The invention belongs to the technology of escalator health condition management, and particularly relates to an escalator fault early warning system and a fault diagnosis method which can achieve real-time online monitoring based on vibration and temperature detection. The system comprises an acceleration sensor, a temperature sensor, a collector, an upper computer and a connecting cable. The collector is connected with the sensors and the upper computer through cables. The collector is used for processing vibration and signals input by the temperature sensor, whether the vibration and the signals are abnormal or not is judged, and if abnormality exists, alarm information and fault data are sent to the upper computer. According to the escalator fault early warning system and the fault diagnosis method, a feature vector formed by a plurality of time domain parameters and having the weight ratio is comprehensively applied to improve the comprehensiveness and accuracy rate of the vibration fault diagnosis. On an intelligent judgment algorithm, whether vibration is abnormal or not is judged according to the information difference obtained in a system training mode and a system work mode. Temperature monitoring is more reasonable and accurate through eliminating of the influence of the environment temperature on a detected object. The above diagnosis method is reliable, effective and good in application effect.