Data Driven Methods Based Fault Diagnosis And Remaining Life Prediction For Ball Screw Module

“Made in China 2025” puts forward the strategic requirement of “accelerating the integration of next-generation information technology and manufacturing industry as the main line to promote intelligent manufacturing as the main direction”.Equipment failure prediction and health management are the key points for the integration of information technology and manufacturing.One of them has become more and more important.The ball screw pair has the function of mutual conversion between linear motion and rotary motion.It is one of the most commonly used devices in mechanical transmission and is often used for mechanical equipment positioning and loading.The fault diagnosis and prediction of the ball screw sub-assembly is realized,which is of great significance for the efficient production and operation of the manufacturing industry.In this paper,after a brief analysis of the structural characteristics of the ball screw sub-assembly and the common failure modes,the research on the non-standard test bench design,ball screw sub-assembly fault diagnosis and remaining life prediction is carried out.In the design of the test bench,the ball screw sub-assembly non-standard test bench was designed based on the premise of not changing the working state and installation structure of the ball screw sub-assembly in the combined press roller machine,combined with the structural size of the sensor.Then select some types of signals that can change with the wear of the ball screw subassembly,and arrange corresponding sensors on the test bench to collect state information during the operation of the ball screw subassembly.According to the sensor output signal and interface type arranged on the test bench,select the appropriate NI acquisition device to complete the sensor signal acquisition.Finally,use the Lab VIEW software to drive the acquisition device,and write the data acquisition program of the ball screw sub-assembly test bench.In the fault diagnosis of the ball screw sub-assembly,different fault diagnosis methods were selected for the rolling bearing and the ball screw pair respectively.Due to the simple structure of the rolling bearing,the characteristic frequency of the fault can be obtained by calculation,so the characteristic frequency of each type of fault can be calculated and the identification of the rolling bearing failure mode can be directly compared with the envelope spectrum.In order to reduce the noise component in the signal,the EEMD method is used to adaptively decompose the vibration signal of the rolling bearing,and then the appropriate IMF component is selected by the correlation coefficient method to obtain the envelope spectrum,so that the fault frequency on the envelope spectrum is more obvious..In the ball screw pair fault diagnosis,the structure of the nut pair on the ball screw pair is complicated,and the diagnosis cannot be completed by the above-mentioned bearing diagnosis method.The machine learning method is selected to complete the ball screw pair fault diagnosis.Using the 3-layer wavelet packet decomposition,the fault signal is decomposed into 8 frequency bands,and the energy value of each frequency band is used as the energy characteristic value of the fault signal,and then some time-frequency characteristic values of the fault signal are calculated by using the formula,together as a ball screw pair fault.The feature is input to the convolutional neural network,and various faults of the ball screw pair are accurately identified.With regard to the life prediction of ball screw sub-assemblies,the long-short term memory network(LSTM)method is used to predict the life of bearings in ball screw subassemblies.The root-mean-square value of each set of IMS rolling bearing life data is extracted,and the long-short term memory network is used to predict the change of the root-mean-square value of the rolling bearing,thereby reflecting the rolling bearing wear trend.Then,by extracting the time-frequency characteristic values of each group of vibration signals and inputting them into the long-short term memory network,the remaining useful life of the rolling bearings is successfully predicted according to the usage duration of each group of data.