Predicting the precision of grinding process, especially correlating surface functionality generation to grinding conditions, would be of great significance to improve grinding accuracy of the end precision product. Huang developed a very promising revolutionary spectral data analysis technique based on the Hilbert transform. The concrete methods of the EMD, the local Hilbert spectrum are introduced. An artificial neural network (ANN) based on back propagation is developed to predict surface roughness Ra. An accelerometer is employed as in-process surface recognition sensor during grinding process to collect the vibration as input neurons. Changing the grinding condition, training and testing within the artificial neural networks to retrieve the weightings, the experimental results show that the proposed ANN surface recognition model is economical, efficient and the model has a high accuracy rate for predicting surface roughness.