The dynamics of the automobile industry and the complex requirements from the automotive Own Equipment Manufacturers (OEMs) demands the product development companies to continuously develop new products and improve the quality of existing products while meeting the automotive standards. As a result, complex testing systems have evolved to ensure consistently high quality and reliability while using these products. One such automotive-related product is Instrument Clusters, which generally displays all the related data for driving, assistance and the status of the vehicular systems. With the advent of new sensors and technologies, Instrument Clusters intricacy have increased with a large number of features and safety aspects. The most imperative and demanding phase of Instrument Clusters development are safety functions of functionalities. Exhaustive and comprehensive testing is indispensable to ensure the product’s high quality, reliability, and in turn driver’s safety. Despite well designed testing processes and strategies that have been followed currently, the error correction effort is still not optimal, primarily due to latency in error identification during product development cycle. Thus, there is dire necessity to identify the cause of these errors and system factors that influence the errors and improve the robustness of the product. For example, a frozen video stream on the car display increases the chances of accidents due to loss of visuals to the driver.
In this talk, I propose to use deep learning in my PhD thesis and compare its performance to conventional approaches. The aim is to build a real-time system that predicts irregularities and supports domain-experts by preventing them from contemplating irrelevant data and rather pointing them to the relevant system factors. The underlying idea is to continuously learn the normal behavior from in-vehicle data and then to autonomously detect unexpected deviations, generally termed as anomalies and analyze them using exploratory data analysis.