Research motivation and current trends on data-driven fault diagnosis are firstly introduced. Then new
classification of existing fault diagnosis approaches, which are data-driven methods, analytical model based methods and
qualitative knowledge based methods, is proposed to show the status of data-driven methods in the framework. The basic
application background is further discussed in terms of data requirements and comparison with analytical model based
methods. The state-of-art methods as well as their connections and differences are analyzed in detail. Finally, several
potential research directions are also summarized.