The company RA Consulting GmbH is an IT service provider and tool specialist located in Bruchsal. Their portfolio consists of solutions for the automotive industry. RA Consulting GmbH is particularly characterized by its know-how in measurements, calibrations and in the field of diagnostics. The QuickCheck will investigate whether statistical analysis and machine learning can be used in the development of these systems. This requires an understanding of the recorded diagnostic data. Based on this understanding, the data can be examined for the occurrence of patterns.
In order to extend the tool functionality, an analysis of the values recorded in the development should take place. The correlations and insights uncovered could then be fed back into the development of new software solutions. Thereby, a potential enlargement of the portfolio of RA Consulting GmbH can be expected. An extension of existing systems and diagnostic tools is also conceivable. In the context of the QuickCheck, an initial investigation and evaluation of possible data sources is carried out. The feasibility of analysis concepts is to be evaluated on the basis of prototypical implementations. The aim is to investigate the extent to which AI-driven methods can positively influence tool development.
The problem under consideration is an open question. Here, an explorative approach is appropriate at first. Through an initial statistical analysis of the recorded diagnostic data, a human-understandable overview is created, and data understanding is improved. By projecting the data with regard to a wide variety of dimensions, patterns and correlations emerge that subsequently need to be investigated further. In addition to classical correlation analyses, elaborate learning methods for visualizing complex high-dimensional data spaces help to assess whether the patterns found are suitable for classifications or automated evaluations.
During the QuickCheck with RA Consulting GmbH, promising patterns were found in the data. In the process, correlations were uncovered at a wide variety of data levels. Analyses and methods were presented that can be used to process and view the recorded data. Based on this understanding of the data, the findings were tested for their usability in automated machine learning techniques. The results were interesting and can be followed up. In addition, some findings can be used as a starting point for analysis by domain experts.