Vibration Signal Fault Diagnosis Based On Convolutional Neural Network

Mechanical vibration is a phenomenon in the operation of mechanical equipment,which can reflect the operating state of the equipment.As the core component of modern machinery and equipment,rotating machinery is widely used in various fields.Its core components such as gears,bearings,etc.are constantly working in harsh environments There must be different degrees of structural damage,which in turn leads to state degradation of the entire mechanical system.Fault diagnosis of mechanical equipment based on vibration signals has become a major form of monitoring the condition of equipment,which can effectively improve system reliability.Neural networks are widely popular in various fields because of their powerful massive data processing capabilities,which simulate the human brain and construct multilayer nonlinear mappings to dig out the information implied by data.Therefore,around the neural network,this article expands the following aspects of the work:1.Aiming at the problem that the vibration signal distribution of the rotating machinery are different because of different working loads,resulting in the degradation of the accuracy of the fault diagnosis model,A transfer-learning-based transfer convolutional neural network model is used,which combines a variety of transfer learning methods to effectively address inconsistencies in the distribution of the source domain dataset and the target domain dataset The resulting model degradation problem was simulated on both the Case Western Reserve University bearing dataset and the PHM 2012 bearing dataset.Simulation results show that compared with the convolutional neural network fault diagnosis model,the transfer-learningbased model has a better performance.2.Aiming at the fact that the fault diagnosis algorithm based on deep learning does not make full use of the characteristics of rolling bearing vibration signal at this stage,a fault diagnosis method based on cyclostationary signal analysis and convolutional neural network is proposed.Rolling bearing fault signal is a typical cyclostationary signal,the use of cyclic spectrum coherence analysis can reveal bearing fault information from multiple dimensions,is an effective data preprocessing method.At the same time,a two-dimensional convolutional neural network model is established.To achieve image fault feature extraction for cyclic spectral coherence input data.The proposed method was simulated on the bearing dataset of Case Western Reserve University.Simulation results show that compared with other methods,The proposed method has stronger learning ability and can capture the occurrence of failures in time.3.From an engineering application perspective,a set of rolling bearings fault diagnosis system based on cyclic spectrum coherence and convolutional neural networks is developed by using the GUI environment of MATLAB.The system combines cyclostationary signal analysis with convolutional neural networks,which can diagnose the input vibration signal,provides great help for engineers,improves the reliability of mechanical systems,and makes the research content more practical.