrial Internet of Things Security and Privacy Issues Using a Layer-Based Hacking Scenario,” IEEE Internet Things J., vol. 10, no. 24, pp. 21917–21941, 2023. [9] M. A. Muhammed Nor et al., “Role of Human Factor for the Implementation of the ‘Vision Zero’ Concept in Railway Transport: An Overview”,” in AIP Conference Proceedings, 2024, vol. 2998, no. 1, p. 030005. [10] G. W. Vogl, B. A. Weiss, and M. Helu, “A review of diagnostic and prognostic capabilities and best practices for manufacturing,” J. Intell. Manuf., vol. 30, no. 1, pp. 79–95, 2019. [11] J. Majumdar, R. S. Mohril, B. K. Lad, and M. S. Kulkarni, “Logistics, Reliability, Availability, Maintainability and Safety (L- RAMS) for Intelligent, Interconnected, Digital and Distributed (I2D2) Empowered Futuristic Military Systems,” IEEE Trans. Eng. Manag., vol. 71, pp. 5869–5884, 2024. [12] F. Su, C. Liu, and H. G. Stratigopoulos, “Testability and Dependability of AI Hardware: Survey, Trends, Challenges, and Perspectives,” IEEE Des. Test, vol. 40, no. 2, pp. 8–58, 2023. [13] E. Brusa, “Digital Twin: Toward the Integration Between System Design and RAMS Assessment Through the Model-Based Systems Engineering,” IEEE Syst. J., vol. 15, no. 3, pp. 3549–3560, 2021. [14] M. Payette and G. Abdul-Nour, “Machine Learning Applications for Reliability Engineering: A Review,” Sustainability, vol. 15, no. 7, pp. 1–22, 2023. [15] I. Hazra et al., “Prognostics and Health Management of Unmanned Surface Vessels: Past, Present, and Future,” J. Comput. Inf. Sci. Eng., vol. 24, no. 8, p. 080801, 2024. [16] J. Perez-Cerrolaza et al., “Artificial Intelligence for Safety-Critical Systems in Industrial and Transportation Domains: A Survey,” ACM Comput. Surv., vol. 56, no. 7, pp. 1–40, 2024. [17] M. Molęda, B. Małysiak-Mrozek, W. Ding, V. Sunderam, and D. Mrozek, “From Corrective to Predictive Maintenance—A Review of Maintenance Approaches for the Power Industry,” Sensors, vol. 23, no. 13, p. 5970, 2023. [18] M. Rodríguez-Hernández, A. Crespo-Márquez, A. Sánchez-Herguedas, and V. González-Prida, “Digitalization as an Enabler in Railway Maintenance: A Review from ‘The International Union of Railways Asset Management Framework’ Perspective,” Infrastructures, vol. 10, no. 4, p. 96, 2025.