A Novel Deep Recurrent Neural Network for Quality Prediction of Cigarette Filter Rod

Ensuring the quality of cigarette filter rods is paramount in tobacco manufacturing, as defects can compromise product integrity and consumer safety. Traditional quality inspection methods often rely on manual sampling and rule-based systems, which may not effectively capture complex patterns in production data, leading to inefficiencies and potential oversights. To address these limitations, this study proposes a predictive model utilizing a Long Short-Term Memory (LSTM) neural network to forecast the quality of cigarette filter rods. LSTM architectures excel in processing time-series data and identifying time-based patterns, making them well-suited for modeling the evolving behaviors in filter rod manufacturing. The model is trained on a dataset encompassing various production parameters, including pressure drop, draw resistance, and ventilation measurements. While real-world production data is often proprietary, publicly available datasets such as the Cigarette Butt Dataset offer relevant features and serve as proxies for developing and testing predictive frameworks. Similar studies have utilized datasets focusing on quality records during production, excluding data from equipment startup and shutdown phases to avoid instability and ensure data reliability. The implemented LSTM model demonstrates significant improvements over traditional methods, achieving an accuracy of 98.8%, precision of 97.5%, recall of 97.9%, F1-score of 97.7%, and an AUC-ROC score of 0.99 in predicting filter rod quality outcomes. Additionally, the recommended model outperforms existing approaches such as ANN, RNN, and CNN not only in accuracy but also in computational efficiency, with reduced inference time facilitating real-time quality monitoring. These results underscore the potential of infusing deep learning (DL) approaches into quality control processes, offering a proactive, scalable, and time-efficient approach to defect detection and enhancing overall production throughput.