An Application of Machine Learning for Evaluating Geometrical Relations in the Mechanical Interation Process

Abstract

The mechanical integration process at Volvo Cars is used for balancing geometric requirements and solutions to ensure correct packaging of all components in the vehicle. To increase efficiency in the task of validating geometrical relations, the application of machine learning is investigated. A study of the users’ behaviours is conducted through 15 interviews. Five concepts are identified as possibilities for utilizing machine learning. In addition, examination of machine learning’s potential is performed by iteratively analysing data from historical integration validation. This process results in the development of a model able to predict critical geometrical relations with an accuracy of approximately 75%. The model is basis for suggestions of possible applications. Users are found to utilize the mechanical integration process inconsistently. Hence, the data lacks certain vital pieces of information to be fully functional. Furthermore, the mechanical integration process serves as an opportunity to give the users awareness of the geometrical environment. If machine learning were to replace the human involvement the geometrical awareness will be lost. Since this project is delimited to use historical data that is available today the results only show what is possible in the current state. The results indicate that machine learning has the potential to be applied in the mechanical integration process. However, the utility is questionable due to the needs not aligning with the current possibilities. Therefore, the recommendation is to not apply machine learning in mechanical integration process today. If Volvo Cars is able to change the users’ behaviours, mechanical integration process’s interface or both, data can be collected making it possible to utilize a machine learning model better by aligning the needs and the possibilities in the future.