Kracker, DavidDavidKrackerDhanasekaran, Revan KumarRevan KumarDhanasekaranSchumacher, AxelAxelSchumacherGarcke, JochenJochenGarcke2022-11-232022-11-232023https://publica.fraunhofer.de/handle/publica/42904110.1080/13588265.2022.20746342-s2.0-85130426331Stricter legal requirements in crash safety lead to more complex development processes in computer-aided engineering and result in an increasing number of simulations. Both, the construction of the simulation models as well as their evaluation are costly and time-consuming. Therefore, an automated workflow is required that significantly facilitates the analysis of the results by the engineer and increases the quality of the evaluation. In this study an automated evaluation process is proposed that detects anomalous crash behaviour in a bundle of crash simulations. The individual states from the simulation are analysed separately from each other and an outlier score is calculated using a kth-nearest-neighbour approach. Subsequently, these results are averaged into one score for each simulation. With the help of different statistical methods, a threshold value is calculated, from which a simulation can be identified as an outlier. The evaluation is carried out on 5 datasets. On average, the precision and recall of the presented method are 1.0 and 0.91, respectively.enCrash simulation analysismachine learningoutlier detectionProcess Automationjournal article