Abstract: The changing trend of surface hardness with process parameters was analyzed， and the surface hardness was tested by grinding low expansion glass in quick-point. Based on BP neural network and single factor tests in quick-point grinding， a series of one-dimensional models were built for surface hardness and process parameters by the least-squares fitting. The accuracy of the model was tested by coefficient of correlation. The results show that the model has high accuracy. The multivariate models about surface hardness and process parameters were proposed after analyzing one-dimensional models. Based on the genetic algorithm， the multivariate numerical models were built for surface hardness according to the results of orthogonal experiments. The accuracy of multivariate model was tested by the verification experiment. The test results indicate that the model has high accuracy.

Key words: surface hardness, numerical fitting, BP neural network, genetic algorithm, point grinding, glass ceramics