Evaluating Manufacturing Systems with Machine Learning Techniques
Loading...
Date
2025-02-04
Authors
Firas Mabroukeh
Basel Al-Sahili
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Abstract:
In the domination of mass production systems, optimizing efficiency is essential for sustainable operation and competitive advantage. However, the complex interdependencies in sequential operations often lead to high variability in output parameters, create challenges for accurate prediction. This project proposes an approach to address this issue by integrating simulation and machine learning techniques to evaluate manufacturing systems efficiency. The project was divided into two phases, phase one was focused on the development of an accurate machine learning model capable of predicting queuing systems output parameters with precision, this model was trained on data generated through simulation, predictions made were compared with theoretical values of queuing systems equations, phase one came to a conclusion that machine learning models and simulations effectively predict and analyze queuing system behaviors, offering a robust integrated toolset with wide-ranging applications. In phase two, the study of manufacturing systems expanded to include serial production lines, these systems lack evaluation formulas. Utilizing five Tree-Based Machine learning algorithms, models were trained and evaluated using simulated data. The study employed Random Forest (RF), Extra Trees (ET), Gradient boosting decision tree (GBDT), XGBoost, and LightGBM models for prediction. Evaluation metrics such as the Coefficient of Determination (r2) and Mean Absolute Percentage Error (MAPE) were used to compare between the models. The findings from phase two underscore the capability of Tree-Based Machine Learning models in handling the complexities of serial production lines, where traditional analytical approaches falter. Among the models tested, Gradient Boosting XGBoost, and LightGBM demonstrated superior performance in predicting key metrics such as average throughput and buffer levels, achieving high R² values and low MAPE across various configurations. However, Tree-Based Machine Learning models struggled to predict Throughput Standard Deviation accurately. Future studies could explore whether this is due to the training data being too small or lacking variety, or if tree-based models are simply not well-suited for predicting this type of output. This could involve creating larger and more detailed datasets or testing other machine learning methods better suited for this task.