Research On Fault Diagnosis Of Rotating Machinery Based On Multi-sensor Information Fusion

With the widespread deployment of sensor technology and data acquisition systems in smart production workshops,data-driven intelligent health monitoring and fault diagnosis has entered the “industrial big data” era.Rotary machinery is widely used in mechanical engineering and plays an important role in engineering applications.With the advancement of science and technology,the structural composition and operating conditions of rotating machinery have become increasingly complex,which has led to the potential for faults occurring under severe production conditions and the difficulty in the diagnosis of faults.There are a large number of conflict data and noise data that affect the comprehensive performance of the fault diagnosis model in the data acquisition process of the multi-source sensing devices.The focus of this article is how to make full use of data of multi-source sensors to monitor the health status of rotating machinery equipment and conduct effective integrated fault diagnosis research under the industrial background of smart manufacturing + big data.The main research work is described as follows.Firstly,an IDS(Improved Dempster-Shafer Evidence Theory)information fusion algorithm is proposed to deal with possible conflict problems in the process of information fusion of multi-source sensors.The IDS algorithm substitutes the evidence body distance matrix into the improved Gini coefficient function,determines the confidence factors of the different evidence bodies through normalization,and effectively fuses the consistent evidence and the conflict evidence in a weighted fusion manner.And through the survival of the fittest mechanism,the confidence in synthetic evidence has been increased.Subsequently,a method based on spectrum slice feature reconstruction is proposed to extract condition information of the device.The method can convert the time domain signal to the frequency domain signal,and then reconstruct the reserved spectrum energy information by slice stacking,which can effectively extract the original signal.The feature extraction method can help reduce the dependence on expert experience,decrease model parameters,reduce sample requirements,accelerate model convergence and improve model diagnostic accuracy.Then,based on the limitations of single-source sensing non-integrated model,a rotating machinery fault diagnosis model IDSCNN(Improved Dempster-Shafer Evidence Theory with Convolutional Nerural Network)based on IDS algorithm and convolutional neural network is proposed.The model is applied to the bearing fault data set from Case Western Reserve University for experimental analysis.Compared with the single-source nonintegrated model and the current mainstream machine learning and deep learning model under various working conditions,the IDSCNN model shows better fault diagnosis performance.It provides a new research route for fault diagnosis of rotating machinery.Finally,aiming at a large number of possible noise data problems in a complicated and changeable production environment,a multi-source sensor noise adaptive model named MACNN(Multi-sensors Adaptive Convolutional Neural Network)is proposed.This model is based on IDSCNN by adding batch-normalization layers and a noise label adaptive network.The method improves the robustness to noise data and noise tags.By performing comparison experiments at different signal-to-noise ratios and different noise tag levels,the model can obtain relatively high fault diagnosis accuracy.