An Optimised Light Gradient Boosting Machine Model for Setup Time Prediction in Electronic Production Lines

Setup time is pivotal in printed circuit board (PCB) assembly line operations. However, PCB production encounters varying setup times due to multiple influencing factors. This paper addresses an uncertain setup time prediction problem in PCB assembly production lines. Unlike existing production time prediction models, our proposed approach integrates a comprehensive range of production features, not only with features related to PCBs but also production line operators, setup procedures and so on. To enhance model accuracy and mitigate overfitting, we implemented some data preprocessing phases and designed a random forest-integrated feature selection method. With the selected features, we used a light gradient boosting machine (LightGBM) as the predictive model and optimised its hyperparameters by a differential evolution (DE) algorithm. We validated our model's performance through extensive computational experiments based on real-world industrial data, focusing on feature selection efficiency and hyperparameter optimisation. The experimental results confirmed that our proposed DE-LightGBM can reduce redundant features and optimise the integral hyperparameters for model training. We also compared the DE-LightGBM model to some well-established machine learning approaches in different setup scenarios. The proposed DE-LightGBM outperformed other machine learning methods being compared, delivering accurate setup time predictions in both standard and complex scenarios.