Abstract: Given the nonlinearity of the chemical production process, to solve the selection of kernel parameter of kernel locality preserving projection algorithm and identify nuclear parameters that are suitable for multiple faults, we propose a new ensemble kernel locality preserving projection (KLPP) algorithm.First, we select a series of kernel functions with different parameters to project nonlinear data into a high-dimensional space for the extraction of nonlinear information of the data.The projection matrix A is obtained.A series of sub-KLPP models is established.Then, the kernel matrix of the data to be detected is calculated and then projected onto the matrix A.The detection results of each sub-model are obtained by statistics.Then, the detection results are transformed into the form of fault probability by using Bayesian decision.Finally, we use the ensemble learning method to combine the detection results.The final detection result is obtained by comparing the transformed result with the control limit.The method is applied to the Tennessee Eastman process and is proven to effectively detect nonlinear faults.