Early Detection of Critical Faults Using Time-Series Analysis on Heterogeneous Information Systems in the Automotive Industry

Thomas Leitner, Christina Feilmayr, Wolfram Wöß
Beside the manufacturing industry's vision of industry 4.0, which is about improving the degree of automation and cus-tomisability depending on a huge amount of data, the automotive industry increasingly advances the after-sales market collecting more and more information about the car using sensors and diagnostics mechanisms. This information can be used to earlier reveal malfunctions and faults with rising quantity that customers experience in order to reduce the solving time of the problem.
more » ... me of the problem. Different heterogeneous data sets exist storing data at various approval stages with different data quality. In order to perform the most accurate detection of critical developing faults it is fundamental to use as much data as possible while weighting their impact by assessing their data quality. For detecting critical performing faults as early as possible time series analysis and forecasting methods are used to analyze their course and predict future values. In this research work, a new approach is proposed, which is subdivided in the following four main tasks: (i) evaluation of data quality metrics of different warranty information systems, (ii) analysis and generation of forecasts on univariate time series based on Auto-Regressive Integrated Moving Average (ARIMA), (iii) weighted combination of different predictions, and (iv) improvement of the accuracy by integrating prediction errors. This solution can be used in arbitrary fields of application, in which different information sources should be analyzed using data quality metrics and prediction errors to determine critical courses as early as possible.