Estimation of time-dependent laser heat flux distribution based on BPNN improved by multiple population genetic algorithm

This study presents a method for estimating the time-dependent laser heat flux distribution of a nonlinear heat conduction system using a Back Propagation Neural Network improved by Multiple Population Genetic Algorithm (MPGA-BPNN) from transient temperature measurements. The primary focus of this research was on establishing a two-dimensional heat transfer model and training the MPGA-BPNN based on temperature distributions derived from finite volume simulations and any known heat flux. A series of numerical simulations were conducted to verify the feasibility of the proposed method, with corresponding experiments designed to test the method's efficacy. A comparative analysis was also conducted among BPNN, GA-BPNN and MPGA-BPNN methods. When the actual heat flux value is 4 × 10⁵ W/m², the relative errors for the BPNN, GA-BPNN, and MPGA-BPNN algorithms are 6.5%, 5.0%, and 1.0%, respectively. Furthermore, the MPGA-BPNN method has shown significant predictive capabilities in reliability and accuracy analyses across a range of cases. The study investigated the influence of network parameters and input temperatures on the heat flux outcomes using the MPGA-BPNN method. The heat flux of the laser on the sample surface was experimentally measured using this method, along with temperature distributions at 1.0–5.0 s obtained via a thermal imager.