The characteristics of petrochemical gearbox fault signals are fuzzy and coupled, and the normal state and abnormal state influence each other, which leads to the problem of difficult fault classification. In order to solve this problem, a highvalue dimensionless feature extraction method for petrochemical gearbox vibration signals was proposed. By using data preprocessing method, the valid data were extracted from the collected data, which were expressed through dimensionless indicators, and then the highvalue dimensionless features were extracted according to the distribution law of dimensionless indicators. In order to ensure the completeness of the sample, a highvalue dimensionless feature compatibilization model was established. By using this method, the bearing fault features of petrochemical gearbox experimental unit can be accurately extracted, and it can provide reference for the petrochemical gearbox feature extraction.