An alarm system plays a fundamental role in safety, quality, and economic profits of chemical processes. Alarm root-cause diagnosis is an essential part in alarm system management to monitor and locate the abnormalities. Because of the high integration and complexity in modern large scale industrial processes, a simplex structure model and, monolithic monitoring methods cannot always meet the requirement of alarm root-cause diagnosis. This work introduces a framework to identify the alarm root cause and visualize the abnormality propagation path. A. novel spatial interpretive structural model (SISM) is proposed to represent the hierarchical organization of Space and Show the causal relationships on different levels of granularity. In SISM, multiple spatial unit blocks are obtained by process decomposition based on the process flow diagram. Each block has one or more different variables. The hierarchical internal causal relationships in each individual block and external causal relationships between any two different blocks are identified by improving the traditional interpretive structural model. Considering the abnormality propagation, each spatial subblock and its child component(s) comprise one module, monitored by its corresponding extreme learning machine. The results from all the individual diagnostic modules may be inaccurate and mutually contradictory. We deploy an adaptation of analytical hierarchical process method to fuse these diagnostic results from various modules. The key benefits of our proposed framework have been demonstrated through seven fault scenarios in a Tennessee Eastman Chemical plant; All alarm root causes are successfully recognized with high diagnostic accuracy and short or even no diagnosis delay. The abnormality propagation paths are dearly visualized in SISM.