Robust Self-Constructing Fuzzy Neural Network-Based Online Estimation for Industrial Product Quality

Abstract

Online estimation for product quality is crucial for improving industrial process efficiency. However, model degradation and outliers usually challenge industrial quality estimation models. To tackle the above problems, a robust self-constructing fuzzy neural network (RSC-FNN) is developed in the article. In the RSC-FNN, the rules can be automatically created or pruned, obtaining the online self-constructing mechanism (OSCM). First, an online error compensation algorithm is developed to generate new rules. Second, the model performance and contribution of existing rules are evaluated online to delete redundant rules. Thus, the OSCM effectively improves the structural adaptability and compactness of the RSC-FNN. Moreover, the correntropy-induced criterion, which can handle complex outliers, is modified for the parameter learning algorithm. Hence, the adverse effect of outliers can be suppressed during the parameter updating process. Besides, we analyze the convergence of the RSC-FNN to ensure its feasibility in industrial applications. Finally, two industrial applications are studied to test the effectiveness of the RSC-FNN. Compared with other FNN <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">s</sub> , the results indicate that the RSC-FNN performs better in learning efficiency, structure compactness, and robustness.