The ASAP Group has once again lived up to its aspirations as an innovation leader and technology company, having been awarded state funding for research and innovation for another of its research projects. The Center of Excellence for Data Science/KI at ASAP’s Munich site is focused on a key question: Against the backdrop of the rising time and cost pressures in the automotive industry, how can insights from large volumes of measurement data be used as quickly as possible? A highly automated DAS must complete millions of miles of test drives before final approval can be issued. This includes both real-world test drives in test vehicles and automated testing using corresponding testing systems. Looking at this from a different perspective, the testing requirements also involve precisely recording and analysing vast quantities of data, with the results incorporated in the development process in order to continuously optimise functions.
Intelligent filter for measurement data
The aim of the funded project is to develop a tool for use in the field of test automation. Automated identification of anomalous measurement data should assist test engineers in their work. The process of validat-ing a DAS includes evaluating all measurement data from different sources (test systems, such as component and system HILs, as well as data from real-world test drives) to identify any anomalies. This, however, is where the ASAP Group’s Data Science Tool comes in. It relies on statistical data analysis and machine learning to automatically examine these vast quantities of trace data and identify any anomalous data points that require inspection by the test engineers. This means that development teams no longer need to sift through trace data: instead, they can concentrate on the anomalies identified by the tool and, ultimately, devote more time to creative activities as development engineers. The tool is not designed for any one specific case or a single vehicle function. In the future, the tool will make it possible to examine all ECU traces from HIL and real-world vehicle tests in the field of electronics development, conducting automated data analysis to identify anomalous data points.
How the tool box works
The toolbox relies on machine learning and statistical data analysis methods. Firstly, the data is loaded and examined for typical error sources such as formatting errors and potential duplicates. The signal values undergo statistical analysis to identify problems such as issues with signal specifications. Data points that logically belong together are then compiled into sequences based on the prepared datasets. Removing non-relevant data from the evaluated dataset reduces the scale of the task, thereby facilitating more efficient calculations. Algorithms are used during this process to identify anomalous data points in the meas-urement data. This involves examining the reduced data from all angles in a high-dimensional space. Mathematical metrics work in the background to ensure the success of the machine learning techniques by facilitating the automated calculation of intervals between all data points. This clusters the data points and ultimately facilitates their evaluation based on statistical anomalies, such as particularly large intervals to all other data points. The use of artificial intelligence (AI) therefore facilitates automated, rapid evaluation of all data and provides the test engineers with concrete recommendations of anomalous data points for review. The tool also generates a very simplified 3D display of the results to make them more easily comprehensible. Continuous feedback from users constantly improves the underlying AI – and thus also supports the intelligent filtering of data points and, in turn, the recommendations for action. As a result, the tool developed by ASAP to automatically detect anomalies in measurement data not only makes it possible to apply data more quickly but also enables test engineers to give faster feedback to development teams in specialist departments. The result is time and cost savings in the development process for vehicles and their functions.