Development of a machine-learning model from a dataset primarily generated from the test data, which enables us performing complex predictions and has a number of applications in the field of engineering. Currently, the headlamp height and corresponding dipped beam height is determined through a physical test for different set of loading conditions, tire pressures and headlamp leveling switch positions as per the vehicle level test regulation. These tests being a part of vehicle certification requirement, falls at the end of a vehicle program. Considering this aspect, the total time consumed, from vehicle development to availability of the prototype test vehicle and the physical test to be close to a year and half.
To enable the test engineer to perform these tests in absence of physical prototype as a front-loading activity, a methodology devised to develop an Artificial Neural Network (ANN) to understand the complex relationship between the vehicle parameters governing the headlamp height and consequently the dipped beam height. Initially, vehicle parameters like vehicle weight, front and rear axle loads, suspension stiffness, tire deformation, headlamp leveling motor voltage etc. are identified that governs the sensitivity of headlamp and dipped beam height. Preparation of data is essentially the first step in development of a data-driven model. Hence, test data is prepared comprising of 25 most influencing parameters of similar class of vehicles. The ANN configuration selected has one input layer, one hidden with seven neurons and one output layer and the algorithm is trained using 80% of test data. The remaining 10% of test data is for validation and testing respectively, using the Bayesian regularization algorithm as a fair amount of correlation established with the physical test data with an accuracy of about 98%. From this work, we can predict the dipped beam height of the vehicle headlamp and finalizing the design specifications for achieving the regulatory requirement in absence of physical prototype.