Predicting servomechanism dynamic performance variation from limited production test data

Abstract

A probabilistic model describing the considerable variation in dynamic performance of a large family of electro-hydraulic servomechanisms is presented. Two fundamentally different techniques are employed to obtain the parameters of the model, both techniques giving very similar results. A third-order dominant transfer function (derived from the production test data) together with high frequency complex poles and a high frequency lag (both factors derived from field testing) form a sixth-order system. Variable parameters for the sixth-order system form the probabilistic model and are presented as a table for use in Monte Carlo simulation. The model is then used to derive experimentally verified production test limits in a new test domain, equivalent to the original frequency domain gates, showing how the use of a novel filtering technique, recently described elsewhere, may be employed to simplify the setting of test limits. The model is also shown to be of use in the prediction of dynamic errors in many domains (whilst only requiring data in one domain) and the initial production test data are shown to predict accurately the dynamic error variation for the complete family.