Applications Of Genetic Neural Network To Machine Process

How to select the rational process parameters to enhance the process precision of products is a research hotspot. Cutting force, grinding force and surface roughness of grinding are the most important parameters to scale machine progress, thus precision methods for them is needed in engineering.After analysis the defects among the traditional methods for predicting cutting force, grinding force and grinding surface roughness, two kinds of front technology—Artificial Neural Network and Genetic Algorithm are introduced, and all algorithms are programed by author with advanced programming language C#, then the theoretic predicting model of Genetic Neural Network is established. Through contrasted with the traditional predicting methods, it was proved that the present method can adapt itself to all kinds of work conditions and has higher flexibility and aptitude. At the same time, only through simply and directly model building and efficiently studying, it can reflect more exactly the varying relations between these physics quantities and their effect factors. Moreover, the method can be extended into predicting other homologous physics quantities in machine process. Several applications in cutting work and grinding work are provided to testify the new method's validity.In the other hand, to accelerate the operation velocity of GNN and improve the global optimization ability, many works were done in this dissertation to improve the normal Error Back Propagation Neural Network and Simple Genetic Algorithm. As for the former, author mends its activation function, introduces momentum item, confirms the topology of network, etc. For the latter, the premature constringency is restrained and the operation velocity is accelerated. Besides, makes use of GA to optimize the initial weights of BP Neural Network to remedy its defects—slow constringency velocity and trending to fall into local minmum. Compared with the normal algorithms or general ones, the operation efficiency of the present mended algorithms is enhanced greatly.