ADs: Active Data-Selection for Data Quality Assurance in Data-Sharing Over Advanced Manufacturing Systems

Machine learning (ML) methods are widely used in manufacturing applications, which usually require a large amount of training data. However, data collection needs extensive costs and time investments in the manufacturing system, and data scarcity commonly exists. With the development of the industrial internet of things (IIoT), data-sharing is widely enabled among multiple machines with similar functionality to augment the dataset for building ML models. Despite the machines being designed similarly, the distribution mismatch inevitably exists in their data due to different working conditions, process parameters, measurement noise, etc. However, the effective application of ML methods is built upon the assumption that the training and testing data are sampled from the same distribution. In this work, we propose an Active Data-selection (ADs) framework to ensure the quality of the shared data among multiple machines. A novel acquisition function is developed for data selection, which integrates the information measure for benefiting downstream tasks and the similarity score for data quality assurance. The simulation study and case study are designed and conducted to validate the effectiveness of the proposed ADs framework. In the simulation study, data samples from different distributions are generated. In the case study, we collected real-world <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">in-situ</i> monitoring data. The results demonstrated that our ADs framework could intelligently share data samples from the target distribution while eliminating the out-of-distribution data samples when training ML methods. Our proposed framework can improve the performance of ML models by selecting a high-quality subset of the full dataset compared with the classical AL method. In addition, the model performance can be comparable to or even better than training with the fully labeled data.