A novel method for condition monitoring of rotating machinery based on statistical linguistic analysis and weighted similarity measures

Defective rotating machinery generally produces complex fluctuations due to non-stationary and nonlinear properties of dynamical systems. Consequently, dynamical structures of vibration data from rotating machinery are hard to disclose. As a result, condition monitoring of rotating machinery is fairly challenging. In this paper, statistical linguistic analysis (SLA), a novel tool for time series analysis, was introduced to analyze dynamical mechanisms hidden in vibration data of rotating machinery. SLA maps original vibration data from rotating machinery to a binary symbolic sequence by exploiting potential of increase and decreases of time intervals. Next, by sliding a window and identifying the elements in each window as a “word”, a group of words is created. Then, by counting the occurrence of each word type, the binary symbolic sequence can be converted into a word frequency sequence. Next, a weighted similarity measure (WSM) defined in this paper serves to detect a change of running conditions of rotating machinery. As a result, this paper proposed a novel method for condition monitoring of rotating machinery based on SLA and WSM. Afterwards, the performance of the proposed method was validated using vibration data from both gearboxes and rolling bearings. Also, the proposed method was compared with conventional temporal statistical parameters, Approximate Entropy and Sample Entropy. The results indicate that the proposed method performs better than the other methods in condition monitoring of rotating machinery. Also, compared with either of Correlation Coefficients and Standardized Euclidean Distances, the WSM gives a somewhat better performance in reflecting a change of dynamical structures.