Extension of Experimentally Assembled Processing Maps of 10CrMo9-10 Steel via a Predicted Dataset and the Influence on Overall Informative Possibilities

Abstract

Processing maps embody a supportive tool for the optimization of hot forming processes. In the present work, based on the dynamic material model, the processing maps of 10CrMo9-10 low-alloy steel were assembled with the use of two flow curve datasets. The first one was obtained on the basis of uniaxial hot compression tests in a temperature range of 1073–1523 K and a strain rate range of 0.1–100 s−1. This experimental dataset was subsequently approximated by means of an artificial neural network approach. Based on this approximation, the second dataset was calculated. An important finding was that the additional dataset contributed significantly to improving the informative ability of the assembled processing maps in terms of revealing potentially inappropriate forming conditions.

Correlation among the Power Dissipation Efficiency, Flow Stress Course, and Activation Energy Evolution in Cr-Mo Low-Alloyed Steel

“…Based on the gained η-range, the deformation course of the studied material can be associated with the specific metallurgical processes, e.g., η-range from 30 to 50% is usually attributed to the DRX course, lower values correspond with the DRV and the η-values above the ca. 60% can have connection to the superplastic behavior [6,23,24]. In the case of the studied steel, the η-values go beyond the 30% threshold (see Figure 5)-Thus, the softening course is probably mediated via DRX.…”

“…Based on the adaptation procedure (analogical to that utilized in [6]), an ideal MLP-network architecture was found to be given by two hidden layers, each with eight perceptrons; these hidden perceptrons were activated via a hyperbolic-tangent sigmoid activation function [37].…”

“…It should be noted that the input vectors ( , , ) are introduced to the network in a normalized form (i.e., dimensionless and with the sample standard deviations equal to 1.0). The applied normalization procedure was analogical to that described previously in [6]. The entire approximation and prediction procedures have been performed by means of MATLAB ® 9.3 software [45] with the embedded Neural Network Toolbox™ 11.0 (MathWorks ® , Natick, MA, USA) [46].…”

“…ε, T) are introduced to the network in a normalized form (i.e., dimensionless and with the sample standard deviations equal to 1.0). The applied normalization procedure was analogical to that described previously in [6]. Based on the assembled MLP network, flow curve prediction has been realized for five additional temperatures (1078, 1158, 1253, 1358, and 1483 K) under the experimentally tested strain rate levels.…”

“…Nevertheless, Quan et al [17] have proposed to extend an experimental dataset by means of a prediction procedure since the limited number of experimental curves can lead to the inferior informative ability of compiled processing maps. This suggestion has been recently studied in the case of 10CrMo9-10 steel [6] when the additional (predicted) dataset revealed potentially inappropriate forming conditions. In connection with the above presented maps, Zhou et al [1] utilized a term-conventional hot processing (CHP) maps.…”