Modern day automotive market demands shorter time to market. Traditional product development involves design, virtual simulation, testing and launch. Considerable amount of time being spent on virtual validation phase of product development cycle can be saved by implementing machine learning based predictive models for key performance predictions instead of traditional CAE. Durability oil canning loadcase for vehicle hood which impacts outer styling and involves time consuming CAE workflow takes around 11 days to complete analysis at all locations. Historical oil canning CAE results can be used to build ML model and predict key oil canning performances. This enables faster decision making and first-time right design.
In this paper, prediction of buckling behaviour and maximum displacement of vehicle hood using ML based predictive model are presented. Key results from past CAE analysis are used for training and validating the predictive model. Commercially available tool is used, and predictions are compared with CAE results. Based on domain expertise, features are selected and cleaned up to make it suitable for training the predictive model. Different algorithms based on ROM (Reduced Order Modelling) and POD (Proper Orthogonal Decomposition) are used for prediction and the best performing algorithm and it’s hyperparameters are selected based on loss function (R2) and acceptable error.
Prediction using Neural Network consists of multi quadratic radial basis function (RBF) which is in good agreement (< 20 % error) with CAE predictions, and it can be improved further by adding more data into the training database. With this predictive model, maximum displacement and buckling can be predicted within 30 mins which resulted in 99% turnaround time savings when compared to existing CAE workflow.