A chaotic system-based signal identification Technology: Fault-diagnosis of industrial bearing system

A set of smart fault-diagnosis technology with features production, extraction, and selection is developed via applying data-mapping strategy and K-Nearest Neighbor (K-NN) algorithm. By designing a nonlinear chaotic mathematics model consisting of a main system and a data-feed system with appropriate parameter settings, the raw vibration signals from time domain can be mathematically mapped into chaotic domain for potential features production, where those tiny differences between each of data set can be clearly observed in the chaotic domain. Further, through appropriate key-features extraction approaches-Euclidean Feature Values (EFV), Squared Euclidean Feature Values (SEFV), and the projections in the x, y and z directions, the related scatter plots of those features can be employed for key-features selections. In the final step, the classification results as well as their corresponding accuracy rates are investigated via K-NN algorithm. The different fault-states of industrial commonly-used rotating machine, rolling-bearing system, are illustrated for performance evaluations. In order to show the well-performance of the proposed technology, three fault conditions with diameter of 7 mil, 14 mil, 21 mil and a depth of 11 mil are all discussed. Experimental results demonstrate that the proposed smart fault-diagnosis technology is effective and feasible for real-time fault-diagnosis, where only K = 1 can be applied to identify different fault states in the fault conditions of 7 mil and 21 mil, the accuracy rate achieve around 100%. Further, for the fault conditions of 14 mil, 97.48 and 97.45% accuracy rate can be obtained via using K = 17 and K = 21. The average accuracy rate of the proposed smart detection system for the open as well as public database is around 99.16%