Application of Learning Vector Quantization (LVQ) in Selecting Mechanism Type in Mechanical Design

摘要

In conceptual design of mechanical systems, selecting an appropriate mechanism type that meets design requirements is a critical problem often encountered. A LVQ neural network based decision-making approach to mechanism type selection is proposed in this paper. Through learning from correct samples extracted from different mechanisms, expert knowledge is acquired and expressed in the form of weight matrix. Then this LVQ network can acts as a classifier and decisionmaker. When selecting mechanism type, through digitizing the design requirements, converting into a characteristic factor set, and fed into the trained LVQ network, a satisfactory mechanism can be automatically recognized from a range of mechanisms achieving a required kinematic function. Under the model, the problem of the accumulation and expression of expert knowledge can be effectively solved, the subjectivity of decision-making can be improved, and the quantitative evaluation of mechanism can be achieved. Compared with other neural networks, LVQ network possesses simpler network structure, faster learning rate, more reliable classification, and better fault tolerance. It is concluded that the LVQ network based decision model developed is appropriate to be used for selecting mechanism type at the early design stage.