AI based prediction of the results of bulk forming simulations

Theme Artificial Intelligence, Forming technology
Project title AI based prediction of the results of bulk forming simulations (KImulation)
Project duration 01.04.2015 – 31.03.2018
Download
Press release

Software and hardware for FEM simulation of bulk forming processes are well developed. However the utilization of FEM simulation in the product and process development of forged parts is still characterized by a long time effort for the calculation. The projects objective is the development of a prediction method, which allows a sufficiently detailed forecast of the results of forming simulations out of a CAD environment in a short period that does not bother the designer. That means e. g. the correlation coefficient should be bigger than 90 percent and the prediction time should be shorter than 60 seconds. In the project algorithms of artificial intelligence (AI) were used to predict the result of the simulations. The results were characterized e. g. by the required forming force, the filling of the cavity or the geometry of the finished part. Selected data mining (DM) algorithms were trained and evaluated with the results of FEM simulations. Therefore, e. g. a geometry model had to be developed that can be interpreted by the DM algorithms. The feasibility of the innovative approach was shown by preliminary investigations. The research results help to reduce the number of iterations between design and time consuming FEM calculation to a necessary minimum.

Publications about the project

In forging industry, the development of new bulk metal forming technologies still is determined by a separation between construction and simulation. The resulting iterations take a lot of time. In this paper, the data mining method neuronal network is used to predict the forming force of a finite element forging simulation of a flange.

simulation, AI, prognosis, forming force

In the forging industry, like in many other economic sectors, it is common to simulate forming processes before executing experimental trials. An iterative simulation process is more economic than trials only but still takes a lot of time. A simulation with realistic parameters takes many hours. For an economical production the idea of predicting some main results of the simulation by Data mining was developed. Within this paper, the use of four different Data mining methods for the prediction of certain characteristics of a simulated flange forging process are presented. The methods artificial neural network, support vector machine, linear regression and polynomial regression are used to predict forming forces and the lack of volume. Both are important parameters for a successful simulation of a forging process. Regarding both, forging forming forces and lack of volume after the simulation, it is revealed that an artificial neural network is the most suitable.

data mining, artificial neural network, linear and polynomial regression, support vector machine

The generation of bulk metal forming processes needs a lot of time. Researchers construct and simulate many days until a forming sequence without defects is found. At IPH a algorithm is supposed to predict a simulation result within one minute based on the constructions made. The basis for the prediction are many simulations, which were set up, executed and analysed automatically. This article decribes one possible way of doing so.

KImulation, simulation, automation

The generation of bulk metal forming processes needs a lot of time. Researchers construct and simulate many days until a forming sequence without defects is found. At IPH a algorithm is supposed to predict a simulation result within one minute based on the constructions made.

Artificial intelligence, FEA

Sponsor

The project no. 267446240 received funding from the German Research Foundation (DFG).

Your contact person

Dr.-Ing.

Benjamin Küster

Manager production automation

Mareile Kriwall
Dipl.-Ing.

Manager process technology