Online Surface Roughness Detection of Cold-Rolled Strip Steel Based on Adaptive Regression Smooth Filtering

This paper presents an adaptive regression smoothing filtering method for on-line surface roughness detection of cold-rolled strip steel with high precision. This method overcomes the shortcomings of the conventional light scattering method using low-pass filter to identify the on-line inspection data of strip surface roughness and improve the accuracy of on-line testing of surface roughness. The initial parameters of on-line surface roughness detection are pre-treated by robust local weighted regression. Establishing the quadratic weight function, to achieve weight updated based on the relative position of historical data and the current data, and also based on the residual of the smoothing value and the actual value. When the adjacent data changes beyond the set threshold, the window width of the smoothing window is reduced automatically to realize the adaptive step-by-step regression smoothing filtering of the on-line roughness detection data. This improves the on-line inspection accuracy of cold-rolled strip surface roughness, and better meets the requirements of downstream users.