Though Fisher discriminant analysis (FDA) is an outstanding method of fault diagnosis, it is usually difficult to extract the discriminant information in a complex industrial environment. One of the reasons is that, in such an environment, the discriminant information can not been extracted entirely due to the disturbances, non-Gaussianity and nonlinearity. In this paper, a method named Joint Fisher discriminant analysis (JFDA) is proposed to address the issues. First, JFDA removes outliers caused by disturbances according to the energy density of each datum. Then, for the non-Gaussianity and weakly nonlinearity, the novel scatter matrices are defined to extract both of the local and global discriminant information based on the manifold learning. Finally, the kernel JFDA (KJFDA) is investigated to hold the manifold assumption because the strongly nonlinearity may weaken the assumption and cause overlapping. The proposed method is applied to the Tennessee Eastman process (TEP). The results demonstrate that KJFDA shows a better performance of fault diagnosis than other improved versions of FDA.Note to Practitioners-The purpose of fault diagnosis is to identify the fundamental reasons which make the production line out of the normal operation. Modern industry process is so complex that it is difficult to establish a model by a mechanism analysis for fault diagnosis. The monitoring data of various operations are easy to gain in modern factories. We can get the diagnosis by comparing with the similarity of various operation data instead of the mechanism model. The work of this paper is motivated by the following reasons: the original data of various operations are nearly the same when it is researched from a global view in a complex industrial environment. However, if they are researched from a local view or in a high-dimensional space, they can be identified better than from a global view. In this paper, a novel method is proposed to research the operation data from both the local and global view in a high-dimensional space. In other words, the operation data are researched in the whole set (from a global view) and local neighborhood (from a local view) simultaneously. If the operation data are still not identified well, they should be researched in a high-dimensional space.The principle seems that though the building is not identified in the ground plan (2D), it can be identified well in a 3D map. We characterize the data mathematically and propose the corresponding method of fault diagnosis. The method is applied to the Tennessee Eastman process (TEP) which is a famous simulation demo of the chemical process, but it is not limited to the chemical industry and can be applied to other fields such as the power industry, the steel industry, etc. The experiments have suggested that the proposed method has a better performance. However, in practice, how to balance the complexity of the method and the superiority of the performance is an open issue in future research.Index Terms-Data-driven modeling, fault diagnosis, feat...