Practical aspects of using an expert system-neural network hybrid system for tuning crystal filters

A knowledge-based system in a rule format, has been developed in order to help an operator during the post-assembly tuning of crystal filters. The generation of the rules was accomplished using the ID3 learning by examples algorithm. A consultation with the system provides the operator with advice as to whether the filter is tuned or as to which screw to turn and in which direction. Unfortunately it was not possible to use ID3 to generate rules for the distance to turn. The distance can be any value in the range of 0 to 2.5 revolutions inclusive and ID3 cannot handle such a large number of classes (Tsaptsinos et al. (1)). It is therefore left to the operator to judge how far to turn the screw. Neural networks were investigated in order to provide the operator with an indication of how far to turn the screws. The results are given for one sub-process of tuning, namely the stopband tuning of the filter