Direct and inverse models in metal forming: A soft computing approach

A non-traditional framework is presented for building inexact models in metal forming using the softcomputing approach. Architecture for building integrated inverse models is developed using neurocomputing.; To date efforts in process modeling have tried to develop exact laws (equations). This dissertation takes the contrasting approach of formulating inexact direct models with their inherent imprecision expressed using fuzzy sets. The approach departs from traditional approaches by replacing equation based relationships with rule based relationships. Fuzzy sets are used to quantify the rules. The theory of approximate reasoning is used to implement inference from these rules. The approach integrates neural networks into the model to learn the inverse model from the direct model.; A Windows® based system is built to implement the framework. The system is object oriented, modular and extensible. It supports the activities involved in building soft models. New models can be built from ground up. The framework is generic and can be used to build inexact rule-based models in any domain. It is shown that models developed using this approach are able to incorporate more complex relationships than is possible with differential or regression equation based approaches. A Model can be built using numerical data or a priori knowledge expressed in terms of rules. The model is stored in a linguistically interpretable form. Modelers and users can modify or adjust a given soft model by adding more knowledge.; The approach is applied to three case studies in metal forming: (1) austenite grain size model in hot forging, (2) damage value model in cold forging and (3) static recrystallization model in hot rolling. The results are analyzed using traditional graphing and compared to traditional approaches. Inverse models are built for the three case studies using neural networks.