A network-based model robustness improvement method for product quality assurance

Since real-time quality prediction is of great importance for preventing defects in manufactured products, it has gained lots of concerns. Data-driven prediction models are commonly used in this field, especially with the increase of available data. However, such methods are vulnerable to production perturbations, which would make the modeling data unmeasured or invalid, thus leading to low-accuracy quality prediction. To solve this problem, the paper designs a new data network-based approach for improving model robustness, considering interactive data relations. Advantages of the proposed method are verified in a case study by using data from a material drying production line. (C) 2022 CIRP. Published by Elsevier Ltd. All rights reserved.

Automated summary

Real-time quality prediction is crucial for defect prevention in manufacturing products, but it is vulnerable to production perturbations. This paper proposes a data network-based approach to enhance model robustness.