Performance prediction of tobacco flavouring using response surface methodology and artificial neural network

• Published in: Journal of engineering (Stevenage, England) Vol. 2019; no. 13; pp. 367 - 372
• Main Authors: Chen, Lin, Yuan, Ruibo, Liu, Ze
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.01.2019; Wiley
• Subjects: 19th International Conference of Fluid Power and Mechatronic Control Engineering; artificial neural network; cigarette manufacturing; design of experiments; Experimental Design software; Expert Response Surface; feed liquid gas ejector flow; MATLAB software; neural nets; optimal solution configuration; pressure 3.7 bar; response data; Response Surface Analysis Method; response surface design; response surface methodology; roller speed; spice ejector flow; tobacco flavouring experiment; tobacco flow; tobacco industry; tobacco process flow; artificial neural network; tobacco process flow; tobacco flavouring experiment; cigarette manufacturing; MATLAB software; response surface methodology; Experimental Design software; tobacco industry; roller speed; optimal solution configuration; tobacco flow; spice ejector flow; response surface design; feed liquid gas ejector flow; pressure 3.7 bar; Response Surface Analysis Method; response data; Expert Response Surface; neural nets; design of experiments
• DOI: 10.1049/joe.2018.9038
• ISSN: 2051-3305

This study was to predict the optimum condition for leaf flavouring in cigarette manufacturing. To this purpose, an integrated research was used by using response surface and artificial neural network. A series of tobacco flavouring experiment's factors were designed by Experimental Design software. The MATLAB software's Neural Network function was used to forecast the responses, and the optimal solution configuration was coming out from the Response Surface Analysis Method. In the optimum condition, moisture removal opening, roller speed and tobacco process flow, pressure and feed liquid gas ejector flow are 18.60%, 10.74 rpm, 5314.11 kg/h, 3.70 bar and 243.63 kg/h, uniformity of the evaluation index and the utilization rate of material liquid distribution are 93.088% and 98.694%. With the corresponding experimental, results are consistent, under the condition of the error to less 7%, the test results show that through a few experimental data of predictive results of the neural network and response surface design has a certain practicability.