In recent years, numerical simulation has been used widely by the metal-forging industry and has become essential in most forging operations. The traditional, time-consuming and costly trial-and-error method has been replaced by increasingly sophisticated simulation software that can accommodate the whole manufacturing process from shearing to multistage forging, flash trimming and through to quenching.

Dramatic improvements in computer hardware capacity, as well as the constant development of more efficient algorithms, have made it possible to simulate the most complex forged parts within a very short computation time. A case in point is the simulation package FORGE 2011, developed by the Centre de Mise en Forme des Materiaux (CEMEF) laboratory of Mines ParisTech and distributed by Transvalor.

    The program is able to give accurate simulation results for a variety of closed-die parts – such as crankshafts, knuckles, axles, connecting rods or camshafts – within a few hours on high-performance computing systems such as clusters or multi-core computers. Common processes such as open-die forging, ring rolling, spinning or cold forging are simulated on a daily basis by numerous users around the world.

    Although trial-and-error techniques have had a long run, their replacement by numerical techniques has thus far brought little change in methodology. Instead of a real trial, a numerical one is made on the computer, and the user runs a number of different simulations until an acceptable result is produced. These numerical input data are then translated into real tool machining and process parameters in order to make the “first time right” real trial, which validates simulation results. Although this virtual technique is faster and more cost effective, it still requires extensive human involvement in data preparation and, most importantly, in result review and analysis.

    Moreover, a lot of experience and know-how of the forming process are needed in order to design the necessary changes that will improve the process.

MAES Algorithm

In order to partly overcome this weakness and improve simulation benefits, different optimization algorithms are available. FORGE 2011 by Transvalor embeds a Meta-model Assisted Evolution Strategy (MAES) algorithm, which is described by the flow chart in Figure 1.

    Instead of the user having to manually define parameters to reach his objectives, the MAES algorithm, coupled with the Finite Element software, generates sets of parameters to reach the objective(s) under given constraints. The relevant parameters, the variations they may have within a specific range and the constraints simulation results must comply with are defined by the user in the optimization preparation stage.

Optimization Examples

Crankshaft

The first example shows the benefit of using automatic optimization in conjunction with forging simulation. The forging sequence of this crankshaft is described in Figure 2. In this example, the objective is to minimize the cut weight of the initial billet, which in this case is 355 pounds. The parameters of the optimization apply to the rolled billet and relate to its length and large diameters. The constraint applies to the finisher stage, where a complete filling of the dies is required.

    The result of this automatic optimization is shown in Figure 3. The flash pattern in the finisher stage using the initial rolled billet (355 pounds) is shown in red. The flash pattern of the optimized rolled billet (311 pounds) is shown in blue.

Open-Die Forging

The second example relates to the optimization of an open-die forging process as described in Figure 4. The material to be forged is stainless steel, and it is forged in four passes.

    In this case, the objective is to improve the microstructure of the bar after cogging (i.e. maximizing the average ASTM grain size). The parameters of the optimization are the initial temperature of the bar (ranging from 1000-1250°C/1800-2282°F) and the velocity of the tools (ranging from 40-60 mm/s). The worst results are shown in Figure 5; the best results in Figure 6. In both cases, the value shown is the ASTM average size, and the legend ranges from 3.5 to 6.5.

    The optimization shows that better microstructure is achieved with a lower initial temperature of the bar and higher forming velocities. It also shows that a high forming temperature generally gives poor microstructural results independent of the press velocity.

    Both preceding examples were run using the commercial features of the Finite Element code of the FORGE simulation package – variation of the billet shape in the crankshaft (Example 1) and variation of the process parameters in the open-die forging (Example 2). In some cases, however, a coupling of the MAES optimizer and CAD packages could be required.

Automotive Part Preform

The final example shows a successful coupling of FORGE software to a commercial CAD software package for the design of an automotive part preform. The objective is to optimize the forged preform after bending in order to fill the finisher die impression with the least possible amount of material and with no defects (e.g., no laps or folds).

    Figure 7 shows the different parameters allowed to vary inside the parametric CAD system and optimized by the MAES algorithm. After each simulation, the CAD system was launched and fed with the new diameters until an optimum was found. Figure 8 shows the results of the optimization.

Conclusion

The application of optimization methods to forging and, more generally, to metal-forming simulation is a relatively new approach. Though R&D centers and universities have been working on these techniques in recent years, they have remained complicated because they usually involve using a variety of very different codes. The newly integrated MAES algorithm in the FORGE software system makes it relatively easy to optimize many parameters in a range of processes such as closed-die forging, open-die forging, heat treatment and others. The industrial examples presented in this article were simulated within hours on recent multi-core systems.

Acknowledgement

This work has been carried out under the auspices of the French National Research Agency (ANR) through its LOGIC program, whose support is gratefully acknowledged. The support of Bharat Forge Kilsta AB (Sweden) is also gratefully acknowledged.