Goal-Based Configuration Design of Fluid Power Systems Using Neural Networks

Abstract

Abstract In this paper, we present a machine learning goal-based methodology for producing configuration designs of fluid power systems. The goal is to produce solution(s) that provide the functionality desired by the user. This methodology allows the construction of an automated tool for producing the designs based on machine learning techniques, specifically artificial neural networks. In order to do this, the set of functional elements that constitute the user requirements must be recognised, as must the elements of the solution space. These sets are described in what is a necessarily context-sensitive method of producing valid mappings from the requirements set to the solution space. The neural networks are applied to extract and encapsulate the necessary design knowledge from examples of good designs, and in doing so, overcome one of the principal problems associated with previous attempts to build automated configuration design tools, namely that of knowledge acquisition. The problem of specifying the time-varying functional requirements was addressed by the adoption of a Time Delay Neural Network algorithm. Two examples are presented as design test cases. The results show that it is possible to construct a configuration design tool based on the methodology and incorporating domain-specific design knowledge in the form of trained neural networks.