In English

Machine learning for vehicle concept candidate population & verification

Björn Grevholm
Göteborg : Chalmers tekniska högskola, 2017. Diploma work - Department of Applied Mechanics, Chalmers University of Technology, Göteborg, Sweden, ISSN 1652-8557; 2017:39, 2017.
[Examensarbete på avancerad nivå]

The aim of this M.Sc. thesis is to evaluate the potential of using machine learning to support concept phase decisions to balance the thermal properties of an automobile. With the use of computer scripts, the relevant measurement data is extracted from repositories and is used to train an artificial neural network which can identify the importance of the different parameters that are involved in tuning the vehicle thermal attributes. After data for several car models has been used to train Machine Learning (ML) tools, this configuration used in predicting parameters affecting engine under hood thermal behaviour. A neural network based ranking procedure which may make it possible to reduce the order of concept decision space is also proposed. After several vehicle families gone through this prediction phase, a clustering of vehicle classes may allow for prediction and optimisation of new families, if errors due to assumptions and underlying mathematics are quantified. The project has the added benefit of allowing Volvo Car Corporation (VCC) to reuse the large amount of data which are seldom used after the initial project delivery date. Measurements collected in VCC’s wind tunnels are the main source of data for this thesis but the open-source script based method can be used on other type of data from other disciplines. A possible outcome of the thesis might be recommendation for updated procedures in creating and storing data to easier integration into machine learning based investigations.

Nyckelord: Machine learning, heat exposure protection, artificial neural networks, vehicle thermal properties, vehicle families, polynomial kernels, linear kernels, prediction, quantification of errors, regularization, radial basis functions, k-means, Support Vector Machines, Support Vector Regression, sorting, data storage, data management, wind tunnel tests



Publikationen registrerades 2017-07-05. Den ändrades senast 2017-07-05

CPL ID: 250510

Detta är en tjänst från Chalmers bibliotek