Digitalizing Assembly Operations Using Agent-Based Modeling Techniques and Digital Twin Technologies

Abstract

Digital Twins promise continuous visibility into assembly operations, but most are built inside commercial simulation platforms that small and medium manufacturers cannot afford, and most monitor equipment rather than individual products. This research develops Agent-Based Digital Twin models and the computational infrastructure to support them, and tests them by digitalizing an educational automotive assembly facility that produces LEGO® cars under both Mass Production (MP) and Lean Manufacturing (LM) configurations. As foundational work, a hybrid Discrete-Event and Agent-Based simulation model of the assembly line is designed, developed, verified, and validated to demonstrate how simulation techniques can be used to study assembly operations. This model analyzes system performance under MP and an alternative Just-In-Time (JIT) scenario, and establishes the modeling techniques used throughout the dissertation. The first contribution presents a systematic approach to designing Digital Twin agents for real-time monitoring of assembly operations. This approach uses Agent-Based Modeling (ABM) techniques to create a virtual representation of both the physical system’s state and the products moving through it. Real-time data from Internet of Things (IoT) devices and automation systems twins the physical and virtual worlds, enabling continuous production tracking. The second contribution generalizes the implementation experience into a Multi-Agent Based Digital Twin (MABDT) architecture. This three-layer framework, comprising Physical, Digital Twin, and Application layers, provides a reusable structure for developing agent-based Digital Twin systems in manufacturing environments. The architecture formalizes agent behavior through statecharts, defines communication patterns between agents, and identifies the components needed at each layer. The third contribution develops a purpose-built computational engine for creating and executing Digital Twin models based on the MABDT architecture, independently of commercial simulation software. This engine provides a simulation environment, an event-driven communication kernel for inter-agent messaging, a user interface layer, and support for real-time applications including dashboards, IoT command interfaces, and intelligent assistants. By relying solely on open-source software, the engine lowers licensing and lock-in barriers that limit Digital Twin adoption among small- and medium-sized manufacturers and academic researchers.