Intelligent Control System for Cigarette Processing Based on Deep Learning

Abstract

∗An intelligent control system was proposed to draw resistance of the cigarette based on the deep learning method to solve the inaccurate control caused by using manual experience. The proposed system based on the Internet of Things platform (Thingworx) achieves millisecond-level single-cigarette data collection from cigarettemaking machines by designing a real-time high-speed data collection process, which a whole-process data interconnection of cigarette production is constructed. A real-time prediction system was developed for draw resistance of the cigarette based on a recurrent neural network. The intelligent control of draw resistance of the cigarette is realized by optimizing the optimal loss function in the rolling prediction domain. The shortcoming rate and the standard deviation per 10000 of draw resistance of the cigarette have been reduced through analyzing the statistical results. The simulation results show that the system can satisfy the current production needs and can perform real-time online intelligent control. The system runs stably and reliably, and high-speed data collection is complete, which improves the consistency of the draw of resistance and the quality of cigarettes.