A Development Of Simulation System Of Insulation Ceramic Si₃N₄EDM Based On ANN

The ceramic material has unique functions in aspect of the electricity, magnetism and so on. It also has the characteristics of high hardness, the wearability, ability of resisting heat and high pressure, which is considered to be one of leading materials which advance industries progresses in the21st century. But it is very difficult to machine and form it shaped. Traditional processing method is being grinded by diamond after burning to form. But it is not only inefficient but expensive in this way, which influences generalization and application of ceramics material. In recent twenty years, the application of EDM in ceramic material has expanded rapidly and EDM of electric ceramics material has become applied.EDM is a non-linear system and it is difficult to describe its technologic disciplinarian, so it is the developing trend to study the processing characteristic and simulation. Artificial Neural Network (ANN), as a new technology, uses computer to carry out some kind of simplification and simulation on function of human being’s cerebrum. It composes a super-large-scale non-linear continuous time driving force system, which provides effective way for the in-depth study of EDM.Compared with other processing method, there are many effect factors and a lot of certain or uncertain factors in the process of EDM. These factors contain polarity of pulse electrical source, pulse on time, pulse off time, peak current, spark area of electrode, processing depth etc. They are close connection with machining effect such as machining rate, machining precision and electrode consumption, which are also important technologic parameters. In actual process of produce and machining, the expectant effect can achieve only when the handlers have rich experience. However, the product is being developed towards high precision, high machining efficiency and diversification of material, which brought forward higher request to handlers. Sometimes, because of the deficiency of the handlers’experience, the capability and function of the equipment can not get sufficient exertion, which brings much waste of resources. Be dead against this condition, the most important problem that should be solved are forecasting of technological effect, optimization of machining parameters and development of intelligent and simulating software in this paper. By using the software, the handlers can forecast the machining effect easily and decide the optimization machining condition according to different machining request.Before proposing the forecasting model of performance effect of EDM insulating ceramics, the experiment should be done to explore factors that influence the performance effect. Moreover, the learning sample that trains ANN can be obtained from much experiment. In this paper, though choosing the experimental results as the learning sample, the performance predictive model of EDM insulating ceramics is proposed, with the BP algorithm of ANN. With choosing the cross section area of electrode, peak current, pulse on time, pulse off time as the inputs, and choosing machining rate, surface roughness as the output parameters, the proposed model is simulated. The comparative results show that the predictable model based on BP algorithm coincides well with optimization combination with multiple performance objectives. Moreover, the proposed predictive model can predict well the experimental observations of EMD insulating ceramics, which provides the important theory bases for optimization control of machining process of EDM insulating ceramics.To get the optimization of electric parameters, the operators always want to obtain the machining rate as fast as possible and make sure the surface roughness is as small as possible. But the fact is that the faster the machining rate is, the bigger the surface roughness will be. So it seems incompatible to make the two targets come true at the same time. In this paper, the machining rate is chosen as target function and other index will be given as restriction conditions. Because if the surface roughness is chosen as target function to minimize, there may be an unacceptable optimization project:the surface roughness is small but the machining efficiency is very low, which is unnecessary in factual production. Moreover, the target function is an ANN model not a specific mathematical expression, so it is not fit to adopt the traditional optimization method to get optimization by computing differential coefficient. Genetic Algorithms (GA) needn’t do deep mathematical analysis to the character of optimization problem and have the ability of global search, which is very fit to solve the problem of the optimization of electric parameters. The ANN and GA are adopted to solve the problem of the optimization of electric parameters of EDM insulating ceramics in this paper. From the results of system simulation, it indicates that this method has high efficiency in search, precision and good practicability.The simulation system of EDM insulating ceramics is composed of following components or sub-models:management of system date, forecasting of technologic effect, optimization of electric parameters and interface. The program relating to numeric calculations is accomplished by software MATLAB and the interface is designed by VC++6.0. The reason is that MATLAB is a tool that can develop the whole program and deal with dates, but MATLAB is only an elucidative executive language and the execution speed is very slow. The operation function of users’interface offered by MATLAB is too simple, so it is unsuitable for development of big software, especially for the basal I/O. Therefore, it is necessary to consider using VC++to design the interface, and then run MATLAB in VC++by engine API. Finally, the two types of program language can be mixed well in the development of the same software.In order to make the communication between the operators and software, a slinky and appropriate interface was also developed. The software is designed as a system having reasonable structure, friendly interface and strong maneuverability. Because the learning samples are all saved in database and can be updated in time, the software is able to transplanted and embedded in other different software. Furthermore, a few promising ways to improve the prediction and controlling of model also have been discussed.