Variance based sensitivity analysis of deep drawing processes based on neural networks using Sobol indices

Abstract

Abstract Today’s deep drawing of car body parts is operated increasingly closer to the process limits, making it more challenging to ensure a high robustness of the process. Disturbances like varying material properties or changing tribological conditions may negatively affect the process, leading to deteriorated product properties and a loss of productivity. In the past, several approaches using different combinations of sensors and actuators to build up a control system have been investigated to fulfil the need of a more robust process. Hence, a thorough process analysis comes to be beneficial to evaluate the expediency of a control system and to make a reasonable preselection of sensors in order to avoid unnecessary costs. This paper presents a method using variant simulations to evaluate the expediency of a control system, including the necessary sensors. The influence of disturbances in the process is evaluated by a numerical sensitivity analysis under consideration of interaction effects. These higher order effects are calculated by Monte Carlo integration. Potential measurands are determined by the maximum achievable observability of not directly measurable quality criteria of a part. For this purpose, different modelling approaches of the observability are considered and compared with regard to their goodness-of-fit.