A framework for computer vision for virtual-realistic multi-axial real-time hybrid simulation

Real-time hybrid simulation has gained popularity over the last 20 years as a viable and cost-effective method of testing dynamic systems that cannot be tested using traditional methods. The emergence of multi-axial Real-time Hybrid Simulation (maRTHS) has led to an increase in the allowable fidelity of the numerical and experimental substructures. The testing community can now replicate multiple-degree-of-freedom (MDOF) responses of both substructures and thus can perform more representative tests. However, with this increased fidelity of the substructures comes an increased complexity of controlling these components. Specifically, multi-axial hydraulic actuator assemblages require nonlinear coordinate transformations to derive plant displacements as the force transducers on the actuators are not capable of performing this task directly. Recently, benchmark problems have been provided to the RTHS community in the form of virtual simulations. Virtual simulation refers to a fully virtual testing methodology where numerical and physical components are represented virtually. This approach enables the RTHS community to evaluate various control algorithms without the need to recreate physical components. This project aims to demonstrate the capability of computer vision-based displacement tracking in a realistic virtual simulation of the experimental substructure in avoiding excess nonlinear coordinate transforms. The tracking algorithm utilizing the Lucas-Kanade optical flow method is tested in the virtual simulation environment which is set up using real-time 3D creation engine, Unreal Engine 4 (UE4), and computer graphics software, Blender. This environment interfaces with MATLAB/Simulink, more specifically “Simulation Tool for v-maRTHS benchmark” developed for multi-axial tests. The result of this study establishes a novel framework for applying computer vision-based tracking algorithms and sensing in v-maRTHS simulations using simulated cameras within virtual simulation environments. A computer vision displacement tracking algorithm is developed and optimized to work in tandem with a MIMO PI controller to reduce tracking time delays within 31.25 milliseconds while tracking the nodal displacement and rotation of the frame within a normalized RMSE of 1.24 and 1.10 respectively.