UC Davis

Digital twins are computer models that can provide accurate digital representations of physical systems. They differentiate from the other forms of computer representations in that they require data from the physical system itself to be integrated into the digital system. Studies regarding the integration of digital twins in the heavy construction vehicle industry have been limited, especially in terms of physics-based digital twins and calibration of physics-based digital twins using a comprehensive set of sensors. Establishing a digital twin using physics-based simulations is not a trivial task, especially for construction vehicles due to their interactions with unknown environments at construction sites. These digital twins would require complex sensors and data acquisition systems to be integrated with the physical system to provide data for proper identification of the physics of the interactions of the machinery with the unstructured construction environments. Another important issue is related to the fact that different sizes of vehicles may be needed to address construction needs at one or more construction sites, but the data obtained from one system cannot, in general, be scaled to other sizes of vehicles. This is especially the case when a digital model is calibrated with respect to one physical system, and then when a larger model of the physical system needs to be used for the construction task, another calibration process from physical to the digital twin is needed. In the field of construction vehicles, where there are different sizes of the same type of machinery is used, calibration of the digital twin increases both the complexity and the cost of the needed sensors and data acquisition systems for making the digital twin become a clear emulator of the physical system. The problem of scaling is addressed in this thesis by applying and enhancing dimensional analysis methods of engineering mechanics. This work represents the development of a comprehensive, calibrated digital twin of a wheel loader by utilizing physics-based simulations. The system developed uses an extensive set of sensors and data acquisition equipment and makes use of dimensional analysis and machine learning to make the physics-based digital twin scalable from small-scale physical machines to much larger-scale machines. The framework utilizes dimensional analysis principles as well as various types of neural networks to carry out the scaling process between pre-existing small-scale vehicles and the larger-scale vehicles to minimize the effect of distorted scaling factors. A physical wheel loader is instrumented as part of this work and is used to illustrate the application of the theories developed in this thesis.