Optimizing Early Detection of Production Faults by Applying Time Series Analysis on Integrated Information

According to the Industry 4.0 initiative, industry aims for total automation and customizability using sensors for data retrieval, computer systems such as clusters and cloud services for large-scale processing, and actuators to react in the production environment. Additionally, the automotive industry is focusing increasingly on gathering information from the aftersales market using sensors and diagnostic mechanisms. All this information enables more accurate classification of faults when cars malfunction or exhibit undesired behavior. Since finding systematic faults as quickly as possible is a key to maintaining a good reputation and reducing warranty costs, techniques must be established that recognize increasing occurrences of fault types at the earliest possible point in time. Several sources of information exist that store heterogeneous datasets of varying quality and at various stages of approval. Using as much data as possible is fundamental for accurately detecting critical developing faults. In order to appropriately support the combination of these different datasets, information should be treated differently depending on its data quality. To this end, a concept to optimizing early fault detection consisting of four components is proposed, each of them with a particular goal; (i) determination of data quality metrics of different datasets storing warranty data, (ii) analysis of univariate time series to generate forecasts and the application of linear regression, (iii) weighted combination of course parameters that are calculated using different predictions, and (iv) improvement of the system accuracy by integrating prediction errors. This concept can be employed in various application areas where multiple datasets are to be analyzed using data quality metrics and forecasts in order to identify negative courses as early as possible.