On Fault Diagnosis Method For Steam Turbines Based On Principal Manifold

With the development of industrial system mechanism as well as the advances in measurement,signal processing,the fault detection and diagnosis technology of industrial equipment is becoming more and more mature.However,the existence of nonlinear relationships between the variables in the industrial system leads to the underlying manifold sampled from the data space.Manifold learning can extract and retain the local spatial features of the input dataset as well as obtain the coordinates of the low-dimensional manifold space,which makes the reconstruction error as small as possible.Based on manifold learning knowledge,a new fault diagnosis method of steam turbine system is proposed,which achieves the following innovative results.1.A method of dimensionality reduction on the ellipsoid of Riemannian manifold is proposed.By summarizing the knowledge of Riemannian manifold,the formulas of exponential mapping and logarithmic mapping on ellipsoid are derived.By describing the characteristics of steam turbine system,especially the dataset of non-linear valve,the ellipsoid manifold characteristics of data are explained.At the same time,the data set is processed by the proposed method,and the corresponding experimental results are obtained.The accuracy of this method is illustrated by comparing with the experimental results of dimensionality reduction on sphere.2.A fault detection and diagnosis method of steam turbine based on principal manifold is proposed.Based on manifold learning,a method to find the tangent space on the local manifold space which corresponds to the input dataset.According to the statistical characteristics of tangent space,two indicators are calculated to detect fault on turbine set compared with the control limit.3.The method of fault detection and diagnosis based on the principal manifold is applied in the industrial environment of steam turbine system.The feasibility and reliability of the method are verified by experiments.