Led by market demand, the industrial process needs to switch to a variety of working modes, the industrial process detection system is becoming more and more complex, and data often presents the characteristics of the multi-mode complex distribution. It is of great significance to study multi-mode fault detection technology for ensuring the safe operation of industrial processes. The statistical process control method represented by principal component analysis (PCA) is a typical fault detection method based on the data drive. It is widely used to analyze whether there is a fault in the production process through the data collected by the system, which does not depend on prior knowledge and mathematical model. However, it requires that the data must conform to the Gaussian distribution, which cannot be satisfied in the multi-mode production process. To improve the performance of industrial process fault detection and eliminate the multi-modal and non-Gaussian characteristics of data, this work proposes a multi-mode process fault monitoring method based on local information LNS-PCA (LLNS-PCA). Firstly, the Gaussian mixture model (GMM) was used to divide the sample into several local samples. Secondly, for each sample data, the mean and variance of the local sample are standardized to make the data follow the Gaussian distribution. Finally, the data of each local sample were combined and PCA model was trained to obtain T 2 statistics and SPE statistics for fault monitoring. The LLNS-PCA algorithm was validated with numerical examples and penicillin production data as training samples. Under the same conditions, PCA, KPCA, and LNS-PCA are used to detect anomalies. The results showed that the LNS-PCA based on local information proposed in this work has a better detection effect. In conclusion, LLNS-PCA was superior to PCA, KPCA, and LNS-PCA, which was worth promoting.