Abstract: A novel technique of intelligent fault diagnosis of turbo-molecular pump,was developed,based on diverse features and multi-source information.The influence of the possible faults on the vibration signals was experimentally evaluated in multipoint sampling method with the lab-built test platform,theoretically analyzed and numerically processed in computation schemes.To be specific,first,the vibration signal spectra,acquired and preliminarily processed,were randomly divided into training and test sets; second,the multiple de-noising self-encoders were formed by changing the activation function; third,the generated deep auto-encoders were used to extract the features from the data set; and finally,the extracted features are used to train a support-vector-machine for fault classification.The test results show that the novel fault diagnosis technique is capable of detecting and diagnosing faults of molecular pump with an accuracy of 98.9%,even under the conditions of unbalanced training-set or high background noises.