A New Approach to Mechanism Type Selection by Using Back-Propagation Neural Networks

Mechanism type selection is a critical problem often encountered in conceptual design stage of mechanical system. A BP neural network based approach to mechanism type selection is proposed, which capitalizes on the features of nonlinearity, self-organization, and fault tolerance of a neural network to implement classification and selection. By using appropriate data sets to train the neural network repeatedly, expertise is acquired and expressed using a trained weight and threshold matrix. Thus the neural network can reflect not only the weight given to each evaluation index, but also the relationship between the characteristic attribute value of a mechanism and the final decision result. When the design requirements are fuzzily quantified, converted into characteristic factor, and fed into the trained BP neural network, an appropriate mechanism can be selected from a list of mechanisms achieving a required function. Under this approach, the problem of the expression and accumulation for expert knowledge can be effectively solved, and the quantitative evaluation for mechanism type is achieved. In a sense, the selecting model can be used as a substitute for decision-making population to carry out evaluation. An engineering example is presented that illustrates the effectiveness of the proposed method.