Neuro-fuzzy Process Control System for Sinking Edm

Electrical-Discharge machining (EDM) stands for highly accurate and very sophisticated metal shaping. The physical process takes advantage of electric field effects between two electrodes, a tool and a workpiece. Material is removed by sequences of electrical discharges. The electrically efficient gap width, which is determined among other parameters by the electric conductivity of the gap and the geometrical distance of the electrodes varies from spark to spark. This leads to a highly nonlinear control problem. Various optimization control algorithms have been developed to improve the performance of EDM sinking machines. Soft computing technologies like Fuzzy logic and Neural Networks gained much popularity in this field. The following article introduces a process control system consisting of a Fuzzy gap-width controller adapted by a Neural Network. By combining a Neural Network with a Fuzzy controller in this way a learning process control system is achieved. Experimental results show the working efficiency of this Neuro-Fuzzy system. INTRODUCTION The operating efficiency of an ED-machine strongly depends on the gap-width controller. This controller has to adjust the size of the interelectrode gap in order to achieve a high workpiece removal rate and a low tool-electrode wear. A model based controller synthesis cannot completely be applied in EDM, because of the insufficient model of the removal system and the chaotic fluctuations of the gap state. To cope with these problems optimization systems have been developed to adapt the gapwidth controller to the continuously changing working conditions and process situations. Most of these systems operate by systematic controller parameter variation using heuristic methods [4], [5], and [12]. Fuzzy technologies follow a different approach. Fuzzy-controllers are designed by integrating the experience of human users in so-called membership functions and rulebases. In EDM Fuzzy-technologies gained much popularity. In numerous research projects it has been shown, that efficiency of ED-machines can be improved by using Fuzzy gap-width controllers [13], [16], [17]. Commercial EDM sinkers with Fuzzy technologies are already available at the market today [11]. For the construction of these Fuzzy controllers optimization algorithms [17] or Neural Networks [13] are often used. In the second case a Neural Network creates automatically on the basis of the desired controller output the membership functions and rulebases of a Fuzzy-Controller. In this context the term "Neuro-Fuzzy" is often used [9]. In fact the role of the Neural Network is limited to the controller development phase only. The Fuzzy gap-width controller carries out all control actions. In another approach the Neural Network is used for the online adaptation of the running Fuzzygap-width controller. In this case the Neural Network becomes a vital part of the process control system itself. In comparison to the above-mentioned “Neuro-Fuzzy” approach the actions of the Neural Network are not limited to the controller development phase any more. By including the Neural Network into the EDM gapwidth control system, the ability of learning is available and working efficiency can be improved. The work presented in this paper follows this approach. A Neural Network adapts the output of a fuzzy-gap-width controller based on the actual process situation measured and computed online. FUZZY GAP-WIDTH CONTROLLER For the multiple cases in which ED-machining is used in practice the gap width controller has to provide a stable and efficient removal process. The electrical efficient gap width is difficult to measure, because of surface roughness and the distribution of the dielectric properties in the gap. Therefore the electrical efficient gap width itself can not be used for controller feedback. Most of modern ED-machines make use of the ignition delay time (td) as input value for the gap width controller [12]. It is easy to measure and td can characterize the most important kinds of ignitions. The disadvantage of td for control operations is its considerable variance. As consequence of this fact a flutter movement of the electrode often may be observed. Because of the very small geometrical gap width this flutter movement can lead to instable process conditions [10]. For this reason the implemented gap-width-controller uses alternative input values. The relative frequency of short circuits and open circuits during an inspected period were used as input parameters. A highly efficient removal process shows a very low number of these pulses [1]. This gap-width-controller is implemented using Fuzzy-technologies. The output of this controller is multiplied by scaling factors to compute the target-speed of the tool-electrode. Different scaling factors can be defined for forward and backward movement. Figure 1 shows the Fuzzygap-width controller. Fuzzy-Controller rel. freq. of open-circuits rel. freq. of short-circuits SV down direction ? electrode speed