Research On Automatic Identification Of Gun-Powder Dose Based On Genetic Algorithms And Neural Network

With the wide application of gun-powder, its quality and reliability are very important such as security problem and capability of use. So it is very important to insure the quality and enhance its reliability. In the traditional procedure of gun-powder product, the deflections such as overloaded, less loaded and no loaded are often occurred. Because its peculiarity, this deflection is difficultly discovered in time. This thing will influence the final capability of the product, even lead to very severe quality accident. So it is of quite consequential value to identify or measure the gun-powder timely in the pressing process of gun-powder after loading materials and to cull the non-eligible product.This paper proceeds the research on automatic identification of gun-powder dose in the pressing process. Considering the severe nonlinear peculiarity of the gun-powder product in the pressing process and the characteristic of sorts of gun-powder category, this paper adopts the neural network method to identify the dose of gun-powder product. To improve the velocity of neural network and avoid the minimal value, genetic algorithms as the training algorithm of neural network is adopted. This system can achieve the task of automatic identification of loading gun-powder dose.Firstly, this paper proceeds the experiment research on peculiarity of gun-powder, and some relevant conclusions are drawed. And then the issue of automatic identification of gun-powder product dose is described and the model algorithms that apply to this issue are expounded: the model system of automatic identification based on genetic algorithms and neural network.This paper ascertains the topology structure of BP network that applies to automatic identification of gun-powder dose: 1-5-1, and the relevant parameters are ascertained: population size N=80, AGA, =0.9,=0.6,=0.1,=0.001,crossover parameter =0.3; neural network parameter: (=0.84,(=0.58. When gun-powder category is different, the sample of training neural network needs change and network parameters , need adjustment in the training process according to the requirement of velocity. But the genetic algorithms parameter and training algorithm of neural network will keep the same.This paper makes use of the trained BP network to identify the dose of gun-powder product and related interpolation algorithms are used, and the better identificationresults are achieved.Experiment indicates that the application of the model system based on genetic algorithms and neural network to identify the gun-powder dose is a effective method, this system has better adaptation, convenient identification, well-extensible, real-time characteristics.