[en] This paper presents the development of fault diagnosis using statistical method. We apply the fault diagnosis to the process data of the uranium enrichment demonstration plant. A uranium enrichment plant is structurally stable and there is a little change in the process data. It is necessary to detect the microscopic fluctuation in the process data before the serious fault. First we apply the auto-regressive model (AR model) to the process data, and estimate the validity of the AR model. Next we attend the cross-correlation in the process data, and construct the physical model using the simple law of the physics in the plant. We estimate the validity of the physical model. As a result we confirm that the fault diagnosis using statistical method is valid in the stable plant. (author)