Data-Driven Prediction of Key Attributes for Tobacco Products.

Draw resistance of the cigarette is one of the critical attributes of tobacco products. It directly affects consumers' health and has a close relationship with the release of cigarette tar, smokes nicotine, and carbon monoxide volume. Therefore, to better monitor the cigarette draw resistance, we have developed a monitoring system for forecasting and tracking the cigarette draw resistance instead of random sampling by a manual that is time-consuming for the security inspector and low accuracy. To realize the interconnection and fusion of the multi-sensors data of the production process with the upper production process data, we used "PTC Thingworx" as an IIoT platform, and designed a cigarette data collection program for getting big and high-speed real-time production data. We performed pre-processing data work in terms of the customized time window for monitoring. We first introduce recurrent neural networks (RNNs) in the field to forecast-based multi-sensor data fusion approaches. It can be seen from the test of various metrics on datasets that the method proposed in this paper has strong robustness and generalizability. The technique can completely replace labor and has an outstanding contribution to replacing the monitors.