A Supervised Adaptive Resampling Monitoring Method for Quality Indicator in Time-Varying Process

In real-world industrial process, the system state is time-varying due to the changeable operating environment, and thus, a steady monitoring model is unsuitable in such a context. Benefiting from the development of sensor and computer technology, the process measurement variables can be collected in real time. In order to realize the online quality indicator monitoring by analyzing the related information in process variable, a supervised adaptive monitoring approach named adaptive resampling principal component regression (ArSPCR) is proposed in this article. First, the key process variables are selected based on mutual information. Second, the weight of each off-line sample is calculated according to its distance from the online sample, and the weighted normalization method is put forward. Considering the different weights of the off-line samples, a novel resampling technology is presented for constructing the modeling sample set, and a quality-related adaptive monitoring model is developed. Afterward, a model update strategy is introduced to balance the model accuracy and computation burden. Finally, the numerical case and Tennessee Eastman process are used to demonstrate the effectiveness of the proposed approach.