skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Development of multilayer perceptron networks for isothermal time temperature transformation prediction of U-Mo-X alloys

Abstract

In this work, a multilayered perceptron (MLP) network is used to develop predictive isothermal time-temperature-transformation (TTT) models covering a range of U-Mo binary and ternary alloys. The selected ternary alloys for model development are U-Mo-Ru, U-Mo-Nb, U-Mo-Zr, U-Mo-Cr, and U-Mo-Re. These model’s ability to predict 'novel' U-Mo alloys is shown quite well despite the discrepancies between literature sources for similar alloys which likely arise from different thermal-mechanical processing conditions. These models are developed with the primary purpose of informing experimental decisions. Additional experimental insight is necessary in order to reduce the number of experiments required to isolate ideal alloys. These models allow test planners to evaluate areas of experimental interest; once initial tests are conducted, the model can be updated and further improve follow-on testing decisions. The model also improves analysis capabilities by reducing the number of data points necessary from any particular test. For example, if one or two isotherms are measured during a test, the model can construct the rest of the TTT curve over a wide range of temperature and time. This modeling capability reduces the cost of experiments while also improving the value of the results from the tests. The reduced costs could result in improvedmore » material characterization and therefore improved fundamental understanding of TTT dynamics. As additional understanding of phenomena driving TTTs is acquired, this type of MLP model can be used to populate unknowns (such as material impurity and other thermal mechanical properties) from past literature sources.« less

Authors:
ORCiD logo;
Publication Date:
Research Org.:
Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
Sponsoring Org.:
USDOE
OSTI Identifier:
1356480
Report Number(s):
PNNL-SA-120740
Journal ID: ISSN 0022-3115; 453040075
DOE Contract Number:
AC05-76RL01830
Resource Type:
Journal Article
Resource Relation:
Journal Name: Journal of Nuclear Materials; Journal Volume: 490
Country of Publication:
United States
Language:
English

Citation Formats

Johns, Jesse M., and Burkes, Douglas. Development of multilayer perceptron networks for isothermal time temperature transformation prediction of U-Mo-X alloys. United States: N. p., 2017. Web. doi:10.1016/j.jnucmat.2017.03.050.
Johns, Jesse M., & Burkes, Douglas. Development of multilayer perceptron networks for isothermal time temperature transformation prediction of U-Mo-X alloys. United States. doi:10.1016/j.jnucmat.2017.03.050.
Johns, Jesse M., and Burkes, Douglas. 2017. "Development of multilayer perceptron networks for isothermal time temperature transformation prediction of U-Mo-X alloys". United States. doi:10.1016/j.jnucmat.2017.03.050.
@article{osti_1356480,
title = {Development of multilayer perceptron networks for isothermal time temperature transformation prediction of U-Mo-X alloys},
author = {Johns, Jesse M. and Burkes, Douglas},
abstractNote = {In this work, a multilayered perceptron (MLP) network is used to develop predictive isothermal time-temperature-transformation (TTT) models covering a range of U-Mo binary and ternary alloys. The selected ternary alloys for model development are U-Mo-Ru, U-Mo-Nb, U-Mo-Zr, U-Mo-Cr, and U-Mo-Re. These model’s ability to predict 'novel' U-Mo alloys is shown quite well despite the discrepancies between literature sources for similar alloys which likely arise from different thermal-mechanical processing conditions. These models are developed with the primary purpose of informing experimental decisions. Additional experimental insight is necessary in order to reduce the number of experiments required to isolate ideal alloys. These models allow test planners to evaluate areas of experimental interest; once initial tests are conducted, the model can be updated and further improve follow-on testing decisions. The model also improves analysis capabilities by reducing the number of data points necessary from any particular test. For example, if one or two isotherms are measured during a test, the model can construct the rest of the TTT curve over a wide range of temperature and time. This modeling capability reduces the cost of experiments while also improving the value of the results from the tests. The reduced costs could result in improved material characterization and therefore improved fundamental understanding of TTT dynamics. As additional understanding of phenomena driving TTTs is acquired, this type of MLP model can be used to populate unknowns (such as material impurity and other thermal mechanical properties) from past literature sources.},
doi = {10.1016/j.jnucmat.2017.03.050},
journal = {Journal of Nuclear Materials},
number = ,
volume = 490,
place = {United States},
year = 2017,
month = 7
}
  • An accelerated learning algorithm (ABP--adaptive back propagation) is proposed for the supervised training of multilayer perceptron networks. The learning algorithm is inspired from the principle of forced dynamics'' for the total error functional. The algorithm updates the weights in the direction of steepest descent, but with a learning rate a specific function of the error and of the error gradient norm. This specific form of this function is chosen such as to accelerate convergence. Furthermore, ABP introduces no additional tuning'' parameters found in variants of the backpropagation algorithm. Simulation results indicate a superior convergence speed for analog problems only, asmore » compared to other competing methods, as well as reduced sensitivity to algorithm step size parameter variations.« less
  • This paper deals with the controversial topic of the selection of the parameters of a genetic algorithm, in this case hierarchical, used for training of multilayer perceptron neural networks for the binary classification. The parameters to select are the crossover and mutation probabilities of the control and parametric genes and the permanency percent. The results can be considered as a guide for using this kind of algorithm.
  • Electrical impedance data obtained with an array of conducting polymer chemical sensors was used by a neural network (ANN) to classify fuel adulteration. Real samples were classified with accuracy greater than 90% in two groups: approved and adulterated.
  • A new method for solving nonlinear programming problems within the framework of a multilayer neural network perceptron is proposed. The method employs the Penalty Function method to transform a constrained optimization problem into a sequence of unconstrained optimization problems and then solves the sequence of unconstrained optimizations of the transformed problem by training a series of multilayer perceptrons. The neural network formulation is represented in such a way that the multilayer perceptron prediction error to be minimized mimics the objective function of the unconstrained problem, and therefore, the minimization of the objective function for each unconstrained optimization is attained bymore » training a single perceptron. The multilayer perceptron allows for the transformation of problems with two-sided bounding constraints on the decision variables x, e.g., a{<=}x{sub n}{<=}b, into equivalent optimization problems in which these constraints do not explicitly appear. Hence, when these are the only constraints in the problem, the transformed problem is constraint free (i.e., the transformed objective function contains no penalty terms) and is solved by training a multilayer perceptron only once. In addition, we present a new Penalty Function method for solving nonlinear programming problems that is parameter free and guarantees that feasible solutions are obtained when the optimal solution is on the boundary of the feasible region. Simulation results, including an example from operations research, illustrate the proposed methods.« less