Abstract: In order to conquer the difficulty of on-line measuring workpiece surface roughness, the surface roughness iden-tification method based on fuzzy-neural networks is put forward. As an example, the identification model of external cylindrical during grinding is built. Through the deep study of theory for-mulae and experimental formulae of external cylindrical surface roughness during grinding built before, it is known that the workpiece velocity, grinding wheel velocity, grinding depth and table feed have significant effect on the roughness. Further, the above grinding parameters measured on-line as the inputs of the identification model is considered. Since the grinding process is too complicated to be built exact mathematic model, the fuzzy-neural networks is introduced to the identification model. T-S type fuzzy inference is adopted to obtain the roughness because there exists linear relation between the roughness loga-rithm and the above grinding parameters logarithm. The model is used in the practical grinding process, and the model accu-racy is 97%. This verifies the feasibility of the proposed method.

Key words: External cylindrical grinding, Fuzzy-neural networks, On-line identification, Surface roughness