SENSOR MONITORING DRIVEN IDENTIFICATION OF HETEROGENEOUS WORKING CONDITIONS FOR MACHINE TOOL

With the development of intelligent manufacturing, real-time and adaptive RUL prediction will become a strong demand for intelligent processing centers. The heterogeneity of working conditions makes the lifetime present highly nonlinear, increasing the uncertainty of RUL prediction. Despite the development of various adaptive RUL prediction methods, the identification technique of heterogeneous working conditions is still a gap. To realize an adaptive or real-time RUL prediction, the identification of current working conditions and state transition trajectories of future working conditions is necessary. Therefore, by utilizing various monitored sensor data, a similarity evaluation method for working-condition identification is proposed. First, two features, the deviation between the probability density functions (PDF) in the amplitude domain and the cross-correlation coefficient in the time domain are constructed. Then, a similarity evaluation method is proposed by constructing the similarity level of two sensor data samples, where two different evaluating indexes are provided for the similarity level. One is evaluated by the average difference of sample features from different sensors under two different working conditions, and another is by the Euclidean distance of sample features under two different working conditions. Then, a comparison method based on 24 time-domain and frequency-domain features is given to verify the performance of the proposed method. At last, a case study based on a degradation dataset of milling insert is provided, proving the effectiveness of our proposed method. Also, the impact of different data manipulations on the accuracy rate of identifications is discussed in the case study.