Sensitivity analysis of CHF parameters under flow instability by using a neural network method

Highlights: • We constructed the predicting model of CHF based on BP neural network. • We found the sensitivity coefficients of different parameters. • We got the comprehensive effect influence of different parameters to CHF. - Abstract: Construct the predicting model of CHF based on BP neural network. The sensitivity coefficients of different parameters could be calculated by solving partial differential of the predicting model. With the method of neural network connection weight sensitivity analysis and the data from other researchers' experiments, the sensitivity of different factors to the critical heat flux (CHF) is analyzed. The result shows that, ΔGmax/G0 has the largest sensitivity coefficients to CHF and the inlet temperature has the smallest sensitivity coefficients in the test range. The sensitivity of ΔGmax/G0 could be 20 times of that of the inlet temperature. The BP predictions of CHF fit well with the experimental data, and the errors fall in the margin of 5%. The BP predictions of the influences of ΔGmax/G0 and τ to CFm fit well with Kim's formula, and the largest error is 12.5%