DIKUL - logo
(UL)
  • Towards intelligent control of electrical discharge machining
    Junkar, Mihael ; Valentinčič, Joško
    One of the basic problems in Electrical Discharge Machining (EDM) is to achieve stable process conditions. To avoid unstable processing and arcing, different control strategies can be applied. We ... have explored the possibilities of using the existing operators' and engineers' knowledge to build the EDM controller. In our paper we present the application of Machine Learning (ML) for the design of the EDM controller. The most successful form of ML has been learning from examples, also called inductive learning or "empirical learning". In learning from examples, the learning program generalizes the examples into general rules. In our case learning examples were taken from actions of the operator guiding the EDM system, and the obtained rules were used for building the EDM controller. Through examination of the EDM process, one notes that the time between the successive pulses has predominant effect on the conductivity of the dielectric in the gap, which consequently affects process stability. For this reason our EDM device was adapted so that the interval time, in addition to the gap and the flushing, became one of the control parameter and was included in the control algorithm as such. Experiments carried out by the upgraded controller demonstrated both process stability, detected by observing the in-process measured parameters (frequency of characteristic pulses), as well as good process performance, obtained by the evaluation of the off-line measured parameters (electrode wear, metal removal rate and surface roughness). Thus the integration of different sources of knowledge improved the EDM process control decisively. Additional advantage of the designed controller is its open architecture, allowing further upgrading by integration of new knowledge.
    Vrsta gradiva - članek, sestavni del
    Leto - 1999
    Jezik - angleški
    COBISS.SI-ID - 4887323