Data-driven approaches to fault detection and isolation are widely used for various process systems. The purpose of this paper is to present a new method to improve the performance of fault diagnosis of chemical plant. This method combines simple process knowledge with data-driven diagnosis by introducing new feature variables. Simple method to create new feature variables is proposed. The proposed method was applied to diagnosis of the Tennessee Eastman plant simulation benchmark problem. Fault diagnosis performance on the extended feature sets are compared with the performance on the original dataset. The result shows that addition of new attributes is effective to improve the accuracy of the diagnosis.