Data-driven quality related prediction and monitoring

The quality or key performance indicator related prediction and diagnosis cover a wide range of practical requirements from industrial applications. Although much effort has been devoted to establishing an analytical model between operating conditions and quality variables based on the first principals, it is still a challenge in practice due to the complexity of large-scale industrial process. To solve this problem, the data-driven quality related prediction and monitoring schemes are proposed in this paper. In order to overcome the drawbacks of standard approach, our focus is firstly concentrated on the modifications of standard partial least squares. Moreover, under industrial operating conditions, a subspace aided data-driven approach is further utilized to construct a soft sensor in the framework of diagnostic observer based residual generator. The proposed approaches are finally applied to quality based prediction and diagnosis on an industrial hot strip mill process. Application results indicate the effectiveness of the proposed methods and demonstrate improvement in performance compared to the standard technique.