Automated Machine Learning Approach for Bottleneck Prediction in Manufacturing Systems

Bottleneck detection and prediction in manufacturing systems are critically important for enhancing operational efficiency and optimizing business resource utilization. Identifying bottlenecks minimizes disruptions in production processes and ensures effective capacity utilization. This study investigates the applicability of automated machine learning (AutoML) methods for bottleneck prediction in production processes. In the study, a dataset from nine machines in a manufacturing system was analyzed in a time-series approach. Forecasting models were developed using AutoML libraries such as EvalML (Evaluation Machine Learning), FLAML (Fast and Lightweight AutoML Library), Prophet, and TPOT (Tree-Based Pipeline Optimization Tool). The performance of these methods was compared to that of the deep learning method LSTM (Long Short-Term Memory). The findings indicate that AutoML libraries offer significant advantages in terms of rapid model development, ease of use, and high accuracy. Additionally, it was highlighted that different AutoML libraries can be optimized based on the dataset's characteristics and possess a broad application potential. AutoML methods not only improve prediction accuracy but also automate expert-driven processes, such as hyperparameter optimization and model creation, enabling these steps to be executed quickly and effectively. The results demonstrate that AutoML technologies hold great potential for wide-ranging applications not only in manufacturing but also across various industrial and commercial domains. This study provides a valuable reference for the use of AutoML methods and lays the groundwork for more comprehensive research in different sectors.