Silicon Carbide Surface Quality Prediction Based on Artificial Intelligence Methods on Multi-sensor Fusion Detection Test Platform

On the basis of the grinding process experiments for SiC ceramic workpiece, grinding process parameters are measured on the multi-sensor fusion detection test platform and experimental results are analyzed. Lempel-Ziv complexity (LZC) is introduced to reflect the integrated grinding process stability due to kinds of factors such as vibration from grinding machine parts and noise from experimental platform. The greater the LZC is, the fewer period factors are in the grinding process, which reflect the nonlinear correlations in the grinding process impacting on grinding process. A method is given based on LZC for analyzing grinding process stability. Under consideration of experimental results, a predictive model for surface quality is given by the Kernel Principle Component Analysis and Modified Extreme learning machine method (KPCA-MELM), and grinding process parameters can be optimized too. KPCA-MELM predictive model overcomes disadvantages of MELM predictive model of the randomness of weight omega and threshold value b by introducing improved genetic algorithm, which makes the roughness predictive error more accurate with the maximum error of 4.803%.