Data-Driven Quality Monitoring Model for Multi Variety and Small Batch Production Process in Cigarette Spinning

Abstract

Since the application of statistical process control (SPC) in production support, the manufacturing field has saved a lot of costs. In recent years, the combination of SPC with multivariate statistics and machine learning has been applied to the detection and diagnosis of industrial process operation and production results. It is one of the most popular fields in the research and practice of statistics. In multivariable quality control (MQC), the use of multivariate statistics for anomaly detection has been deeply studied. The general method is to use the principal component analysis model. Most MQC literatures mainly focus on the monitoring of quality variables and the detection of quality problems. However, conventional MQC methods are difficult to determine the process variables that lead to quality problems. For the multi-variable, non-linear and large delay characteristics of cigarette silk production data, the final product quality control is a very challenging task. This paper proposes a data-driven modeling method for the real-time final product quality prediction model in the multi-variety and small batch process operation.