Abstract: To improve the performance of locality preserving projections (LPP) for fault detection of a nonlinear process, we propose a fault detection method of anonlinear process based on differencial LPPs.First, the training data of the batch process are preprocessed.The nearest neighbor of each sample is found.The differencial operation is conducted between the sample and its nearest neighbor.Then, the LPP algorithm is used for dimensionality reduction and feature extraction.The squared prediction error (SPE) statistics of the samples is calculated, and kernel density estimation is used to determine the control limits.The new test sample data are projected onto the LPP model after differencial processing.The SPE statistics is calculated and compared with the control limits for fault detection.Simulation experiment results of numerical examples and semiconductor process data verify the effectiveness of the algorithm.