Abstract: The performance of complex products is always synthetically affected by multiple feature parameters. Thereby, it is the key to create a mapping between feature parameters and product performance for product performance prediction. It proposes a method of multiple parameters correlation analysis and prediction of product performance based on multi-criteria modification. In the proposed method, the product performance prediction is realized with four steps that are correcting the gross error of feature parameters, reducing the number of feature parameters, modifying the feature parameters by finite element method (FEM) and improving the product performance prediction model based on neural network, respectively. The gross error processing of feature parameters is implemented based on topological distance of the measured data, the key feature parameters, influencing product performance, are selected by associative analysis of gray entropy to reduce the complexity of the model. Following that, the feature parameters changed after assembling are modified by FEM. The product performance is predicted based on generalized regression neural network with AMESim. The validity of the proposed method is verified by an instance of performance prediction for a double nozzle baffle electro-hydraulic servo valve.

Key words: Associative analysis of gray entropy, Finite element, Gross error, Multi-criteria, Neural network, Performance prediction