Optimization of particle acceleration parameters of special cold spray nozzles via neural network and genetic algorithm

Abstract

Cold spray technology is a new technology that deposits supersonic solid particles on the surface of materials. There are many factors that affect particle acceleration, such as the geometric structure of the cold spray nozzle (contraction section, throat, expansion section, special nozzle angle, etc.), the parameters of the propellant gas (gas type, gas temperature, gas pressure, etc.), the material properties of particles (metal and non-metal), particle size (generally 10...50 microns), and particle morphology (spherical, irregular shape, etc.). The objective of this study was to investigate the influence of particle acceleration parameters on cold spraying. This work aims to predict and optimize the particle velocity at the outlet of the special nozzle to meet the critical velocity requirements of various metal particles and thus meet the deposition conditions. The task to be solved is to optimize the particle parameters of the special nozzle and obtain the particle velocity at the outlet of the special nozzle. The methods used are as follows: Three key parameters that affect particle velocity were selected as research objects: helium gas temperature and pressure when selecting helium gas, and titanium particle diameter as the third parameter. First, 30 sets of particle exit velocity data were sampled using Latin Hypercube Sampling, of which 24 were training data and 6 were prediction data. Then, the neural network was analyzed to obtain the minimum neuronal error value, thereby determining the number of hidden layers. At the same time, the parameters were normalized, and finally, the nozzle exit particle parameters were optimized using genetic algorithm. The results showed that after three rounds of optimization and taking the average value, the particle velocity at the outlet of the special nozzle was 591m/s. The optimized parameters were: helium temperature of 694...865 K, a helium pressure of 3.3...3.7 MPa, and a titanium particle diameter of 12...20 microns. When the optimized parameters were input into the numerical simulation software, the result was close to the predicted value. Therefore, the neural network and genetic algorithm optimized the parameters with high accuracy (with an error of 4%) and can be used as a reference for relevant workers.