The invention discloses a fault detecting method based on a scattered OSC-PLS regression model and aims to establish an effective fault detecting model for carrying out more reliable fault detection from the perspective of equal footing of measurement variables. Specifically, the fault detecting method uses a scattered modeling strategy for reference, removes orthogonal irrelevant signals betweeneach measurement variable and other measurement variables, then utilizes a PLS regression algorithm to establish a regression model about the two parties and utilizes estimation errors of each measurement variable to carry out fault detection. Compared with a traditional method, the method utilizes scattered modeling to equivalently establish respective regression models for each measurement variable, removes orthogonal irrelevant signals in the regression models, guarantees the soft measurement precision of the regression models to the most degree and accordingly ensures follow-up fault detecting performance. In other words, the method is a more optimized data driven fault detecting method.