Estimating product-to-product variations in metal forming using force measurements

The limits of production accuracy of metal forming processes can be stretched by the development of control systems for compensation of product-to-product variations. Such systems require the use of measurements from each semi-finished product. These measurements must be used to estimate the final quality of each product. We propose to predict part of the product-to-product variations in multi-stage forming processes based on force measurements from previous process stages. The reasoning is that final product properties as well as process forces are expected to be correlated since they are both affected by material and process variation. In this study, an approach to construct a moving window process model based on historical data from the process is presented. These regression models can be built and updated in real-time during production. The approach is tested with data from a demonstrator process with cutting, deep drawing and bending stages. It is shown that part of the product-to-product variations in the process can be predicted with the developed process model.