Analyzing data of endurance tests is a time consuming task during the product development of steering systems. Such tests are performed to ensure the quality of each product design. If it is not possible to automate specific analysis of the measured data to the greatest possible extent, it has to be performed manually. This thesis applies machine learning methods to an analysis, which could not be automatized by a previous classical analysis, e.g. threshold rules. The models, one and two dimensional convolutional neuronal networks, predict individually for each second of a long endurance test of steering systems the probability of containing squeak noise events. The advantage of using such models instead of manual labelling the data is, that data can be classified automatically, the results can be visualized in a report and the classification can be extended to further classes or similar problems with comparably little effort.
Titelaufnahme
Zugriffsinformation
Klassifikation
Abstract
Lizenz-/Rechtehinweis