Validation of a first Hyperloop prototype by early estimation of product characteristics and definition of important key components by means of scaling
Bachelor thesis by Tobias Panitz
The Hyperloop is a train that travels through a partially vacuumized tube, making it faster than airplanes. With the help of the Hyperloop, travel in the future should not only become faster, but also more sustainable and prevent environmentally harmful short-haul flights.
Especially at the beginning of the product development process and prototyping, there are high uncertainties to what extent the model meets the requirements and properties of a real demonstrator. Therefore, the focus in this work was placed on scalability. The goal of this work was to reduce these uncertainties as much as possible by analyzing and defining the most important key components and by thinking ahead about the future system properties.
The focus was primarily on the mechanical and electrical components. Fluid mechanical systems were neglected for this consideration. Significant differences between the properties of a model and a real demonstrator were found, particularly in the areas of infrastructure, the structural frame and the drive in the form of a linear induction motor. Based on the results obtained there, measures were derived for future product generations, for example in the form of a change in technology or by identifying further research potential, which aim to achieve the highest possible similarity between model and demonstrator.
Analysis of Machine Learning Based Surrogate Models for FEAStructural Models in the Design Pro-cess of a Hyperloop Prototype
Bachelor thesis by David Kubeneck
The student initiative mu-zero HYPERLOOP e.V. is researching on the Hyperloop principle. The Hyperloop principle describes a transport system in which transport capsules are moved in partially evacuated tubes with the help of electromagnetic propulsion systems. Prototype development in the initiative involves a process in which the various subsystems of the prototype are developed in parallel. This leads to conflicts when development progress on one component leads to new requirements on another component. One possible solution to such problems is a holistic system mapping. For structural analyses, this requires fast FEM simulations. One way to accelerate these is the application of surrogate models based on machine learning algorithms.This thesis investigates the applicability of such surrogate models in the context of product development in the student initiative. A surrogate model is created for the frame of the prototype. For this purpose, it is modeled and described using parameter coding. The five descriptive parameters span a design space of over 2.3 million variants. Of these, an FEM simulation is performed for 2500 variants. These serve as a training data set for the machine learning model.The model learns the relationships between parameters and FEM results. The results are reduced to maximum displacement and stress. Comparatively, two algorithms are implemented:Artificial Neural Networks (KNN) and Support Vector Machines (SVM).
The results show that the SVM is more economical to build and has higher prediction quality. At the same time, it requires only 100 FEM examples (about 3 hours of simulation time) to completely cover the design space. Predictions for individual design variants can be made within milliseconds. In conclusion, the surrogate model has a high potential for large design spaces in product development. For these, it allows statements to be made about all variants based on a few example simulations. This allows flexibility in the design selection.