Part-to-Part model predictive control - using a modified Gauss-Newton scheme

Abstract

Over the last two decades there has been and increasing academic and industrial focus on process robustness and process control applied on sheet-metal forming processes and a high number of control schemes has been proposed. However, many of proposed control schemes rely on process data sampled during the punch stroke (in case of stamping and deep drawing) e.g. draw-ill sensors and the absent of robust sampling technologies is on of the main barriers preventing industrial implementation. One approach to avoid in-process sampling is to apply an part-to-part control strategy, where the process parameters are updated based on post-processing samples e.g. flange geometry. Different algorithms has been proposed, the current article will adapt the algorithm proposed by Endelt (2017) which are based on a Gauss-Newton formulation (non-linear least-square). The control scheme is applied on the NumiSlieet’08 S-rail benchmark, the algorithm has previously been applied on a square deep-drawing with good results and the present paper illustrate the flexibility of the proposed part-to-part control scheme, as the algorithm can be directly adopted to s-rail process using the design rule from the original paper.