Font Size: a A A

The Methodology Of Discharging-state Identification In Micro Electrical Discharging Machining (MicroEDM)

Posted on:2007-03-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:M ZhouFull Text:PDF
GTID:1101360182982451Subject:Mechanical and electrical engineering
Abstract/Summary:PDF Full Text Request
Micro electrical discharge machining (micro EDM), one branch of EDM, focusing on micro machining, has received much attention around the world, led by the trend of micro mechanics development. There are still amounts of research work needed to be done in micro EDM for a stable machining with high efficiency and high precision. The stable control of the micro EDM process calls for precise discrimination of the discharging states in gap between electrode and work piece and building a prediction model reflecting the variation patterns of discharging states under micro EDM conditions. The requirement of high efficiency and high precision in electrical discharge machining involves the research work conducted on the optimization of parameters in EDM with multiple objectives such as machining rate, machining precision and tool consumption etc. Therefore, this thesis paper focuses its researches first on experiments with electrical discharge machining micro-and-small holes, the process of which is appropriate for the study of how to achieve a stable micro EDM process with high precision and high efficiency, and then on generalizing the results and setting up the models of interest from the experimental data.High frequency and low electrical power of micro EDM causes the waveforms of voltage and current highly distorted, thus indistinguishable by the conventional EDM discrimination systems due to the low ratio of signal to noise. Such uncertainty of signals in discrimination of discharging states really brought a new challenge in micro EDM field. From the observation of the signals of voltage and current, we found that the information contained in current is complementary to the information in voltage for discriminating the different discharging states. Thus a new intelligent discrimination system was established. This intelligent system consists of three modes, a fuzzy discrimination mode, a LVQ neural network mode and a fuzzy generalization mode. To solve the uncertainty in the sampled signals from the gap, the signals of voltage and current are taken simultaneously as inputs of the fuzzy discrimination mode. The inputs to fuzzy discrimination mode are then treated by the established fuzzy inference mechanism and the discharging state of the sampled signal is thus denoted by a value in one of the four intervals, eachrepresenting separately one of the four discharging states, namely short state, off state, spark/arc state and open state. The value then is fed into the LVQ neural network mode and converted into one of four vectors of four by one, each of which represents a different discharging state by defining the corresponding element being one and the other three being zeros in the vector. After the signals from a group of pulses are serially treated in the same way, summed one by one and finally divided by the number of the sampled signals, the elements in the vector represent the ratios of four discharging states. The ratios of short state and spark/arc state are then put into the final fuzzy generalization mode and the discharging state in the gap can be statistically generalized accordingly.The ratios of discharging states in the vectors, corresponding to the settings of parameters (voltage, current, pulse on time, pulse off time, servo reference voltage and electrode up-and-down time), in an experiment with orthogonal array can be easily gained from this intelligent micro EDM discrimination system. With simple manipulations of the ratios of discharging states in the vectors in terms of orthogonal properties, the trend lines of the ratios of discharging states with each varied parameter can be formed. By analyzing the trend lines of short ratio, spark/arc ration and open ration, we found the way of how a varied parameter shapes the ratios of discharging states and hence come up with principles of setting the discharging parameters (voltage, current and pulse on time) and the adjustment of the process controlling parameters (pulse off time, servo reference voltage and electrode up-and-down time) in terms of performance characteristics. Then the quantification of such relationship between discharging states and the parameter settings followed. After the models of linear regression and ARX010 created and the loss functions compared by fitting the data of discharging states and the parameter settings in these models, a conclusion is reached that the ARX010 model is better in describing this relationship than the model of linear regression. Further study of the loss functions of the above models shows that the big loss functions in these models largely originate from the fact that the errors between the simulated outputs of the models and the real values from the intelligent EDM discrimination system are really not unbiased. In this case, the paper proposed to use the Instrumental-Variable method, by which the reiterative application of LS method to ARX model results in an unbiased guess of the coefficients in the model, and thus obtained an IV010 model with its loss function less than the half of the loss function from ARXO1O model. If we think EDM process as a dynamic system, because of lack of full knowledge of its internal machining mechanism, this system can be regarded as a black box. One order, second order and even more orders of IV models have been tried to describe this black box by fitting some of the experimental dataof discharging states to these models and comparing the predictions of these models with the other experimental data of discharging states. The results showed that the model IV110 is the most appropriate to describe the relation between discharging states and the parameter settings than the other models mentioned above.In this thesis paper, an experiment with orthogonal array was carried out to investigate the effects of such parameters as voltage, current, pulse on time and pulse off time, upon the individual performance characteristic. Through the arrangement of guch kind of experiment, a full knowledge of the implementations on each of the performance characteristics like electrode consumption, holes clearance and machining rate, etc. can be obtained. Further analysis of the experimental data found that the different emphasis on one of the performance characteristics leads to a quite different combination of the parameter settings involved, and some times may even lead to a quite poor performance for another characteristic, that is to say, that the meet of one performance characteristic with optimized parameter settings may induce the sacrifice of another performance characteristic. In such cases an inter-culture approach was tried out to balance this contradiction. The theory of grey relational analysis was used to study the geometrically corresponding relationships among these characteristics. The grey relational coefficients from the manipulations of the experimental data in terms of grey relational formulations denote the degree of geometric approaches for an individual performance characteristic under different parameter settings in the experiment and hence dictate the optimized parameter settings for an individual performance characteristic with more specified and flexible way than that in the conclusion from the manipulations of the experiment data in terms of orthogonal properties only. A grey relational grade, obtained by averaging the grey relational coefficients, is used to evaluate the performance with multiple characteristics;that is, the optimization of the parameter settings with multiple performance characteristics can be transformed into the maximization of the grey relational grades and the biggest grade for each parameter dictates the best arrangement of parameters corresponding to it. Thus, the optimization method and the scheme of parameter settings with multiple performance characteristics can be gained.The property of high frequency in micro EDM makes the discharging states varied much faster than in conventional EDM, so that the controlling schemes in conventional EDM can not be used in micro EDM for real and precise control, since the lagging of control, though working normally in conventional EDM, may cause the controlling system unstable in micro EDM or even causes it breakdown. As with such problem, we proposed to build a mathematic model of discharging states based on the observed data from EDM discrimination systemwhich monitors the machining process. The model is able to provide the data of discharging states one step ahead or more steps ahead with prescribed precision. The control system adapts its behavior in response to the ahead-of-time data such that the lagging of control can be avoided and therefore it makes sense to consider the stable control in micro EDM.Then follow the research work on how to build such a model which can be used for real and precise control in micro EDM. Normally the relationship between the discharging states and the parameter settings is that of multiple inputs and single output. A fundamental weakness of the multivariable system is that as the input space dimension increases the degree of sophistication of the model also increases and the calculations of identification algorithms on the parameters in the model increase potentially, limiting the application of the model in real-time control. From a pragmatic attitude, the model should be generated with a simpler structure, while still retaining the inherent relationships. We propose to use the initial value of the discharging state as input, determined by the parameter, settings, and use timely varied values of discharging states as output. Based on the observed data, this single input and single output (SISO) relation after spectral analysis proposes many models with different structures and different orders. After the comparisons of outputs of these models with the original data in time and frequency domains and by the analysis of zero and pole distribution in these models, BJ11321 is finally selected as the right model to describe dynamic discharging system. Once the structure of the model BJ11321 has been determined, it is applied to real, online prediction. By comparisons of the prediction data and the real data, the ratio of averaged error to the averaged value of online discharging states is odouj^d f>Yt , which qualifies this model structure being used in the real, online control with prescribed precision.
Keywords/Search Tags:electrical discharge machining, micro electrical discharge machining, discrimination of discharging states, parameter optimization, system identification
PDF Full Text Request
Related items