基于3σ准则与逻辑回归的片剂包衣终点判断方法研究

Study on endpoint determination method of tablet coating based on 3σ criteria and logic regression

该文通过探索衣膜厚度的统计分布特点,提出了一种基于3σ准则与逻辑回归的片剂包衣终点判断的新方法。首先,采集4个批次样品的近红外光谱,并用显微镜图像法测量衣膜厚度;其次,采用3σ准则得到合格包衣片的光谱区间,以测试样品在上述区间内的波数NI作为光谱特征;然后,由K折交叉验证考察模型的合格性判断性能,得到最佳建模条件与包衣终点的合格率判断阈值,建立基于3σ准则与逻辑回归的定性分析模型;最后,由上述模型计算测试集样品在不同时间的合格率,将其变化趋势与上述合格率判断阈值相结合,实现包衣终点判断。经KS检验得P=0.081>0.05,即合格片的衣膜厚度服从正态分布,可采用3σ准则确定衣膜厚度范围。而当包衣终点的合格率判断阈值设为90%时,模型对测试集的包衣终点判断正确率为100%。该研究揭示了包衣过程存在一定的随机性的现象,提出的方法具有准确性高、分析速度快、可解释性好的特点,为未来在线检测与质量评价提供参考。

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.