In this paper, a fault detection method based on One-Class Support Vector Machines(OCSVM) using im-proved particle swarm optimization to find optimal parame-ters is proposed. The feature recursion elimination support vector machine method is used to reduce unnecessary fea-tures, and the particle swarm algorithm combining particle mutation ideas and adaptive particle velocity improvement methods is used to optimize the relevant parameters of One-Class Support Vector Machines for fault detection. After being applied to the benchmark Tennessee Eastman problem dataset, the proposed algorithm is proved to have superior performance over One-Class Support Vector Machines with deterministic parameters,and it can take into account the detection effect of the training set and the test set.