An Original Computational Framework for Digital Twin Simulation of Complex Mechatronic Assemblies in the Nigerian Automotive Industry

Digital twin technology, representing a dynamically synchronized virtual replica of a physical mechatronic system updated in real time from operational sensor data, holds transformative potential for product design validation, production process optimization, and predictive maintenance in the emerging Nigerian automotive assembly and component manufacturing industry. The application of digital twin frameworks to the heterogeneous and data-sparse industrial context of Nigerian automotive manufacturing, where formal product lifecycle data systems are often absent, presents foundational theoretical and computational challenges unaddressed in the existing digital twin literature developed for fully digitized Western manufacturing environments. This dissertation develops an original computational framework for digital twin construction and simulation of complex mechatronic assemblies under data-sparse conditions, integrating physics-based high-fidelity simulation, sparse Bayesian identification of assembly parameter uncertainties, and Gaussian process-based surrogate modelling for real-time state estimation. A novel adaptive synchronization protocol is derived that dynamically adjusts the digital twin update frequency and model fidelity level based on the information content of available sensor streams and the predicted divergence between virtual and physical assembly states. The framework is developed and validated on a body-in-white robotic welding cell at an automotive assembly plant in Lagos State. Digital twin prediction accuracy for weld joint dimensional quality was validated against physical measurement data at a mean absolute error of 0.31 millimetres. A formal information-theoretic metric for digital twin fidelity quantification is developed as an original contribution, providing a principled basis for comparing digital twin implementations across varying sensor data availability levels. Keywords: digital twin, automotive manufacturing, Bayesian identification, Gaussian process, mechatronic assembly