Data-driven process monitoring methods use only process data to build monitoring models, while useful process knowledge is completely ignored. Consequently, data-driven monitoring models usually have poor interpretability, which may reduce fault detection and diagnosis capabilities. To overcome such drawbacks, a novel knowledge data integrated sparse monitoring (KDISM) model and two-level deviation magnitude plots are proposed for industrial process monitoring. The basic idea of KDISM is to build a sparse and interpretable monitoring model by integrating process data with process knowledge. The procedure for building the KDISM model consists of four steps: (1) use basic process knowledge to analyze meaningful connections among process variables; (2) divide process variables into different groups according to variable connections; (3) construct a knowledge-based sparse projection (KBSP) matrix on the basis of the variable grouping results; (4) apply the KBSP matrix to process data to build the KDISM model. The KDISM model has good interpretability because the KBSP matrix reveals meaningful variable connections. The KDISM model is also able to eliminate redundant interference between variables owing to the sparsity of the KBSP matrix. These two advantages make the KDISM model well-suited for fault detection and diagnosis. Two fault detection indices are defined based on the KDISM model for detecting the occurrence of faults. The variable deviation magnitude (VDM) is defined to quantify deviations of variables from the normal values. Based on the VDM, two-level deviation magnitude plots are proposed for fault diagnosis, with the first-level groupwise VDM plot used to identify faulty variable groups while the second-level square VDM plot is used to identify faulty variables. The effectiveness and advantages of the proposed methods are illustrated by an industrial case study.