CN113435322A - Main shaft bearing fault detection method, system, equipment and readable storage medium - Google Patents
Main shaft bearing fault detection method, system, equipment and readable storage medium Download PDFInfo
- Publication number
- CN113435322A CN113435322A CN202110713654.9A CN202110713654A CN113435322A CN 113435322 A CN113435322 A CN 113435322A CN 202110713654 A CN202110713654 A CN 202110713654A CN 113435322 A CN113435322 A CN 113435322A
- Authority
- CN
- China
- Prior art keywords
- network
- vibration signal
- value
- main shaft
- shaft bearing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 238000001514 detection method Methods 0.000 title claims description 17
- 230000006870 function Effects 0.000 claims abstract description 53
- 238000000034 method Methods 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012545 processing Methods 0.000 claims abstract description 28
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 26
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 23
- 230000004913 activation Effects 0.000 claims description 20
- 238000007781 pre-processing Methods 0.000 claims description 16
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 8
- 238000013145 classification model Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 230000000717 retained effect Effects 0.000 claims description 5
- 230000001133 acceleration Effects 0.000 claims description 4
- 238000003745 diagnosis Methods 0.000 abstract description 18
- 238000005516 engineering process Methods 0.000 abstract description 15
- 238000013135 deep learning Methods 0.000 abstract description 10
- 238000004364 calculation method Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 230000009467 reduction Effects 0.000 description 6
- 238000000605 extraction Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000015556 catabolic process Effects 0.000 description 4
- 238000006731 degradation reaction Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000005096 rolling process Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000012800 visualization Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a method, a system, equipment and a readable storage medium for detecting faults of a main shaft bearing, which are based on a deep learning wide residual error network and a signal processing technology, firstly, a vibration signal is subjected to collective empirical mode decomposition, modal components are screened based on kurtosis values, singular value decomposition optimization is carried out on the screened components, signal characteristics in an image form are output through short-time Fourier transform after the signals are reconstructed, then, a wide residual error network is built, image characteristic data is input into the wide residual error network for training, and finally, the wide residual error network with the fault diagnosis function is obtained. The accuracy and efficiency of fault diagnosis are improved.
Description
Technical Field
The invention belongs to the field of main shaft bearing fault diagnosis, and particularly relates to a main shaft bearing fault detection method, a main shaft bearing fault detection system, main shaft bearing fault detection equipment and a readable storage medium.
Background
With the continuous development of modern mechanical equipment, the trend of large-scale, precise and complex equipment is realized, the service conditions of the equipment are various, and the equipment is often accompanied with severe working conditions such as variable load, noise interference, impact and the like. The bearing is a key part of the main shaft, and related data show that about 30% -40% of equipment faults occur on the bearing, and if the faults cannot be found in time, serious economic loss and even safety accidents are caused. Therefore, monitoring the state of the main shaft bearing and timely fault diagnosis and judgment are of great importance, the safety service of mechanical equipment is guaranteed, the production efficiency can be improved, and the economic benefit is increased.
The conventional signal processing technology has certain limitation when being applied to monitoring the state of a main shaft bearing, and common monitoring methods such as empirical mode decomposition, envelope analysis, wavelet transformation, Fourier transformation and the like mostly need manual identification after feature extraction operation, so that the efficiency of fault diagnosis has a space for further improving. For example, in publication No. 1CN110514441A, "a rolling bearing fault diagnosis method based on vibration signal denoising and envelope analysis", the diagnosis of the fault is completed by performing empirical mode decomposition on the vibration signal after the secondary filtering denoising, selecting the first intrinsic mode function for envelope analysis, and separating the characteristic frequency of the fault.
In addition, the deep learning technology can adaptively extract the feature parameters in the data, and compared with the traditional feature extraction algorithm, the deep learning technology has the advantages that: by establishing a deep model, the hidden characteristics of data are learned from a large amount of data, the intelligence of computer processing problems is realized, the dependence on a large amount of signal processing technologies and diagnosis experiences is eliminated, meanwhile, the generalization capability of the fault diagnosis method using the deep learning technology is greatly improved, and the actual requirement of the universality of mechanical fault diagnosis is met.
The traditional convolutional neural network obtains stronger analysis capability by increasing the number of layers, but with the increase of depth, the phenomenon of gradient explosion or gradient disappearance becomes more and more serious, the network degradation becomes more obvious, the training effect of the network is seriously influenced, but the extraction and identification effect of the shallow network on the fault characteristics is poor, and therefore the number of layers of the network needs to be set reasonably. Meanwhile, the vibration signal of the bearing usually contains strong noise, the noise has randomness and complexity, and the deep learning model usually takes the noise as information to be learned to be trained together, so that extremely low training efficiency and accuracy are caused.
Disclosure of Invention
The invention aims to provide a method, a system, equipment and a readable storage medium for detecting faults of a main shaft bearing, so as to overcome the defects of the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
a main shaft bearing fault detection method comprises the following steps:
s1, acquiring a main shaft bearing vibration signal, performing ensemble empirical mode decomposition processing on the acquired vibration signal to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, and keeping the inherent modal component with the kurtosis value larger than a set threshold value;
s2, constructing a Hankel matrix for the reserved inherent modal components, and performing singular value decomposition to complete vibration signal data preprocessing;
s3, carrying out short-time Fourier transform on the preprocessed vibration signal, and outputting the vibration signal in a characteristic image form;
and S4, selecting the width and depth of the network, constructing a wide residual error network, inputting the characteristic image obtained in the step S3 into the deep wide residual error network for training, finishing training to obtain a classification model after the network loss function value tends to be stable, and detecting the fault of the main shaft bearing by using the trained classification model.
Further, the acceleration sensor acquires a vibration signal of the main shaft bearing, EEMD processing is performed on the vibration signal, normally distributed white noise is added to the original signal, and EMD processing is performed on the original signal as a whole:
obtaining N natural modal components IMFi。
Further, a kurtosis value is calculated for each eigenmode component:
the threshold value is set to 0, and the natural mode component having a kurtosis value smaller than the threshold value is discarded.
Further, IMF is performed on the retained intrinsic mode componenti＝[a1,a2,a3,…,an]
A Hankel matrix is constructed and,
SVD (singular value decomposition) is carried out to obtain H ═ U ∑ VTObtaining a singular value matrix sigma-diag1,σ2,…,σn,0,…,0]The matrix of singular values after the singular values of the first K orders are reserved is sigma ═ diag [ sigma [ ]1,σ2,…,σK,0,…,0]And performing an inverse process of the SVD and the Hankel matrix so as to reconstruct the inherent modal component.
Further, short-time fourier transform is performed on the preprocessed vibration signal:
the window function is a Hanning window, the characteristics of the signals on a time frequency domain are output in an image form, and a vibration signal characteristic image of the fault main shaft bearing with time represented by a horizontal axis, frequency represented by a vertical axis and energy represented by color is obtained; the size of the output vibration signal characteristic image is 320 multiplied by 240, and the number of image channels is 3.
Further, the output corresponding to the input image data is a label of a fault category, and different types of labels are set by adopting one-hot coding for different types of faults.
Further, adding a batch normalization layer BN and an activation layer ELU to the front end of the main path of the basic unit residual block of the wide residual network;
the activation function of the network is an ELU function:
the value of a is set to 1;
the loss function of the deep wide residual network is a classified cross entropy function:
and returning the error along the minimum gradient direction, and correcting the weight matrixes of the convolution layer and the full-connection layer of the deep wide residual error network to reduce the cross entropy loss function value of the next iterative training until the loss function value is smaller than the set value.
A spindle bearing fault detection system comprising:
the signal preprocessing module is used for performing ensemble empirical mode decomposition processing on the acquired vibration signals to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, reserving the inherent modal component with the kurtosis value larger than a set threshold value, constructing a Hankel matrix for the reserved inherent modal component, and performing singular value decomposition to complete vibration signal data preprocessing;
the characteristic image training module is used for carrying out short-time Fourier transform on the preprocessed vibration signals and outputting the vibration signals to the classification module in a characteristic image form;
and the classification module is used for carrying out optimization training according to the characteristic images until the network loss function value tends to be stable to obtain a classification model, and outputting a corresponding bearing fault result according to the input vibration signal of the main shaft bearing.
A terminal device comprising a memory, a processor and a computer program stored in said memory and executable on said processor, said processor implementing the steps of the above-mentioned spindle bearing failure detection method when executing said computer program.
A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the above-mentioned spindle bearing failure detection method.
Compared with the prior art, the invention has the following beneficial technical effects:
the invention relates to a main shaft bearing fault detection method, which is based on a deep learning wide residual error network and a signal processing technology, firstly carries out ensemble empirical mode decomposition on a vibration signal, screens modal components based on a kurtosis value, singular value decomposition optimization is carried out on the screened components, signal characteristics in an image form are output through short-time Fourier transform after signals are reconstructed, then building a wide residual error network, inputting image characteristic data into the wide residual error network for training, and finally obtaining the wide residual error network with the fault diagnosis function, the method combines the signal processing technology and the deep learning technology, reduces the influence of bearing vibration signal noise, meanwhile, the wide residual error network relieves the requirement of a common artificial neural network model on the high performance of the computer, and the performance degradation problem which often occurs along with the increase of the number of network layers improves the accuracy and efficiency of fault diagnosis.
According to the method, a signal processing technology and a deep learning technology are combined, noise reduction preprocessing on vibration data is firstly realized, a deep wide residual error network is built, self-adaptive extraction of fault characteristics is realized, the problem of noise influence contained in data in the prior art is solved, and the fault identification accuracy and the generalization capability of a model are improved by applying the deep wide residual error network.
Furthermore, the method is widened on the basis of the residual error network, namely the number of convolution kernels is increased, the number of layers of the network is reduced, and the wide residual error network not only increases the network training speed, but also reduces the high requirements of the multi-layer network on the computer performance.
Furthermore, the ELU function has a negative value part, so that the bias of the output of the activation function is reduced, and the calculation speed is increased.
Furthermore, the weight of the convolution layer and the full-link layer of the depth wide residual error network is adjusted, so that the cross entropy value is continuously reduced in the iterative training process.
One-dimensional time sequence signals are converted into characteristic images through short-time Fourier transform, the visualization degree of data processing is increased, and meanwhile, the feasibility and the accuracy of fault diagnosis of the rotary machine are improved by introducing a deep learning technology; by using the wide residual error network, the number of the network layers is reduced and the network parameters are not reduced at the same time by increasing the number of convolution kernels, so that the wide residual error network keeps higher accuracy and obviously improves the running speed at the same time, and meanwhile, the problem of network degradation caused by the increase of the number of the network layers is solved, and in addition, a pre-activation layer is added at the front end of a residual error path of a residual error block, so that the effects of reducing the calculation difficulty and the occurrence of overfitting are achieved; a pre-activation link is added to the deep wide residual error network, and a group of batch normalization layers and activation layers are additionally added before convolution operation, so that the calculation difficulty and the possibility of overfitting are further reduced.
Drawings
Fig. 1 is a flowchart of a method for diagnosing a fault of a main shaft bearing according to an embodiment of the present invention.
Fig. 2 is a frequency domain waveform of an inner ring single point fault signal in an embodiment of the invention.
FIG. 3 is a frequency domain waveform of an inner ring single point fault signal processed only by EEMD in an embodiment of the present invention.
FIG. 4 is a frequency domain waveform of an inner ring single point fault signal after EEMD + SVD processing in the embodiment of the present invention.
FIG. 5 is a time-frequency domain characteristic diagram of an inner ring single-point fault signal after EEMD + SVD processing in the embodiment of the present invention.
Fig. 6 is a residual block of a residual network in an embodiment of the present invention.
FIG. 7 is a convolutional layer of the deep wide residual network WRN-22-4 in an embodiment of the present invention.
Fig. 8 is a depth wide residual network with pre-activation added in the embodiment of the present invention.
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings:
a main shaft bearing fault detection method comprises the following steps:
s1, acquiring a main shaft bearing vibration signal, performing Ensemble Empirical Mode Decomposition (EEMD) processing on the acquired vibration signal to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, and keeping the inherent modal component with the kurtosis value larger than a set threshold value;
specifically, an acceleration sensor is arranged on a main shaft, and operating equipment collects vibration signals of a main shaft bearing;
the acceleration transducer obtains a one-dimensional time sequence vibration signal, EEMD processing is carried out on the vibration signal, normally distributed white noise is added into an original signal, and EMD processing is carried out on the original signal as a whole:
obtaining N natural modal components IMFi。
The kurtosis values are calculated for each of the natural modal components:
the threshold value is set to 0, and the natural mode component having a kurtosis value smaller than the threshold value is discarded.
S2, constructing a Hankel matrix for the reserved inherent modal components, and performing Singular Value Decomposition (SVD) to complete vibration signal data preprocessing;
specifically, the retained natural modal component is subjected to SVD optimization processing, and the retained natural modal component IMF is firstly subjected to IMFi＝[a1,a2,a3,…,an]
A Hankel matrix is constructed and,
SVD (singular value decomposition) is carried out to obtain H ═ U ∑ VTObtaining a singular value matrix sigma-diag1,σ2,…,σn,0,…,0]The matrix of singular values after the singular values of the first K orders are reserved is sigma ═ diag [ sigma [ ]1,σ2,…,σK,0,…,0]And performing an inverse process of the SVD and the Hankel matrix to reconstruct the inherent modal component, and then combining the inherent modal components to perform the next signal preprocessing.
S3, performing Short-time Fourier Transform (STFT) on the preprocessed vibration signal, outputting the vibration signal in a characteristic image form, and dividing the vibration signal into a training set and a test set;
and carrying out short-time Fourier transform on the preprocessed vibration signal:
the window function is a Hanning window, the characteristics of the signals on a time frequency domain are output in an image form, and a vibration signal characteristic image of the fault main shaft bearing with time represented by a horizontal axis, frequency represented by a vertical axis and energy represented by color is obtained. The size of the output vibration signal characteristic image is 320 multiplied by 240, and the number of image channels is 3.
S4, selecting the width and depth of the Network, building a Wide Residual error Network (WRN), inputting image training set data into the deep Wide Residual error Network for training, finishing training to obtain a classification model after the Network loss function value tends to be stable, inputting test set data into the trained Network, obtaining the result of fault diagnosis for verification, and utilizing the trained Network for detecting the fault of the main shaft bearing.
The built depth wide residual error network is WRN-22-4, namely the depth of the wide residual error network is 22 layers, and the width of the wide residual error network is 4 times of that of a common residual error network. To reduce the requirements on computer performance, image data is loaded at the input of the network using a Batch method, with the single Batch import number (Batch size) set to 100. The input data of the depth-width residual error network is a characteristic image, namely the input data of the depth-width residual error network is in the form of.
The output corresponding to the input image data is a label of a fault category, and the labels of different categories are set by adopting One-Hot coding (One-Hot) for the faults of different categories. The wide residual error network increases the number of convolution kernels in the residual error block on the basis of a general residual error network, namely the number of convolution kernels is increased to 4 times. The basic unit residual block of the wide residual network comprises two branches, namely a Main Path (Main Path) and a Shortcut Path (Shortcut Path), wherein the Shortcut has two categories including identity mapping and convolution. The structure of the basic unit residual block of the wide residual network is improved, and a batch normalization layer BN and an activation layer ELU are added at the front end of a main path to play a role in pre-activation.
The activation function of the network is an ELU function:
the value of a is set to 1.
The loss function of the depth-wide residual network is a Categorical cross-entropy function (category cross-entropy):
and returning the error along the minimum gradient direction, and correcting the weight matrixes of the convolution layer and the full-connection layer of the deep wide residual error network to reduce the cross entropy loss function value of the next iterative training until the loss function value is smaller than the set value.
Adding L2 regularization to change the loss function to
In the network iterative training process, an Adaptive Moment Estimation algorithm (Adam) is adopted as a gradient descent algorithm of loss function convergence.
The validity of the diagnostic method of the present invention is verified using existing bearing data as an example. The data set comprises normal data, fault data of a 12kHz driving end, fault data of a 48kHz driving end and fault data of a fan section, the fault types comprise three states of bearing inner ring faults, bearing outer ring faults and bearing rolling body faults, the fault data of the driving end with the sampling frequency of 12kHz is adopted, the working condition of the bearing is shown in a table 1, and the specification parameters of the bearing are shown in a table 2.
TABLE 1 operating conditions
TABLE 2 bearing information
Referring to fig. 1, the method for detecting a fault of a main shaft bearing according to the present invention includes the following steps:
firstly, collecting vibration signals;
secondly, preprocessing the signals, including collecting empirical mode decomposition and singular value decomposition;
thirdly, performing short-time Fourier transform on the signal subjected to noise reduction preprocessing, and outputting a time-frequency domain characteristic image of the fault bearing signal;
and fourthly, determining an input structure and an output structure, setting the width and the depth of the network, and constructing a depth wide residual error network.
Wherein the networkThe activation function of (2) is an ELU function:
the parameter a is set to be 1, compared with the ReLU function, the method has the advantages that the problem of gradient disappearance can be relieved, the offset effect is reduced due to the negative value part, and meanwhile, the calculation speed is increased.
The loss function of the network in the training process is a classified cross entropy function, the weights of the convolution layer and the full-connection layer of the network are optimized by continuously reducing the loss function, and the gradient descent algorithm of the optimization target of the minimized cross entropy function adopts self-adaptive moment estimation.
The basic unit residual block of the network is shown in fig. 6, and the principle is as follows: assuming that the input is x, the neuron output is h (x), a general neuron is directly fitted by convolution, an identity mapping path is added in a wide residual unit, the fitted output is composed of two parts, namely h (x) ═ f (x) + x, as the number of network layers is increased, the phenomenon of gradient disappearance occurs, the network performance is seriously degraded, the output of a convolution function, namely h (x) ═ 0, and the output of a residual block, namely h (x) ═ x, ensures that the network performance is not degraded at least.
According to the method, a processing method combining EEMD and SVD is used for a spindle bearing vibration signal containing strong noise, the EEMD solves the modal mixing problem of EMD, IMF components with the kurtosis value larger than 0 are extracted after decomposition, then a Hankel matrix is constructed for the screened IMF components, and SVD is used for further optimizing data; one-dimensional time sequence signals are converted into characteristic images through short-time Fourier transform, the visualization degree of data processing is increased, and meanwhile, the feasibility and the accuracy of fault diagnosis of the rotary machine are improved by introducing a deep learning technology; by using the wide residual error network, the number of the network layers is reduced and the network parameters are not reduced at the same time by increasing the number of convolution kernels, so that the wide residual error network keeps higher accuracy and obviously improves the running speed at the same time, and meanwhile, the problem of network degradation caused by the increase of the number of the network layers is solved, and in addition, a pre-activation layer is added at the front end of a residual error path of a residual error block, so that the effects of reducing the calculation difficulty and the occurrence of overfitting are achieved; a pre-activation link is added to the deep wide residual error network, and a group of batch normalization layers and activation layers are additionally added before convolution operation, so that the calculation difficulty and the possibility of overfitting are further reduced.
In one embodiment of the present invention, a terminal device is provided that includes a processor and a memory, the memory storing a computer program comprising program instructions, the processor executing the program instructions stored by the computer storage medium. The processor is a Central Processing Unit (CPU), or other general purpose processor, Digital Signal Processor (DSP), Application Specific Integrated Circuit (ASIC), ready-made programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc., which is a computing core and a control core of the terminal, and is adapted to implement one or more instructions, and in particular, to load and execute one or more instructions to implement a corresponding method flow or a corresponding function; the processor of the embodiment of the invention can be used for the operation of the main shaft bearing fault detection method.
A spindle bearing fault detection system, comprising:
the signal preprocessing module is used for performing ensemble empirical mode decomposition processing on the acquired vibration signals to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, reserving the inherent modal component with the kurtosis value larger than a set threshold value, constructing a Hankel matrix for the reserved inherent modal component, and performing singular value decomposition to complete vibration signal data preprocessing;
the characteristic image training module is used for carrying out short-time Fourier transform on the preprocessed vibration signals and outputting the vibration signals to the classification module in a characteristic image form;
and the classification module is used for carrying out optimization training according to the characteristic images until the network loss function value tends to be stable to obtain a classification model, and outputting a corresponding bearing fault result according to the input vibration signal of the main shaft bearing.
In still another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a Memory device in the terminal device and is used for storing programs and data. The computer-readable storage medium includes a built-in storage medium in the terminal device, provides a storage space, stores an operating system of the terminal, and may also include an extended storage medium supported by the terminal device. Also, one or more instructions, which may be one or more computer programs (including program code), are stored in the memory space and are adapted to be loaded and executed by the processor. It should be noted that the computer-readable storage medium may be a high-speed RAM memory, or may be a Non-volatile memory (Non-volatile memory), such as at least one disk memory. One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement the corresponding steps of the method for detecting a failure of a spindle bearing as may be used in the above-described embodiments.
The specific embodiment is as follows:
a main shaft bearing fault detection method comprises the following specific processes:
preprocessing of signals: EEMD and SVD processing is carried out on the acquired one-dimensional time sequence vibration signals, firstly, the vibration data is subjected to ensemble empirical mode decomposition into a plurality of IMF components, and the IMF components are subjected to equation decomposition
Calculating the kurtosis of each IMF component, and screening out the components with the kurtosis larger than 0, wherein in this example, the failure characteristic frequency of the inner ring is about 160Hz, and as more noise still exists in each IMF component, as shown in FIG. 3, if the retained IMF components are directly merged, namely only through EEMD processing, the noise reduction effect is still not ideal, so SVD optimization is further performed on the IMF components, and the intrinsic modal component IMF is subjected toi＝[a1,a2,a3,…,an]Constructing a Hankel matrix
Then SVD is carried out to decompose H ═ U ∑ VTObtaining a singular value matrix sigma, and reserving the maximum first 5 singular items in the sigmaThe signals are optimized to obtain new IMF components, and the new IMF components are merged, so that the noise reduction pretreatment of the one-dimensional time sequence signals is completed as shown in FIG. 4.
And performing short-time Fourier transform on the signal subjected to noise reduction preprocessing. The time resolution and the frequency resolution are influenced by window length parameters of short-time Fourier transform, the longer the window length, the higher the frequency resolution and the lower the time resolution, and vice versa, the lower the frequency resolution and the higher the time resolution, in this example, a window function selects a Hamming window, and the window length Block size is set to 2048, as shown in FIG. 5, the time-frequency domain features of the fault vibration signal are output in an image form, a data set is constructed, and the data set is divided into a training set and a test set.
Setting the width and depth of the network, and constructing a depth wide residual error network. In the example, the size of the vibration signal characteristic image is 320 x 240, the corresponding output result is a fault type of one-hot coding, the example has three types of outer ring fault, inner ring fault and rolling element fault, the labels are [0,0,1], [0,1,0], [1,0,0], the number of layers of the network is set to 22, the width is set to 4 times, namely a wide residual error network WRN-22-4 is built, then a training set is led into the network for iterative training, after a loss function value tends to be stable, the network training is finished, a system with different types of fault diagnosis is obtained, and then a test set is input into a diagnosis model to obtain a fault diagnosis result.
Specifically, the network structure of WRN-22-4 includes 5 blocks, which are 1 convolution segment Conv _1 and 3 wide residual segments, respectively, including Conv2_ x, Conv3_ x, Conv4_ x, and 1 full-link Layer FC Layer.
Specifically, as shown in fig. 7, the data first enters a first wide residual segment Conv _1, and the calculation formula is as follows:
g (x) is an ELU function,
which represents the input data, is,
in order to output the data, the data is output,
is a weight matrix of the convolution kernel,
the bias execution matrix of convolution kernel represents convolution operation, and the convolution layer is followed by BN layer and activation function ELU. The convolution block Conv _1 has a convolution kernel size of 3 × 3 and a number of 16.
Specifically, the structures of Conv2_ x to Conv4_ x are similar, and are all the superposition of convolution of 3 × 3 and Dropout layers, except that the number of convolution kernels is different, the number of convolution kernels of each unit of Conv2_ x is 16 × 4, the number of convolution kernels of each unit of Conv3_ x is 32 × 4, the number of convolution kernels of each unit of Conv4_ x is 64 × 4, and 22 layers are provided, wherein the multiple 4 is the depth of WRN-22-4.
Specifically, as shown in fig. 8, an improvement is made in the residual path of the residual block, and a group of batch normalization layers BN and activation layers ELU are additionally added before the convolution operation starts, so as to achieve the purpose of pre-activation, thereby reducing the difficulty of calculation and the possibility of occurrence of over-fitting, and accelerating the speed of network training.
Specifically, the end of the wide residual error network is a fully connected layer, after the output of the residual error block passes through the activation function, the fully connected layer performs "Average Pooling" (Average Pooling) and "flattening" (Flatten) "on the output, and finally outputs the classification result through the classification function SoftMax. The output of the SoftMax function is the probability of a certain class, assuming an input array ViThe SoftMax value of the jth element is:
specifically, the L2 regularization method is to add a correlation term of weight sum of squares after the loss function
In trainingLoss function is iterated continuously in the training process, and L2 regularization terms
The method has the function of reducing omega, and the complexity of the network can be reduced by the smaller weight omega, so that the function of reducing the occurrence of the overfitting phenomenon is achieved.
According to the method, a signal processing technology and a deep learning technology are combined, noise reduction preprocessing on vibration data is firstly realized, a deep wide residual error network is built, self-adaptive extraction of fault characteristics is realized, the problem of noise influence contained in data in the prior art is solved, and the fault identification accuracy and the generalization capability of a model are improved by applying the deep wide residual error network.
Claims (10)
1. A main shaft bearing fault detection method is characterized by comprising the following steps:
s1, acquiring a main shaft bearing vibration signal, performing ensemble empirical mode decomposition processing on the acquired vibration signal to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, and keeping the inherent modal component with the kurtosis value larger than a set threshold value;
s2, constructing a Hankel matrix for the reserved inherent modal components, and performing singular value decomposition to complete vibration signal data preprocessing;
s3, carrying out short-time Fourier transform on the preprocessed vibration signal, and outputting the vibration signal in a characteristic image form;
and S4, selecting the width and depth of the network, constructing a wide residual error network, inputting the characteristic image obtained in the step S3 into the deep wide residual error network for training, finishing training to obtain a classification model after the network loss function value tends to be stable, and detecting the fault of the main shaft bearing by using the trained classification model.
2. The method according to claim 1, wherein the acceleration sensor obtains a vibration signal of the main shaft bearing, the vibration signal is subjected to EEMD processing, normally distributed white noise is added to the original signal, and the original signal is subjected to EMD processing as a whole:
obtaining N natural modal components IMFi。
3. A method for detecting a failure of a main shaft bearing according to claim 1, wherein a kurtosis value is calculated for each of the natural mode components, respectively:
the threshold value is set to 0, and the natural mode component having a kurtosis value smaller than the threshold value is discarded.
4. A method of spindle bearing fault detection as claimed in claim 1 wherein the retained natural modal component is IMFi＝[a1,a2,a3,…,an]
A Hankel matrix is constructed and,
SVD (singular value decomposition) is carried out to obtain H ═ U ∑ VTObtaining a singular value matrix sigma-diag1,σ2,…,σn,0,…,0]The matrix of singular values after the singular values of the first K orders are reserved is sigma ═ diag [ sigma [ ]1,σ2,…,σK,0,…,0]And performing an inverse process of the SVD and the Hankel matrix so as to reconstruct the inherent modal component.
5. The method for detecting the fault of the main shaft bearing according to claim 1, wherein the preprocessed vibration signal is subjected to short-time Fourier transform:
the window function is a Hanning window, the characteristics of the signals on a time frequency domain are output in an image form, and a vibration signal characteristic image of the fault main shaft bearing with time represented by a horizontal axis, frequency represented by a vertical axis and energy represented by color is obtained; the size of the output vibration signal characteristic image is 320 multiplied by 240, and the number of image channels is 3.
6. The method of claim 1, wherein the corresponding output of the input image data is a label of a fault category, and different types of labels are set by using unique hot coding for different types of faults.
7. The method according to claim 1, wherein a batch normalization layer BN and an activation layer ELU are added to the front end of the main path of the wide residual network basic unit residual block;
the activation function of the network is an ELU function:
the value of a is set to 1;
the loss function of the deep wide residual network is a classified cross entropy function:
and returning the error along the minimum gradient direction, and correcting the weight matrixes of the convolution layer and the full-connection layer of the deep wide residual error network to reduce the cross entropy loss function value of the next iterative training until the loss function value is smaller than the set value.
8. A spindle bearing fault detection system, comprising:
the signal preprocessing module is used for performing ensemble empirical mode decomposition processing on the acquired vibration signals to obtain a plurality of inherent modal components, calculating the kurtosis value of each inherent modal component, reserving the inherent modal component with the kurtosis value larger than a set threshold value, constructing a Hankel matrix for the reserved inherent modal component, and performing singular value decomposition to complete vibration signal data preprocessing;
the characteristic image training module is used for carrying out short-time Fourier transform on the preprocessed vibration signals and outputting the vibration signals to the classification module in a characteristic image form;
and the classification module is used for carrying out optimization training according to the characteristic images until the network loss function value tends to be stable to obtain a classification model, and outputting a corresponding bearing fault result according to the input vibration signal of the main shaft bearing.
9. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the steps of the method of any of claims 1 to 7 are implemented when the computer program is executed by the processor.
10. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110713654.9A CN113435322B (en) | 2021-06-25 | 2021-06-25 | Method, system, equipment and readable storage medium for detecting faults of spindle bearing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110713654.9A CN113435322B (en) | 2021-06-25 | 2021-06-25 | Method, system, equipment and readable storage medium for detecting faults of spindle bearing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113435322A true CN113435322A (en) | 2021-09-24 |
| CN113435322B CN113435322B (en) | 2024-04-02 |
Family
ID=77754680
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110713654.9A Active CN113435322B (en) | 2021-06-25 | 2021-06-25 | Method, system, equipment and readable storage medium for detecting faults of spindle bearing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113435322B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114305343A (en) * | 2022-01-18 | 2022-04-12 | 山东大学 | Single-lead electroencephalogram sleep staging method based on complementary set empirical mode decomposition |
| CN114648044A (en) * | 2022-03-18 | 2022-06-21 | 江苏迪普勒信息科技有限公司 | Vibration signal diagnosis and analysis method based on EEMD and depth domain countermeasure network |
| CN114795258A (en) * | 2022-04-18 | 2022-07-29 | 浙江大学 | Child hip joint dysplasia diagnosis system |
| CN114858467A (en) * | 2022-05-26 | 2022-08-05 | 上海交通大学 | Anti-noise and cross-noise-domain misfire diagnosis method and system for diesel engine |
| CN115326397A (en) * | 2022-07-28 | 2022-11-11 | 沈阳顺义科技有限公司 | Method for establishing crankshaft bearing wear degree prediction model and prediction method and related device |
| CN116728291A (en) * | 2023-08-16 | 2023-09-12 | 湖南大学 | Robot polishing system state monitoring method and device based on edge calculation |
| CN117973904A (en) * | 2024-03-29 | 2024-05-03 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334948A (en) * | 2018-02-09 | 2018-07-27 | 武汉理工大学 | A kind of mechanical bearing fault diagnosis technology based on wide residual error network learning model |
| CN111797567A (en) * | 2020-06-09 | 2020-10-20 | 合肥工业大学 | Deep learning network-based bearing fault classification method and system |
| WO2020244134A1 (en) * | 2019-06-05 | 2020-12-10 | 华南理工大学 | Multi-task feature sharing neural network-based intelligent fault diagnosis method |
-
2021
- 2021-06-25 CN CN202110713654.9A patent/CN113435322B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108334948A (en) * | 2018-02-09 | 2018-07-27 | 武汉理工大学 | A kind of mechanical bearing fault diagnosis technology based on wide residual error network learning model |
| WO2020244134A1 (en) * | 2019-06-05 | 2020-12-10 | 华南理工大学 | Multi-task feature sharing neural network-based intelligent fault diagnosis method |
| CN111797567A (en) * | 2020-06-09 | 2020-10-20 | 合肥工业大学 | Deep learning network-based bearing fault classification method and system |
Non-Patent Citations (2)
| Title |
|---|
| 王奉涛;邓刚;王洪涛;于晓光;韩清凯;李宏坤;: "基于EMD和SSAE的滚动轴承故障诊断方法", 振动工程学报, no. 02 * |
| 陈仁祥;汤宝平;杨黎霞;周广武;: "自适应参数优化EEMD机械故障特征提取方法", 振动.测试与诊断, no. 06 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114305343A (en) * | 2022-01-18 | 2022-04-12 | 山东大学 | Single-lead electroencephalogram sleep staging method based on complementary set empirical mode decomposition |
| CN114648044A (en) * | 2022-03-18 | 2022-06-21 | 江苏迪普勒信息科技有限公司 | Vibration signal diagnosis and analysis method based on EEMD and depth domain countermeasure network |
| CN114795258A (en) * | 2022-04-18 | 2022-07-29 | 浙江大学 | Child hip joint dysplasia diagnosis system |
| CN114858467A (en) * | 2022-05-26 | 2022-08-05 | 上海交通大学 | Anti-noise and cross-noise-domain misfire diagnosis method and system for diesel engine |
| CN114858467B (en) * | 2022-05-26 | 2023-05-26 | 上海交通大学 | Diesel engine anti-noise and cross-noise domain fire diagnosis method and system |
| CN115326397A (en) * | 2022-07-28 | 2022-11-11 | 沈阳顺义科技有限公司 | Method for establishing crankshaft bearing wear degree prediction model and prediction method and related device |
| CN115326397B (en) * | 2022-07-28 | 2023-10-27 | 沈阳顺义科技有限公司 | Method and related device for establishing crankshaft bearing wear degree prediction model and prediction method |
| CN116728291A (en) * | 2023-08-16 | 2023-09-12 | 湖南大学 | Robot polishing system state monitoring method and device based on edge calculation |
| CN116728291B (en) * | 2023-08-16 | 2023-10-31 | 湖南大学 | Robot polishing system state monitoring method and device based on edge calculation |
| CN117973904A (en) * | 2024-03-29 | 2024-05-03 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
| CN117973904B (en) * | 2024-03-29 | 2024-06-07 | 深圳市联特微电脑信息技术开发有限公司 | Intelligent manufacturing capacity analysis method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113435322B (en) | 2024-04-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113435322A (en) | Main shaft bearing fault detection method, system, equipment and readable storage medium | |
| Chen et al. | ACDIN: Bridging the gap between artificial and real bearing damages for bearing fault diagnosis | |
| CN108231201B (en) | Construction method, system and application method of disease data analysis processing model | |
| CN111797567B (en) | Bearing fault classification method and system based on deep learning network | |
| CN108334948B (en) | Mechanical bearing fault diagnosis technology based on wide residual error network learning model | |
| CN104155108B (en) | A kind of Fault Diagnosis of Roller Bearings based on vibration time frequency analysis | |
| CN108827605B (en) | Mechanical fault feature automatic extraction method based on improved sparse filtering | |
| CN111444988B (en) | Rolling bearing fault diagnosis system | |
| Zhang et al. | Enhanced sparse filtering with strong noise adaptability and its application on rotating machinery fault diagnosis | |
| CN110991424A (en) | Fault diagnosis method based on minimum entropy deconvolution and stacking sparse self-encoder | |
| CN109389171B (en) | Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology | |
| CN111397902B (en) | Rolling bearing fault diagnosis method based on feature alignment convolutional neural network | |
| Yu et al. | PCWGAN-GP: A new method for imbalanced fault diagnosis of machines | |
| CN112836604A (en) | Rolling bearing fault diagnosis and classification method, system and equipment based on VMD-SSAE and storage medium thereof | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| CN113822139A (en) | Equipment fault diagnosis method based on improved 1DCNN-BilSTM | |
| CN114861740B (en) | Self-adaptive mechanical fault diagnosis method and system based on multi-head attention mechanism | |
| Liu et al. | A rotor fault diagnosis method based on BP-Adaboost weighted by non-fuzzy solution coefficients | |
| CN112380932B (en) | Vibration signal characteristic value selection method and elevator health state evaluation or fault diagnosis method | |
| CN117110446A (en) | Method for identifying axle fatigue crack acoustic emission signal | |
| CN116429426A (en) | Bearing fault diagnosis method, device and medium for multi-domain feature fusion | |
| CN115326398B (en) | Bearing fault diagnosis method based on fuzzy width learning model | |
| CN116595465A (en) | High-dimensional sparse data outlier detection method and system based on self-encoder and data enhancement | |
| CN116296338A (en) | Rotary machine fault diagnosis method | |
| Xiao et al. | Health assessment for piston pump using LSTM neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |