This paper explores the statistical distribution characteristics of coating film thickness, so as to present a new method for determining coating endpoint based on 3σ criterion and logic regression. Firstly, the spectrum and thickness of 4 batch samples were collected. Secondly, the spectral range of normal products was obtained by 3σ criterion, with the spectral feature NI as the number of test spectrum in the above range. Then, the model based on 3σ criterion and logic regression was built according to the best condition in K-fold cross-validation and the determined threshold of qualified rate in the coating endpoint. Finally, the qualified rate of test set samples at different time points was calculated by the above model, and the above change trend and the threshold value were combined to determine the coating endpoint. The results of KS analysis showed the distribution of thickness of the qualified products followed the normal distribution(P=0.081>0.05). The accuracy of the coating endpoint determination was as high as 100% by the model based on 3σ criterion and logic regression when the determined threshold of qualified rate was 90%. Therefore, the 3σ criterion was feasible to the research of coating eligibility. This paper reveals certain random phenomena in the coating process, and the method features a high accuracy, quick analysis and a good interpretability, which provides a reference for online detection and qualify evaluation in future.
Keywords: 3σ criterion and logic regression; end point determination; near-infrared spectroscopy; statistic distribution.