A Hybrid Feedback Linearization and Neural Network Control Algorithm Applied to a Hydraulic Actuator

Abstract

This paper report a research investigation that proposes to replace the inversion set present in the traditional feedback linearization approach by an artificial neural network resulting in a hybrid composition approach with a neural network and an analytical term. The method is applied into a hydraulic actuator position system together with a friction compensation approach also built using neural networks. The control strategy used is based on a cascade methodology that consists of interpreting the hydraulic positioning system model as two interconnected subsystems: a mechanical subsystem driven by a hydraulic one. As experimental results have indicated a significant system behavior dependence on the oil temperature, its effects are also studied and the proposed method was improved by the inclusion of the oil temperature information as an input for the neural network functions. Experimental results show the effectiveness of the proposed controller and their advantages when compared with the traditional analytical schemes with feedback linearization approaches.

A neural network-based inversion method of a feedback linearization controller applied to a hydraulic actuator

“…Intelligent strategies, such as fuzzy logic or neural networks, have been widely used in many different applications, mainly because of their ability of "learning by themselves" how to adapt to their working environment. In the control of hydraulic actuators, such methods form the core of many algorithms [16][17][18][19][20][21][22], and many of them rely on fully online learning procedures to compensate for the unknown disturbances and unmodeled dynamics. Some works are based on radial basis function (RBF) neural networks [20,21], others use simplified feedforward multi-layer perceptron neural networks linearizated by means of Taylor's series expansion [19,23], an approach based on [24] and similar to that of several works in other areas, such as [25,26].…”

“…The effectiveness of the proposed strategy is demonstrated both analytically, with a rigorous stability proof using Lyapunov's second method, and in practice, with comprehensive experimental results. This approach was introduced in [16], being expanded here by including its main theoretical aspects: the Lyapunov stability proofs and an enhanced discussion on the development of the neural network.…”