Automated Outlier Detection in Multidimensional Driveability Data Using AVL-DRIVE

With the increased number of variants, the preservation of a brand-specific vehicle DNA becomes more and more important. Paired with growing customer expectations, brand DNA can be a crucial point in the decision-making process of buying a new vehicle. Whereas the customer will assess the DNA subjectively during driving by evaluating the vehicle drive quality (“driveability”), most manufacturers are not merely relying on subjective evaluations by having test drivers perform maneuvers with prototype vehicles. Nowadays, the assessment is performed objectively during the vehicle development process. As a supporting measure, the Anstalt für Verbrennungskraftmaschinen List (AVL) has made the objective assessment tool AVL-DRIVE commercially available. Up to now, the AVL-DRIVE ratings had to be manually analyzed and checked for outliers. Low ratings and high deviations to a priori specified target values are a good starting point for the search of outliers. Yet there is more expert knowledge required, which is merged in the AVL-DRIVE Outlier Editor. Based on distance and density metrics paired with domain knowledge, automated detection of outliers in multidimensional driveability data is performed. As a result, the number of outliers present in the data and the improvement potential given by the elimination of the outliers are calculated. With customizable outlier thresholds, individualization of the outlier analysis can be performed. By relying on computational power rather than manual outlier screening operations, the presented methodology allows to save time and resources, thus beneficiating the development process from start to finish.