Analysis And Diagnosis Of Short-time Vibration Signal Facing Wireless Vibration Sensor

The motor running in normal condition is an important premise to ensure the normal operation of industrial production and operation and avoid economic losses.In today’s increasingly extensive use of motors,it is particularly important to monitor the operation state of the motor in real time and diagnose its faults.This paper uses vibration signal to reflect the motor and has complex wiring and short service life.A low-power wireless vibration sensor was designed.The MEMS three-axis accelerometer ADXL357 is used as the sensor chip;the Lo Ra module serves as the wireless transmission module;and the chip STM32L431CBT6 with low power mode is selected as the processor core chip.A corresponding data integrated gateway is also designed.This paper proposes the concept of a short-time vibration signal to reduce the operating power consumption of the sensor.The collected short-time vibration data were spectrum analyzed by FFT to obtain the corresponding characteristics of the frequency domain.To diagnose short vibration signal,this paper introduces three machine learning methods,random forest,XGBoost and LSTM neural network algorithm.The motor fault data set is cited and preprocessed to meet the characteristics of short-term vibration signal,build four intelligent diagnosis algorithm models,and extract preprocessed data with 1 D convolution,the extracted data is divided into training set and test set,and brought into the four model comparison results,the LSTM neural network algorithm performs better in the fault classification of short-term vibration signal.Using the multi-scale feature fusion theory,the LSTM neural network was improved,and the data was brought into the model training and test.After analysis and comparison,the improved LSTM neural network model has stronger generalization ability,accuracy of 96%,and the Loss curve converges faster and less oscillation during convergence.