Research And Implementation Of Fault Detection System For Miniature Gear Reduction DC Motor Based On Deep Learning

In many industries,the demand for gear reducer motor is increasing,so it is necessary to improve the production efficiency of speed reduction geaerd motor for these domestic manufacturers.At present,domestic enterprises generally use manual detection methods to diagnose faults of geared motors,especially miniature gear reduction dc motor.This method not only limits the production efficiency,but also easily leads to detecting errors.Taking JL-12FN20-37 gear reduction DC motor as an example,combining the theory of motor fault diagnosis and deep learning,the fault detection system of micro gear reduction dc motor based on convolutional neural network is researched and implemented.This system can effectively realize fault detection by analyzing the vibration signal of JL-12FN20-37 gear reduction DC motor,which has certain application value.The work of the dissertation mainly includes the following three parts:(1)Feature Extraction and Selection of Vibration Signals: For the fault detection system designed in this dissertation,it is necessary to extract the appropriate vibration signal characteristics of JL-12FN20-37 gear reducer DC motor to improve the recognition rate of the system.Therefore,this dissertation extracts the non-stationary time-varying vibration signals by three feature extraction methods which are based on short-time fourier transform,based on hilbert-huang transform and based on 1/3 octave spectrum-principal component analysis.These three feature extraction methods all achieve data dimensionality reduction of vibration signals,reducing data redundancy.Finally,by analyzing these features,the input characteristics of the fault detection classification model are finally obtained.(2)Research on Fault Detection Classification Model: Firstly,this dissertation builds a fault detection classification model based on support vector data description as a benchmark for experiments.Secondly,two kind of fault detection classification models based on deep learning are designed.One of them is a classification model based on convolutional neural network.When improving the parameters of typical convolutional neural network model,it includes five optimized structure including false detection penalty and multi-segment detection,which all improve the accuracy and the precision of the classification model.The other is a fault detection classification model based on long short-term memory neural networks.The classification performance of the classification model is also improved after adding four optimized structures including dropout on recurrent connection.Finally,by analyzing the result of the experiment,this dissertation selects the convolutional neural classification model,which has the better result,as the classification model of fault detection system.(3)Implementation of Fault Detection System for JL-12FN20-37 Gear Reducer DC Motor: In this dissertation,according to the hardware performance and implementation requirements,the selection of hardware equipment such as DC power supply and vibration sensor is completed,and the motor measurement platform based on the spring damping frame is designed,which can helps to reduce the interference of external vibration sources.In the aspect of system software,this dissertation designs an easy-to-operate graphical user interface program for the fault detection system.Finally,the advantages and disadvantages of the system are evaluated through actual tests.