基于近红外光谱结合多变量统计过程控制的一种针对混合过程误投情况的判断方法

A Judgment Method for Detecting Incorrect Input Situations in a Mixed Process Based on Near-Infrared Spectroscopy Combined with Multivariate Statistical Process Control

摘要: 目的：本研究旨在建立一种近红外光谱结合多变量统计控制判断混合生产过程误投的方法。方法：利用近红外光谱仪对混合过程进行光谱扫描，对其光谱进行预处理和波段选择，并结合多变量统计过程控制(Multivariate Statistical Process Control, MSPC)建立混合过程在线监测模型。结果：所建立的相似度匹配值(Similarity match value, SMV)模型波段选取1304.54~1502.76 nm，采用二阶导数(Second derivative)光谱预处理方式，判定阈值为0.9960。以40批正常批次建模，经30批正常生产批次及4批模拟误投批次(含漏投与±5%投料偏差)验证，模型对正常批次与误投批次均实现了正确判定，无漏报与误报。结论：该研究开发的混合过程误投判断方法稳定、可靠，可实现生产中混合过程状态的快速判断方法。

Abstract: Objective: This study aims to establish a method that combines near-infrared spectroscopy with multivariate statistical control to judge the mis-feeding errors in the mixed production process. Methods: Using a near-infrared spectrometer to conduct spectral scanning during the mixing process, pre-processing and band selection of the spectra are carried out, and an online monitoring model for the mixing process is established by combining multivariate statistical process control (MSPC). Results: The established similarity match value (SMV) model selected the wavelength range of 1304.54~1502.76 nm, adopted the second derivative spectral preprocessing method, and set the judgment threshold at 0.9960. Using 40 normal batches for model building and 30 normal production batches and 4 simulated mis-feeding batches (including drug omission and ±5% dosing deviation) for validation, the model correctly judged both normal and mis-feeding batches without false negatives or false positives. Conclusion: The hybrid process mischarge judgment method developed in this study is stable and reliable, and can enable rapid judgment of the state of the mixing process in production.