Applications of Machine Learning in Automotive Verification and Validation: A Review

In more aspects of daily life than one may think, the use of machine learning (ML) is widespread. More decisions are being made using solid evidence as a result of the use of ML in research, technology, and business. It is one of the technological sectors that is expanding the fastest right now in a variety of industries, for example manufacturing, health care, education, financial modelling, policing, and marketing. Data science, statistics, and computer science all converge in machine learning (ML) (Jordan and Mitchell in Science 349:255–260, 2015). ML’s potential to transform Automotive Product Development Process (PDP) cannot be overlooked. Though ML in the automotive industry typically linked to self-driving or autonomous cars, there are many other applications behind the scenes. The use of ML techniques in PDP improves the effectiveness of numerical solutions by fusing them with human intelligence. The agility needed in the PDP due to ever changing customer expectations can be achieved with smart deployment of the ML systems. Automotive Verification and Validation (V&V) is a complex process involving a number of lengthy tasks requiring expertise. Human inaccuracy at any stage in the development could lead to flawed engineering decisions. ML tools could assist engineers to make accurate, robust and timely engineering decisions. In this paper the examples of ML implementation specific to V&V process are discussed. These examples should give a general though process for enthusiasts in this area. A concise summary of the fundamentals of Machine Learning concept and related terms is discussed in the beginning to make the readers familiar with the basic concept behind ML. The common stages needed to create a Machine Learning model through DNN beginning with scratch are listed in a typical ML workflow that is presented.