Predictive Modeling for Advanced Virtual Metrology: A Tree-Based Approach

Abstract

The rapid development of industry 4.0 has promoted the extensive adoption of big data analytics for manufacturing industry. In this domain, virtual metrology is a critical technique that is able to reduce manufacturing cost over a large amount of practical applications. In this paper, we propose a novel tree-based approach for simultaneous feature selection and predictive modeling to facilitate efficient virtual metrology. The proposed method accurately identifies multiple feature sets and then chooses the best candidate to minimize modeling error. As demonstrated by the experimental results based on two industrial examples, the proposed method can achieve higher modeling accuracy and find a more complete feature set than the conventional approach implemented with orthogonal matching pursuit (OMP).