Date of Award
5-2026
Document Type
Thesis
Degree Name
Master of Science in Engineering (MSE)
Department
Civil Engineering
Committee Chair/Advisor
Dr. Laura Redmond
Committee Member
Dr. Christopher McMahan
Committee Member
Dr. John Wagner
Committee Member
Dr. Pamela Murray-Tuite
Abstract
Digital twins can be used for system performance optimization, health monitoring, predictive maintenance, and investigating system anomalies. A ground vehicle digital twin may be assembled using representations of the subsystem models (tires, suspension system, powertrain, etc.) of different fidelity. Lower fidelity Digital Twins have relatively simpler design and physics, hence are generally computationally less expensive, but may not represent the physical system accurately. On the contrary, higher fidelity Digital Twins are generally more precise in their ability to represent the physical system, but due to complexity, are usually computationally more demanding. This thesis describes the application of Bayesian Classification within the novel Bayesian Assembly Tool for selecting subsystem models to construct suitable digital twins. This tool consists of two phases: first, each subsystem model is calibrated to a set of experimentally collected field or laboratory data (or against the highest fidelity model if test data is not available); second, Bayesian classification is used to select a suitable digital twin based on the likelihood of observing the experimental data from that model. If the probability of observing the data from a digital twin assembled from lower fidelity subsystem models is similar to the probability of observing the data from a digital twin composed of higher fidelity subsystem models, then using the lower fidelity option has several significant advantages, including computational time and engineering cost. The proposed tool can assess many possible mixtures of fidelities among subsystems and thus select only the high-fidelity representations where they are truly needed. The Bayesian Assembly Tool is validated on two vehicle dynamics models: a wheeled vehicle model and a tracked vehicle model, both created in Matlab/Simulink. For both models, Digital Twin assemblies with combinations of varying fidelity subsystem models are created and calibrated to laboratory test data using Griddy Gibbs calibration methodology. Then, using the posterior distributions of the uncertain model parameters, synthetic field data are simulated from each of the Digital Twin assemblies. Bayesian classification is implemented to obtain the optimal Digital Twin model that best represents the field test data. For the wheeled vehicle Digital Twin, both laboratory test data and field test data are obtained from the simulation of the highest-fidelity Digital Twin assembly. Physical test data were used to validate the tool on the tracked Digital Twin assembly.
Recommended Citation
Sharma, Shishir, "Bayesian Assembly Tool for Intelligent Digital Twin Selection" (2026). All Theses. 4785.
https://open.clemson.edu/all_theses/4785