Verification Experiment of Data-driven Dynamic Quantizer

— Application to Motor Control Systems with Difficult-to-model Elements —

Abstract

In this paper, we discuss practical applicability of the data-driven designed dynamic quantizer (D4Q), proposed recently by Fujimoto and Minami. In many industrial control applications, the target systems only accept discrete-valued control inputs, mainly to suppress manufacturing cost, system complexity and network bandwidth. Quantizers, which converts control-valued inputs to discrete-valued ones, are thus required in those applications. One of a promising algorithm for this purpose is the so-called “optimal dynamic quantizer (ODQ)”; it minimizes the discrepancy between the outputs with and without discretization, based on the precise linear model of the target system. Then Fujimoto and Minami proposed a novel approach to design a dynamic quantizer, not from the system model, but directly from the input-output data. In this paper, we apply the D4Q design method to an electro-mechanical control system, possibly containing elements that are difficult to model, such as friction. Through the experimental results, we show that the D4Q is effective compared to the existing quantization methods, and is more robust against unmodelled element than the conventional ODQ.