Data-Driven Method For Predicting The Health State And Trend Of Winder Chucks

Winding machine chemical fiber filament production of the key equipment,the clip is winding machine high-speed rotation of the core components.It is of great significance to study the prediction method of the health status trend of the clip to ensure the normal operation of the clip,so as to realize the high quality and high quantity production of chemical fiber.Traditional intelligent operation and maintenance methods are difficult to solve the problem of health status trend prediction due to the high noise and weak health degradation trend characteristics of the monitoring data.Therefore,in view of the characteristics of high operation monitoring data,this paper studies the method of constructing the health index of the head with time-frequency feature fusion under high noise.Aiming at the weak characteristics of health degradation trend,this paper studied the construction of health prediction model of card head strengthened by state trend under the weak characteristics.A prototype system for health status trend prediction of winding machine clip is developed.The specific work is as follows:(1)In view of the characteristics of high frequency strong interference and limited signal source under the construction of the health indicators of the winding machine,the time-frequency data feature fusion method for the construction of the health indicators of the winding machine was proposed.Aiming at the high frequency and strong interference characteristics of the vibration signal of the clip,a signal decomposition and reconstruction method based on EMD-permutation entropy was designed.The permutation entropy threshold was set to initially filter the interference information in the original signal.For card head take on the characteristics of the signal source is limited,the vibration signal through the signal time-frequency analysis,contains abundant bearing degradation characteristics of the original feature set,using correlation and monotonicity index trend time-frequency characteristics of the relativity of preliminary screening,reuse of both short-term and long-term memory module mapping network LSTM-HI get health indicators of winding machine clip.In experiment,public data set and instance data set are used to verify the validity of the method.(2)According to the characteristics of weak trend and time-varying non-constant speed in the early stage of winding machine clip prediction model construction,a clip health prediction model with enhanced state trend was proposed.In order to enhance the early weak features,firstly,the time spectrum of the continuous wavelet coefficients was obtained by the continuous wavelet transform,and the early weak trend was enhanced by the time-frequency enhancement,and the enhanced time spectrum was sent to the convolutional neural network to identify the health status of the card head.In order to overcome the non-adaptive threshold caused by time-varying non-constant velocity,an attention-mechanism-based method was designed in the trend prediction stage,and the attention-enhanced prediction results were taken as the output.In the experiment,the public data set and instance data set are used to verify the designed state trend prediction method.(3)According to the actual engineering requirements of a chemical fiber plant in Shaoxing,a prototype system for health state trend prediction of winding machine clip was developed.The efficiency and accuracy of state trend prediction of winding clip vibration signals could be effectively improved by data processing functions such as health indicator construction and state trend prediction.In this way,real-time monitoring and predictive maintenance of device status can be realized.