SPEAKER:	Jie Yu
TITLE:	Computational Intelligence based Smart Plant Monitoring, Diagnosis and Beyond
DATE:	January 26 2011
TIME:	11:15 am
PLACE:	JHE 342

ABSTRACT

Due to the increasing complexity of modern industrial processes, preventive monitoring and fault diagnosis have become highly critical to ensure safe operation, improve product quality and sustain economic profit of manufacturing. New generation of digital instruments, data devices, automation systems and high-performance computing together have promoted the development of a smart platform for plant monitoring and diagnosis.
In this presentation, a novel data-driven covariance benchmark is first proposed to assess performance of multivariable control systems with alleviated process model requirements. A series of multivariate statistical methods are further developed to identify the worst performance subspace and diagnose the underperforming control loops or variables. Moreover, the estimation of multivariable minimum variance control benchmark is simplified through the newly defined left/right diagonal interactor matrix. Meanwhile, a new Bayesian inference and finite Gaussian mixture model integrated framework is established for predictive monitoring of complex multimode processes. It is shown that the proposed inferential probability index can precisely detect different kinds of abnormal events and the corresponding mixture contribution method is able to isolate the faulty variables with high fidelity.  The application examples of the developed process and control performance monitoring methods will be demonstrated.
Finally, the plan to extend the computational intelligence techniques from industrial-scale systems to molecular-level biofactories with applications in medicine and renewable energy will be briefly discussed.