Optimized detection of tar content in the manufacturing process using adaptive neuro-fuzzy inference systems

Area of Science:

• Analytical Chemistry
• Computational Intelligence
• Tobacco Science

Background:

• Cigarette manufacturing involves complex chemical processes.
• Accurate detection of tar and harmful compounds is crucial for product regulation and safety.
• Existing methods for tar detection can be time-consuming or lack real-time processing capabilities.

Purpose of the Study:

• To model and optimize the detection of cigarette tar during manufacturing.
• To compare the efficacy of an Adaptive Neuro-Fuzzy Inference System (ANFIS) with High-Performance Liquid Chromatography (HPLC) for tar detection.
• To investigate the relationship between tar yield, nicotine, and Benzene, Toluene, Xylene (BTX) levels in cigarettes.

Main Methods:

• Development of a neuro-fuzzy system, specifically an Adaptive Neuro-Fuzzy Inference System (ANFIS), for tar detection.
• Utilizing five basic features as inputs for ANFIS classifiers to detect tar in smoke condensate.
• Application of High-Performance Liquid Chromatography (HPLC) as a classical benchmark method for comparison.
• Comparative performance analysis of ANFIS and HPLC methods.

Main Results:

• ANFIS classifiers successfully learned to differentiate new cases based on input features for tar detection.
• Low yield cigarettes exhibited similar nicotine levels compared to high yield cigarettes.
• Benzene, Toluene, and Xylene (BTX) levels were found to increase proportionally with increasing tar yields.

Conclusions:

• The ANFIS model provides an effective approach for modeling and optimizing cigarette tar detection.
• The study highlights a significant correlation between increased tar yield and elevated BTX levels.
• ANFIS demonstrates potential as an efficient alternative or complementary method to HPLC for real-time tar analysis in cigarette manufacturing.