Optimization of Residual Stress of High Temperature Treatment Using Genetic Algorithm and Neural Network

Abstract

In a nuclear industry area, high temperature treatment of materials is a factor which requires special attention. Assessment needs to be conducted on the properties of the materials used, including the strength of the materials. The measurement of material properties under thermal processes may reflect residual stresses. The use of Genetic Algorithm (GA) to determine the optimal residual stress is one way to determine the strength of a material. In residual stress modeling with several parameters, it is sometimes difficult to solve for the optimal value through analytical or numerical calculations. Here, GA is an efficient algorithm which can generate the optimal values, both minima and maxima. The purposes of this research are to obtain the optimization of variable in residual stress models using GA and to predict the center of residual stress distribution, using fuzzy neural network (FNN) while the artificial neural network (ANN) used for modeling. In this work a single-material 316/316L stainless steel bar is modeled. The minimal residual stresses of the material at high temperatures were obtained with GA and analytical calculations. At a temperature of 650 0 C, the GA optimal residual stress estimation converged at -711.3689 MPa at a distance of 0.002934 mm from center point, whereas the analytical calculation result at that temperature and position is -975.556 MPa . At a temperature of 850 0 C, the GA result was -969.868 MPa at 0.002757 mm from the center point, while with analytical result was -1061.13 MPa. The difference in residual stress between GA and analytical results at a temperature of 650 o C is about 27%, while at 850 o C it is 8.67%. The distribution of residual stress showed a grouping concentrated around a coordinate of (-76; 76) MPa. The residuals stress model is a degree-two polynomial with coefficients of 50.33, -76.54, and -55.2, respectively, with a standard deviation of 7.874.

The Analysis of Optimal Crack Ratio for PWR Pressure Vessel Cladding Using Genetic Algorithm

“…To analyze the relationship of Stress Intensity Factor, crack ratio, and fracture toughness, the J-integral can be calculated numerically or using finite element method. Furthermore, for optimization purpose, the J-integral can be calculated using genetic algorithm [9,10]. In elastic plastic fracture mechanics, J-integral is used to evaluate the crack.…”

“…The results of the repetition of the above steps are expressed in a set of binary numbers or binary digits in the range of (binary digit/bit) as well as in the ranks of real numbers (the real). The algorithm ends when the maximum number of generations has been produced, or the best approximation level has been reached [9,10].…”

“…In this method, the selection probability of individuals depends on the fitness value, starting with the largest probability for the largest fitness value. The largest fitness value is taken most frequently so as to have the best chance to reproduce [10].…”

“…In addition, GA was applied to optimize crack detection for the beam structures [9]. In another study, the residual stress in high-temperature treatment was optimized using GA and Neural Network [10]. Finally, the mechanical fracture of PWR pressure vessel was evaluated and modeled based on neural network [11].…”