Optimizing Bottle Filling Efficiency Through a Machine Learning Approach: a Case of National Alcohol and Liquor Factory

No Thumbnail Available

Date

2026-02

Journal Title

Journal ISSN

Volume Title

Publisher

Addis Ababa University

Abstract

The bottle filling process is an important process in beverage production since it has direct influence on the quality of the product, compliance with regulations, and the efficiency of production. Nonetheless, major deviations on the bottle filling process are noticed in the National Alcohol and Liquor Factory (NALF). Records of inspections of 35 production days in January- February 2025 found that 27,972 bottles were underfilled and 18,648 overfilled, with an average of about 1332 bottles in need of reprocessing per day. These variations raise manpower, energy use, equipment depreciation, operation expenses and lowers the total output. The factory used in the 2024 annual performance report has reached an output of 85 percent of the intended production capacity, which indicates the necessity of the better monitoring and control of the process. In this study, the aim was to maximize the efficiency of bottle filling based on a supervised machine learning approach. The study involved both quantitative and qualitative research methods using both exploratory and experimental research designs. Primary data were gathered with the help of structured interviews, structured discussions with experts of the company, and direct observations of the filling process, whereas secondary data contained records of maintenance, production records, and reports on quality inspection. A supervised machine learning approach was used in this investigation to identify filling results. A Random Forest classifier was used to distinguish filling outcomes as underfill, correct fill, or overfill. The model had a training accuracy of 99.6% and a testing accuracy of 99.3%. A macro and weighted average performance and reliability were used to validate model performance and reliability, and out-of-bag error estimation and stratified k-fold cross-validation were used to validate. To facilitate real-time monitoring and dynamic process control, the developed model was coupled with the filling process by use of a three-layer architecture. The conclusions reveal that machine learning can be used to enhance the stability of the process, minimize deviations during filling, and decrease the use of resources, as well as assist in making decisions that are supported by data. This Study adds a viable machine learning framework to enhance filling efficiency and promote intelligent manufacturing behaviors in the production of beverages

Description

Keywords

Bottle filling process, beverage manufacturing, Machine Learning, Random forest classification, Process optimization

Citation