The invention discloses an escalator fault diagnosis method based on the sequential probability. The escalator fault diagnosis method based on the sequential probability comprises the steps that vibration signals of key components of an escalator are collected through accelerator sensors mounted on the components of the escalator in vertical direction and the horizontal direction; the vibration signals are treated, and the peak values of the processed signals are used for characteristic values, namely indicator data of the actual operating state of the escalator; the indicator data are calculated through a sequential probability ratio algorithm to obtain a likelihood ratio; and the likelihood ratio is compared with the threshold value of a monitoring node, and fault diagnosis of an escalator mechanical system is conducted. According to the escalator fault diagnosis method based on the sequential probability, a wavelet packet algorithm is used for achieving smoothening and noise reduction of escalator data at different frequencies; and for the escalator in different operating states, the characteristic parameters are calculated through the sequential probability ratio algorithm, andthus differences caused by different operating states of the escalator are eliminated. In the detection process, the number of samples used for diagnosis does not need to be preset, the influence ofsample deficiencies or sample redundancy on the diagnosis result is avoided, the checking efficiency and accuracy are improved, and safety operation of the escalator is ensured.