![]() The frequency resolution Δ w of a signal has relations to the sampling number N and the sampling frequency F s, based on the formula Δ w = F s / N. This will lead to errors in frequency, amplitude, and phase, and eventually, affect the extraction accuracy. However, when performing Fourier transform of signals on computers, which can only process discrete data, it is unavoidable to suffer spectrum leakage and barrier effect because of time domain truncation and limited samples. ![]() Through spectrum analysis, various frequency components contained in the vibration signal can be clearly viewed. In order to ensure the real-time performance, the deep learning models require more optimized algorithms, better hardware and enough data. A large amount of data will lead to a long running time of algorithms. However, in the case of a limited amount of data, the deep learning algorithms cannot make unbiased estimates of the laws of the data. The deep learning models reduce the incompleteness caused by artificial design through self-learning and building feature models. A hybrid unsupervised feature selection (HFS) approach demonstrated its effectiveness in the fault diagnosis of rolling bearing. This method may be applied to the fault diagnosis of rolling bearing. The results indicate that this method is able to learn features adaptive from frequency data and achieve higher diagnosis accuracy. A CNN model can learn features from frequency data directly and detect faults of gearboxes. However, one of its obvious shortcomings is that it requires large amount of computation. The results showed that the average accuracy rate in the testing dataset reached more than 99%. Besides, the input of this method is the raw sampling signal without any pre-processing or traditional feature extraction. This method has the advantages of much higher prediction accuracy, faster iteration and more efficient to prevent over-fitting. For instance, based on a convolutional neural network (CNN, Address) and a long-short-term memory (LSTM) recurrent neural network, an improved bearing fault diagnosis method is proposed. With the rapid development of computer technology and machine learning, deep learning algorithms are increasingly applied to the fault diagnosis of rolling bearing. In addition to the traditional methods, new techniques and methods are also applied to the fault diagnosis of rolling bearing. The application of these methods in the fault diagnosis of rolling bearing suffers from their defects and requirements for data. However, there are some defects in EMD, such as over-envelope, under-envelope, mode mixing, end effect, IMF criterion, and may produce unexplained negative frequencies after calculating the instantaneous frequency. HHT includes empirical mode decomposition (EMD) and HT, and it is a self-adaptive time-frequency analysis method. While WT has a variable time-frequency window, the results are fixed-band signals when the wavelet basis and decomposition scale are selected. For example, the DRT is difficult to find out the best main resonance frequency band accurately, and the time-frequency window size of STFT is fixed. While these methods are effective and useful in the fault diagnosis of rolling bearing, there are still some limits. Many signal processing methods are used to process the vibration signal, such as demodulated resonance technique (DRT), short time Fourier transform (STFT), wavelet transform (WT), and Hilbert-Huang transform (HHT). The third part is the noise and interference. The second part is the vibration information of the rotating machines except the faulty rolling bearing. The first part is the fault information of rolling bearing with the characteristics of non-stationary, nonlinear, and modulation. The vibration signal of rolling bearing in the fault state usually consists of three parts. Among the three steps, feature extraction is the most critical one. The process of fault diagnosis can be divided into three steps, which are signal acquisition, feature extraction, and diagnosis decision. When a fault occurs, rotating machines should stop to repair or replace the faulty rolling bearing in time to avoid serious results. The purpose of fault diagnosis for rolling bearing is to determine the type of faults, the degree of damage, and the cause of faults. Condition monitoring and fault diagnosis of rolling bearing has become an attractive research topic. Rolling bearing is one kind of core parts in rotating machines, which plays an important role in industrial production.
0 Comments
Leave a Reply. |