Precise Prediction of Microstructural Properties with Minimal Experimental Effort for the Nickel-Base Alloy Inconel 718

The prediction of microstructure evolution in addition to the macroscopic material strength, material flow and temperature evolution is becoming increasingly important as more and more complex materials, with properties that are heavily influenced by their microstructure, are being used. This in turn requires refined microstructure models to be parameterized. Compared to flow curve models, the experimental effort for the parameterization of microstructure models increases due to the inclusion of grain size and recrystallization effects. Therefore plenty of experiments are usually performed to fully characterize the material at hand. The increasing versatility of testing machines, like dilatometry with easily variable temperatures, in addition to the growing expenses that go along with increasing the number of experiments for high cost materials, leads to the question whether performing all those experiments is really justified. In this paper the microstructure model StrucSim is parameterized for the nickel-base alloy Inconel 718 and coupled online with a finite-element (FE) simulation to predict the material behaviour during double compression tests. StrucSim combines multiple constitutive equations into a single consistent material model representing also the microstructure. Therefore these constitutive equations are parameterized to their respective metal-physical phenomena to find the initial parameters for StrucSim. Afterwards the set of final parameters is determined by optimizing the initial parameters using the StrucSim algorithm interconnecting the constitutive equations to define a reference model. The reference model is later compared to different final parameter sets parameterized based on reduced experimental data. Beforehand the reference model is coupled with the FE software Simufact.forming to simulate double compression tests and compare them to experiments as a validation of the reference model. Here forces are predicted with a mean deviation (root of the sum of squared relative errors) of 7.6 % and grains sizes with a mean deviation of about 8 μm from the measurements. Afterwards the influence of reducing the available data during parameterization of StrucSim is investigated to evaluate the possibility of reducing the experimental effort. It is shown that when using only 50 % of the data the quality can be maintained with the reduced model. When simulating the double compression tests a comparable deviation regarding the forces and grain sizes is achieved. Reducing the number of experiments by 50 % during materials characterization therefore appears feasible.