Abstract: The disclosure provides a method, a device and a system for fault detection. The fault detection method includes following steps: acquiring interactive data generated during a working process of a target Internet of Things system, and generating data to be detected based on the interactive data; detecting whether the data to be detected is fault data or not, and performing an emotion analysis on the data to be detected which is judged as the fault data to obtain a corresponding emotion label; and extracting a corresponding fault type from a preset fault emotion mapping table based on the emotion label and outputting the fault type. In this disclosure, the interactive data is analogized to human emotions, and the faults are accurately found through the emotional analysis on the data, thereby achieving the fault detection of the Internet of Things system in a convenient, flexible and accurate manner.