A predictive maintenance architecture for TFT-LCD manufacturing using machine learning on the cloud service

Abstract

The rise of Industry 4.0 has brought the world to intelligent manufacturing. The manufacturing industry combines technologies such as the Internet of Things, big data, and AI. Recent developments can further analyze equipment maintenance work by collecting real-time machine statuses, such as temperature and other parameter information. To achieve predictive machine maintenance, perform device maintenance and repair in advance to avoid unexpected downtime and affect production line operation. This paper will take the industry of TFT LCD panel component manufacturing as an experimental field and implement the predictive maintenance system of the TFT LCD machine through the Azure cloud service platform. First, the Pearson correlation was run to find a strong correlation for parameter training. In this case, Spark was used to reduce the processing time that initially took 2 h to 43 s and increase the speed by 99.4%. An optimization of the partition of the data table increased the operating cost, the IO cost, and the CPU cost by 98.77%, 98.78%, and 98.74%, respectively. Different training data and nodes are also compared to find excellent results. KNN, RF, XGBoost and SVM were compared to select a model that would be most suitable for use in the TFT LCD case. Finally, the results of the data and model analysis were visualized in real-time Azure Kubernetes scoring.