Towards Prescriptive Maintenance Using Digital Twins and Artificial Intelligence

Licentiate thesis, 2025

In the era of Industry 4.0, industrial sectors worldwide face increasing complexity and operational challenges driven by rapid technological advancements and evolving market demands. Among these challenges, maintenance plays an important role in ensuring system reliability and sustaining productivity in manufacturing environments. While the evolution toward prescriptive maintenance promises significant value, its adoption is hindered by persistent technical and organizational barriers. This thesis develops and validates an integrated framework that leverages artificial intelligence and digital twins to enable this transformation, providing a systematic pathway from predictive insights to actionable prescriptions.

Employing a design research methodology based on five studies, this research systematically addresses three key research questions. It first identifies critical barriers to industrial implementation, including data scarcity, system integration complexity, and organizational skill gaps. To overcome these challenges, a stakeholder-weighted decision model is applied to test what-if scenarios, and the thesis proposes an AI-enhanced digital twins framework complemented by a five-layer conceptual framework that guides scalable adoption. The effectiveness of these solutions is demonstrated through two empirical studies encompassing automated maintenance prioritization using deep reinforcement learning, and unsupervised anomaly detection and localization. This work bridges the academic-industry gap by delivering both theoretical insights and validated frameworks that transform maintenance from a reactive cost center into a strategic driver of operational resilience, accelerating the adoption of intelligent, autonomous systems and supporting the evolution toward sustainable, self-optimizing manufacturing.