| Implementing an effective process monitoring algorithm is essential in minimizing downtime, increasing the safety of plant operations, and reducing manufacturing costs. Data-driven techniques based on multivariate statistics such as principal component analysis and partial least squares have been applied in many industrial processes and their effectiveness for fault detection is well-recognized. There is an inherent limitation on the ability for data-driven techniques to identify and diagnose faults, especially when the abnormal situations are associated with unknown faults and multiple faults. To improve the proficiency of data-driven techniques for fault identification and diagnosis, algorithms based on Fisher discriminant analysis and principal component analysis are proposed. In addition, a technique which integrates a causal map and data-driven techniques is proposed. The proficiencies of the methods are tested by application to the Tennessee Eastman process simulator and the results indicate that the new measures are better for monitoring the process compared to the existing measures. |
