Development of industrial networks and computational power of automation systems in recent years, have eliminated many previous limitations in collection and transferring of high volumes of data. These data are stored for a long time. This immense volume of data in industrial automation systems is an abundant source of information. With recent developments in data sciences and understanding importance of processing raw data, tendency to use these concepts for processing industrial data has increased. Fault diagnosis and identification of fault propagation path are among the most important applications of analyzing industrial data. Development of these solutions has important role in reducing accidents and failures and also preventing financial, life and environmental consequences. For this purpose, many different approaches have been recommended, most of which make use of process variables. Process variables data has some limitations for practical applications. For instance, industries compress this data source over time which cause loss of information. Also these approaches usually have high computational load and should be in execution consistently. In this research, alarm data source from history of alarm management system has been used for introduction of new approach to fault diagnosis. Proposed algorithm makes use of process mining approaches for the final goal. Process mining is still a young field related to data science and mostly is used for analysis of event-based databases. First, a framework for using process mining is introduced in which models are discovered for activated alarms due to known fault scenarios based on historical alarm data. After choosing appropriate model according to evaluation metrics, these models are used for online fault diagnosis. Online conformance checking using incremental prefix alignments is used to check conformance of activated alarms and discovered models for fault scenarios. Tennessee Eastman chemical process is used as a simulation of real chemical process to illustrate performance of the algorithm. An alarm management system for this process is designed in Matlab which produces alarms for known fault scenarios and stores them. Online fault diagnosis algorithm is deployed with Python programing language. Fault Identification, Causalities, Alarm Management System, Process Mining, Process Modeling