APPLICATION RESEARCH ON RANDOM FOREST ALGORITHM TO VIBRATION SIGNAL BASED FEATURE SELECTION AND PATTERN RECOGNITION

摘要

Random Forests Algorithm (RFA) is a combined classifier including a lot of classification trees, which has been widely used in the area of medical, biology and machine learning, but there are only several application researches on fault diagnosis for mechanical equipment. In this paper, based on introducing principle of RFA, by extracting feature parameters from vibration signals obtained from different condition of a mechanical gearbox, RFA is applied to evaluate importance of each feature parameters to pattern recognition. Simultaneously, a RFA based classifier is formed, and influence of the number of decision trees on classification accuracy and calculating time are discussed. Conclusion can be drawn from real measuring vibration signal from a gearbox, when the number of decision trees reach some value, classification accuracy will be near 100%.