An Improved Fractal Dimension Algorithm and its Application on Gear Fault Recognition

The gear´s vibration signal is nonlinear and long-range correlation. The quartile deviation fractal dimension (QDFD) is used to indicate this characteristic. A new feature vector, which is made up of QDFD and intercept that is got when the QDFD is calculated, is used to fault identification. The experiment shows that the classification results are fine if the data were acquired in the same gear´s working condition and bad if the data was acquired in different gear´s working condition. An improved algorithm of QDFD (IQDFD) is introduced to overcome the shortcoming of new feature vector. The improved algorithm combined with Gaussian mixture model is used to the fault identification. The result shows that the proposed method can recognize the common gear´s failure mode correctly.