A generic method for servo tuning based on dynamic modeling and task description

Highlights

• This paper deals with a new tuning method adapted to robots and machines with regards to the task and structure specificities.

• This tuning method is validated experimentally on a vertical axis.

• The tuning method is based on the mechanical structure dynamic modeling.

• The dynamic parameters are identified under task solicitations to ensure reliability with the real behavior of the machine.

Abstract

In robotics or machining, task quality is strongly linked to servo tuning. Indeed, geometrical and dynamic behavior of the mechanical structure could be degraded when a coarse tuning is achieved. Several P/PI or PID tuning methods can be used but generally uncorrelated with industrial context or mechanical considerations. In this paper, we propose a generic tuning method based on dynamic modeling of the mechanical structure and model parameters experimental identification. Therefore, the proposed tuning is adapted to the real dynamic behavior of the mechanical structure under task load while being sufficiently simple to be implemented in an industrial context. The method is illustrated on a vertical axis.

Introduction

Control of industrial systems is a well-known and active field of research works. Thus, PID controller gain tuning has been already treated many times by the automatic control community [[1], [2], [3], [4], [5], [6]]. Many methods for gain tuning are employed from time domain tuning to frequency domain analysis, dealing with stability, robustness and digital control effects. However, in industrial context, systems are mainly tuned by empiric methods based on the experts skills without any modeling or computation. Thus, improvements are still expected [7]. Furthermore, there is an important gap between academic theoretical approaches and industrial needs increases.

For this work, the main applications are machine-tool or robotics feed drive based on P/PI position control. Thus, the classical tasks requires geometrical accuracy (steady state) and dynamic accuracy (tracking error) under high loads sollicitations. The global performances are thus linked to the mechanical behavior of the machine structure and servo tuning loaded with task sollicitations. On the one hand, automatic control works are based on high level control theory such as genetic algorithms [8,9], fuzzy control [10], robust control theory with classical models (first or second order with delay) [5,11,12]. These methods requires a high level of skills in modeling and computation. The theoretical concepts are often tested on dedicated controller in ideal conditions. Therefore, the versatility and adaptability to an industrial context could be questionable. On the other hand, mechanics community works on modeling and identification in order to simulate tasks or to plan trajectory generally without controller architecture considerations [[13], [14], [15]]. Last but not least, Electrical engineering works deal with actuators and drive tuning without task load considerations [16,17]. Thus, the task quality realized by a control system can be improved by a tuning method which takes into account control scheme, mechanical behavior of system structure and task load for a given technology of actuators. For example, in Ref. [6], the tuning is obtained with simple considerations. The PID tuning aims an over-damped second order with time delay system behavior. The rules used for PID tuning computation are applied to general systems with approximations and empirical considerations. However, the work focuses on general application and the tuning performances criteria are not adapted to machining or robotic tasks.

Our approach is to propose a tuning methodology based on mechanical modeling and model parameter identification coupled with simple theoretical tuning dealing with task quality. The aim is to develop a tuning method which takes into account the mechanical structure behavior under task load and the technological structure of the actual servo-drives to guarantee a task quality and a short implementation time. Moreover, our method should be simple to implement in an industrial context without controller modifications.

This article is organized as follows: a first part deals with the proposed tuning thus presenting method. A second part presents the experimental validation on a vertical axis.