Rolling bearing fault recognition method based on stochastic resonance and convolutional neural network

Abstract

The application discloses a rolling bearing fault recognition method based on stochastic resonance and a convolutional neural network, comprising the following steps: setting artificial fault points on inner and outer rings and rolling bodies of a rolling bearing, collecting bearing vibration signals of the rolling bearing under different rotating speeds, constructing an adaptive variable scale stochastic resonance detection system based on an improved signal-to-noise ratio, performing uniform scale processing on a horizontal axis on spectrum graphs respectively, composing an original image set by using the processed spectrum graphs, marking and classifying spectrum graphs in the original image set based on the artificial fault points, dividing the original image set after marking and classifying into a training set and a test set, building a convolutional neural network model and performing training and testing, determining parameters of the convolutional neural network model according to a test result and representing the parameters as a recognition model, and performing fault recognition on acquired bearing vibration signals to be recognized based on the recognition model and acquiring a fault type. The application solves the problem of weak fault recognition of the rolling bearing under strong noise and variable rotating speed working conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image recognition, and particularly relates to a rolling bearing fault recognition method based on random resonance and a convolutional neural network. BACKGROUND

[0002] The rolling bearing is an important part widely used in mechanical equipment, which is mainly used for supporting rotating shaft parts and needs to bear complex and cyclically repeated alternating stress for a long time. Therefore, the rolling bearing is a kind of wearing parts, and the running state of the rolling bearing will play a crucial role in the healthy and reliable operation of the whole equipment. Therefore, it has very important practical significance to diagnose the fault of the rolling bearing, especially the early fault of the rolling bearing.

[0003] The surface damage type fault (such as pitting, peeling and scratch, etc.) is a common fault in the operation process of the rolling bearing. For this kind of fault, the working surface damage point of the bearing element repeatedly hits the surface of the other element in contact during the operation process, which will produce "through vibration". The occurrence period is regular, and the frequency produced by different fault positions is different. Detecting these signals is an effective means to diagnose the bearing fault. Usually, the fault of the rolling bearing is diagnosed by experts in various fields according to professional experience. The emergence of shallow intelligent models such as neural network and support vector machine replaces the traditional artificial diagnosis process and opens the intelligent era of fault diagnosis. However, since the fault diagnosis of the rolling bearing is usually in the early stage, the fault signal is often accompanied by large noise, and the working condition of the rolling bearing is relatively complex, and the working condition is often unstable, such as fluctuation of rotating speed, which leads to fluctuation of the fault frequency of the rolling bearing within a certain range, and the accuracy of identifying the fault type of the rolling bearing is low. SUMMARY

[0004] The present application provides a rolling bearing fault recognition method based on random resonance and a convolutional neural network to overcome the above technical problems.

[0005] The rolling bearing fault recognition method based on random resonance and a convolutional neural network comprises

[0006] S1, artificial fault points are arranged on the inner and outer rings and rolling elements of the rolling bearing, and bearing vibration signals of the rolling bearing at different rotating speeds are collected based on an acceleration sensor above the rolling bearing,

[0007] S2, an adaptive variable scale random resonance detection system based on improved signal-to-noise ratio is constructed, and the adaptive variable scale random resonance detection system is used for detecting the collected bearing vibration signals to obtain a frequency spectrum corresponding to the artificial fault points,

[0008] S3, respectively, the horizontal axis of the spectrum graph is uniformly scaled, the processed spectrum graph is composed of the original image set, the spectrum graph in the original image set is marked and classified based on the artificial fault point, and the original image set after marking and classification is divided into a training set and a test set,

[0009] S4, a convolutional neural network model is built, the convolutional neural network model is trained based on the training set, the trained convolutional neural network model is tested based on the test set and the test result is obtained, the parameters of the convolutional neural network model are determined according to the test result and are represented as an identification model,

[0010] S5, obtaining the bearing vibration signal to be identified, identifying the fault of the obtained bearing vibration signal to be identified based on the identification model and obtaining the fault type.

[0011] Preferably, the S2 comprises,

[0012] S21, a particle swarm optimization model is built, including setting the number of particles as M, the maximum number of iterations as N, the search range of the particle and the maximum search speed, the objective function and the detection range and detection position of the objective function, and initializing the particle swarm optimization model,

[0013] S22, obtaining the bearing vibration signal, setting the classification condition based on amplitude, frequency or noise, classifying the bearing vibration signal based on the classification condition, determining the variable scale stochastic resonance solving method, and decomposing the classified bearing vibration signal in the bearing vibration signal according to the variable scale stochastic resonance solving method,

[0014] S23, selecting a stochastic resonance system, inputting the M particles in the initialized particle swarm optimization model and the variable scale decomposed bearing vibration signal into the stochastic resonance system and obtaining system parameters, and obtaining the optimal system parameters after N iterations,

[0015] S24, based on the stochastic resonance system of the optimal system parameters, inputting the bearing vibration signal into the stochastic resonance system to obtain the spectrum graph corresponding to the artificial fault point.

[0016] Preferably, the variable scale stochastic resonance solving method is the Runge-Kutta method.

[0017] Preferably, the horizontal axis of the spectrum graph is uniformly scaled, the total length of the coordinate axis of the horizontal axis of the spectrum graph is a fixed multiple of the rotating speed, that is, the horizontal axis range is 0-nr, n is a constant value, r is the rotating speed, and the three kinds of fault characteristic frequencies are divided by nr respectively to represent the horizontal axis, that is, the position of the fault point in the spectrum graph does not change with the change of the rotating speed.

[0018] Preferably, the stochastic resonance system is a bistable stochastic resonance system.

[0019] This invention provides a rolling bearing fault identification method based on stochastic resonance and convolutional neural networks. An adaptive variable-scale stochastic resonance detection system detects the collected vibration signals. The spectrum output from this system is then processed by standardizing the horizontal axis to construct a training set. A convolutional neural network model is used for training, enabling fault identification in the early stages of rolling bearing failures, even under strong noise conditions. The improved signal-to-noise ratio adaptive variable-scale stochastic resonance system can detect fault signals within a certain range, thus avoiding interference from unstable operating conditions. The standardized horizontal axis processing of the stochastic resonance output spectrum avoids the decrease in identification accuracy caused by speed variations, solving the problem of identifying weak rolling bearing faults under strong noise and speed variation conditions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the rolling bearing fault identification method of the present invention;

[0022] Figure 2 This is a random resonance detection spectrum diagram of the inner ring, outer ring, and rolling element faults in an embodiment of the present invention;

[0023] Figure 3 (a) is a random resonance spectrum diagram of a rolling element fault generated at a rotational speed of 100 r / min without undergoing uniform horizontal axis scaling, where the arrow points to the detected fault characteristic frequency;

[0024] Figure 3 (b) is a random resonance spectrum diagram of a rolling element fault generated at a rotational speed of 200 r / min without undergoing uniform horizontal axis scaling, where the arrows point to the detected fault characteristic frequencies.

[0025] Figure 4 (a) is a random resonance spectrum diagram of rolling element faults generated at a rotational speed of 100 r / min and processed with uniform horizontal axis scale in an embodiment of the present invention.

[0026] Figure 4 (b) is a random resonance spectrum diagram of rolling element faults generated at a rotational speed of 200 r / min and processed with uniform horizontal axis. Detailed Implementation

[0027] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0028] Figure 1 For the method flowchart of the present application, as shown in Figure 1 The method of the present embodiment includes:

[0029] S1, artificial fault points are arranged on the inner and outer rings and rolling elements of the rolling bearing, and bearing vibration signals of the rolling bearing at different rotating speeds are collected based on an acceleration sensor above the rolling bearing,

[0030] S2, an adaptive scale random resonance detection system based on improved signal-to-noise ratio is constructed, and the adaptive scale random resonance detection system is used to detect the collected bearing vibration signals to obtain a frequency spectrum corresponding to the artificial fault points,

[0031] S3, the frequency spectrum is processed in a uniform scale of the horizontal axis respectively, the processed frequency spectrum is composed into an original image set, the frequency spectrum in the original image set is marked and classified based on the artificial fault points, and the original image set after marking and classification is divided into a training set and a test set,

[0032] S4, a convolutional neural network model is built, the convolutional neural network model is trained based on the training set, the trained convolutional neural network model is tested based on the test set to obtain a test result, parameters of the convolutional neural network model are determined according to the test result and expressed as an identification model,

[0033] S5, the bearing vibration signals to be identified are obtained, and the bearing vibration signals to be identified are fault-identified based on the identification model to obtain a fault type.

[0034] The application provides a rolling bearing fault identification method based on stochastic resonance and a convolutional neural network, detects collected vibration signals through an adaptive variable-scale stochastic resonance system, then constructs a training set after frequency spectrum output by the stochastic resonance system is processed through uniform scale processing of a horizontal axis, and trains a convolutional neural network model, so that fault identification can be performed in the early stage of rolling bearing fault, and identification can also be completed under strong noise. The adaptive variable-scale stochastic resonance system based on improved signal-to-noise ratio can detect fault signals in a range, so that the interference of unstable working conditions can be avoided. After the frequency spectrum output by the stochastic resonance is processed through uniform scale processing of the horizontal axis, the problem of decreased identification accuracy caused by speed variation can be avoided, and the problem of weak rolling bearing fault identification under strong noise and speed variation working conditions can be solved.

[0035] Specifically, the embodiment gives detailed contents of the rolling bearing fault identification method, including

[0036] S1, artificial fault points are arranged on the inner and outer rings and rolling bodies of the rolling bearing, and bearing vibration signals of the rolling bearing under different speeds are collected based on an acceleration sensor above the rolling bearing. Specifically, artificial faults with a width of 0.5 mm are cut on the inner and outer rings and rolling bodies of the rolling bearing by a wire cut electrical discharge machine to simulate weak faults in the actual operation of the rolling bearing. The vibration signals of different bearing faults are collected by an acceleration sensor arranged above the bearing seat of the rolling bearing. The speed regulator is used to adjust the bearing speed from 40 r / min to 200 r / min at a step of 20 r / min, and the bearing vibration signals of different faults are collected under each speed condition for 10 min.

[0037] S2, an adaptive variable-scale stochastic resonance detection system based on improved signal-to-noise ratio is constructed, and the adaptive variable-scale stochastic resonance detection system is used to detect the collected bearing vibration signals to obtain a frequency spectrum corresponding to the artificial fault point,

[0038] The S2 includes:

[0039] S21, a particle swarm optimization model is built, including setting the number of particles as M, the maximum number of iterations as N, the search range of the particle and the maximum search speed, the objective function and the detection range and position of the objective function, and initializing the particle swarm optimization model,

[0040] S22, the bearing vibration signals are obtained, the classification conditions based on amplitude, frequency or noise are set, the bearing vibration signals are classified based on the classification conditions, the variable-scale stochastic resonance solving method is determined, and the classified bearing vibration signals in the bearing vibration signals are decomposed according to the variable-scale stochastic resonance solving method.

[0041] Specifically, the bearing vibration signal is divided into small-scale signals and large-scale signals based on amplitude, frequency or noise, the small-scale signal refers to a small amplitude, small frequency, small noise signal, the large-scale signal refers to a large amplitude, large frequency, large noise signal, an amplitude classification threshold, a frequency classification threshold and a noise classification threshold are set, the signal with an amplitude less than the amplitude classification threshold is represented as a small-scale signal, the signal with an amplitude greater than or equal to the amplitude classification threshold is represented as a large-scale signal, the signal with a frequency less than the frequency classification threshold is represented as a small-scale signal, the signal with a frequency greater than or equal to the frequency classification threshold is represented as a large-scale signal, the signal with noise less than the noise classification threshold is represented as a small-scale signal, and the signal with noise greater than or equal to the noise classification threshold is represented as a large-scale signal. Since the classical stochastic resonance system is constrained by the adiabatic approximation theory, it is only applicable to small parameter signal conditions, but the fault signals in actual engineering applications are usually large parameter signals, so it is necessary to transform the large parameter signals into small-scale signals by the variable-scale stochastic resonance solving method. The variable-scale stochastic resonance solving method is the Runge-Kutta method. Since the transformation scale R is too large, the iteration step is too large, and the iterative solution does not converge, the embodiment dynamically adjusts the transformation scale R according to the Runge-Kutta method, so that the iterative solution tends to converge.

[0042] S23, selecting a stochastic resonance system, the stochastic resonance system being a bistable stochastic resonance system, inputting each particle after initialization and the bearing vibration signal after variable-scale decomposition into the stochastic resonance system and obtaining system parameters, obtaining optimal system parameters after N iterations,

[0043] S24, based on the stochastic resonance system with optimal system parameters, inputting the bearing vibration signal into the stochastic resonance system to obtain a frequency spectrum corresponding to the artificial fault point.

[0044] Specifically, the maximum search speed of the particle is taken from 20% of the variable feasible region, and the number of particles, i.e. the population number, is 50, wherein the variable is the stochastic resonance system parameters a and b. The objective function of the particle swarm optimization algorithm, the detection range and the detection position of the objective function are determined; wherein the objective function of the particle swarm optimization algorithm is the improved signal-to-noise ratio (ISNR), and the ISNR is represented as:

[0045]

[0046] In the formula: ft is the fault theoretical frequency; S(ft) represents the total energy in the interval [ft-lΔf, ft+lΔf] with ft as the center, l represents the detection range, Δf represents the proportion value, and N(ft) represents the total noise energy excluding S(ft).

[0047] Further, the detection range is determined according to the detection position, and the detection position selection formula is:

[0048] kt=fr(K / fs) (2)

[0049] In the formula: kt is the detection position, K is the number of sampling points, and fs is the sampling frequency; kt is an integer, and when kt has a decimal, it is rounded off.

[0050] Randomly initialize the particle's position x i (0) and velocity v i (0), where each position represents a set of system parameters a and b for random resonance, and the velocity represents the update step size of system parameters a and b. The individual extreme value pbest of each particle. i Set the position to the optimal position calculated in the current iteration, and set the position with the largest individual extreme value in the entire particle swarm to gbest. i ;

[0051] The positions of each particle and the preprocessed vibration signals are input into the stochastic resonance system, and the improved signal-to-noise ratio is selected as the fitness function. The fitness value of each particle is then calculated. If the fitness value of a particle exceeds the fitness value of the current individual extreme value, then the position of that particle is taken as the new individual extreme value, pbest. i If the maximum value of an individual extreme value in the entire particle swarm exceeds the current global optimum, then that individual extreme value is taken as the new global optimum, gbest. i Based on the obtained individual extreme values ​​and global optimal values, the velocity and position of all particles are recalculated; it is determined whether the current iteration number has reached the maximum iteration number N. If the termination condition is met, the optimal system parameters a and b are directly output. The optimal system parameters a and b obtained in the previous step and the bearing vibration signal are input into the bistable random resonance system to detect the weak periodic signal.

[0052] The stochastic resonance model mainly includes three elements: a nonlinear system, a weak periodic signal, and noise. In this embodiment, the nonlinear system used in the stochastic resonance system is a bistable system.

[0053] S3. Perform horizontal axis scaling on the spectrum graphs respectively. The horizontal axis scaling on the spectrum graphs includes making the total length of the horizontal axis coordinate axis of the spectrum graphs a fixed multiple of the rotational speed, that is, the horizontal axis range is (0-nr), where n is a fixed value and r is the rotational speed. The three fault characteristic frequencies are divided by nr and then represented as the horizontal axis. That is, the position of the fault point in the spectrum graph does not change with the rotational speed.

[0054] The processed spectrograms were then used to form the original image set. Based on artificially identified fault points, the spectrograms in the original image set were labeled and classified. The labeled and classified original image set was then divided into a training set and a test set.

[0055] Specifically, Figure 2is the fault stochastic resonance detection spectrum of inner ring, outer ring and rolling body in the present application. Taking 22209 bearing as an example, according to the theoretical fault frequency calculation formula, the theoretical fault frequency calculation formula is shown in formula (3), (4), (5):

[0056] Inner ring fault characteristic frequency:

[0057]

[0058] Outer ring fault characteristic frequency:

[0059]

[0060] Rolling body single fault frequency:

[0061]

[0062] Wherein Z is the number of rolling bodies, f r is the rotating frequency, d is the diameter of rolling body, D is the pitch diameter of bearing, and alpha represents the contact angle of bearing. Three kinds of fault characteristic frequencies can be calculated as: f I = 0.2538r, f o = 0.3462r, f c = 0.0529r; the fault characteristic frequency is fixed in the spectrum to avoid the interference of rotating speed change.

[0063] Further, the total length of the coordinate axis of the horizontal axis of the stochastic resonance output spectrum is a fixed multiple of the rotating speed, that is, the horizontal axis range is (0-nr), so the positions of the three kinds of fault characteristic frequencies in the spectrum can be expressed as f I / nr = 0.2538n, f o / nr = 0.3462n, f c / nr = 0.0529n, n is a constant; that is, the position of the fault characteristic frequency spectrum in the spectrum does not change with the change of rotating speed. Figure 3 (a) is the rolling body fault stochastic resonance spectrum generated in the present application without horizontal axis uniform scale processing under the rotating speed of 100r / min, wherein the arrow points to the detected fault characteristic frequency; Figure 3 (b) is the rolling body fault stochastic resonance spectrum generated in the present application without horizontal axis uniform scale processing under the rotating speed of 200r / min, wherein the arrow points to the detected fault characteristic frequency; Figure 4 (a) is the rolling body fault stochastic resonance spectrum generated in the present application with horizontal axis uniform scale processing under the rotating speed of 100r / min; Figure 4 (b) is the rolling body fault stochastic resonance spectrum generated in the present application with horizontal axis uniform scale processing under the rotating speed of 200r / min.

[0064] The original image set is classified according to the fault category and the corresponding label is set, a training set is formed, and 30% of the images in the original image set are randomly extracted to form a test set; the images of the training set and the test set are subjected to uniform scale processing on the horizontal axis, that is, the horizontal axis range is (0-nr), and r is the rotating speed.

[0065] S4, a convolutional neural network model is constructed, the convolutional neural network model is trained based on the training set, the trained convolutional neural network model is tested based on the test set and a test result is obtained, parameters of the convolutional neural network model are determined according to the test result and are expressed as an identification model,

[0066] Specifically, the convolutional neural network model is composed of a convolutional layer 1, an activation layer 1, a pooling layer 1, a convolutional layer 2, an activation layer 2, a convolutional layer 3, an activation layer 3, a pooling layer 2, a deep concatenation layer 1, a deep concatenation layer 2, a pooling layer 3, a deep concatenation layer 3, a deep concatenation layer 4, a deep concatenation layer 5, a deep concatenation layer 6, a pooling layer 4, a deep concatenation layer 7, a deep concatenation layer 8, a pooling layer 5, a full connection layer, and a classification output layer.

[0067] Each convolutional layer: feature extraction is performed on the images in the set to obtain a feature map;

[0068] Each activation layer: non-linear mapping is performed on the results of the corresponding convolutional layer;

[0069] Each pooling layer: the features extracted by the corresponding convolutional layer are selected;

[0070] Each deep concatenation layer: the results processed by the corresponding pooling layer are concatenated;

[0071] Full connection layer: all feature matrices of the pooling layer are converted into a one-dimensional feature vector;

[0072] If the test result of the test set and the training result of the training set differ by more than 1% compared with a threshold value, the learning rate, the sample number, and the number of training rounds are changed and the training is performed again.

[0073] S5, obtaining a bearing vibration signal to be identified, performing fault identification on the obtained bearing vibration signal to be identified based on the identification model and obtaining a fault category.

[0074] The overall beneficial effects are:

[0075] The application provides a rolling bearing fault identification method based on random resonance and a convolutional neural network. The collected vibration signals are detected by an adaptive variable scale random resonance system, then the frequency spectrum output by the random resonance system is processed by uniform scale processing of the horizontal axis to construct a training set, and the convolutional neural network model is trained, so that the fault identification can be performed in the early stage of rolling bearing fault, and the identification work can also be completed under strong noise. The adaptive variable scale random resonance system based on the improved signal-to-noise ratio can detect the fault signals in a range, so that the interference of unstable working conditions can be avoided. After the random resonance output frequency spectrum is processed by uniform scale processing of the horizontal axis, the problem of the decrease of the identification accuracy caused by the speed variation can be avoided, and the weak rolling bearing fault identification problem under the strong noise and speed variation working conditions can be solved.

[0076] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying rolling bearing faults based on stochastic resonance and convolutional neural networks, characterized in that, include S1. Artificial fault points are set on the inner and outer rings and rolling elements of the rolling bearing. The bearing vibration signals at different speeds are collected by an acceleration sensor above the rolling bearing. S2. Construct an adaptive variable-scale stochastic resonance detection system based on improved signal-to-noise ratio. This system is used to detect the acquired bearing vibration signals to obtain a spectrum corresponding to the artificial fault point. Specifically, S2 includes... S21. Construct a particle swarm optimization model, including setting the number of particles to M, the maximum number of iterations to N, the search range between the particles and the maximum search velocity, the objective function, and the detection range and detection position of the objective function, and initialize the particle swarm optimization model. S22. Obtain the bearing vibration signal, set classification conditions based on amplitude, frequency or noise, classify the bearing vibration signal based on the classification conditions, determine the variable-scale stochastic resonance solution method, and decompose the classified bearing vibration signal in the bearing vibration signal according to the variable-scale stochastic resonance solution method. The variable-scale stochastic resonance solution method is the Runge-Kutta method. S23. Select a stochastic resonance system. Input the M particles from the initialized particle swarm optimization model and the bearing vibration signal after variable-scale decomposition into the stochastic resonance system and obtain the system parameters. Obtain the optimal system parameters after N iterations. S24. Based on the optimal system parameters, the bearing vibration signal is input into the optimal system parameters of the stochastic resonance system to obtain the spectrum diagram corresponding to the artificial fault point. S3. Perform horizontal axis scaling on the spectrum graphs separately, and assemble the processed spectrum graphs into an original image set. Mark and classify the spectrum graphs in the original image set based on artificial fault points. Divide the marked and classified original image set into a training set and a test set. The horizontal axis scaling on the spectrum graphs includes setting the total length of the horizontal axis coordinate axis of the spectrum graph to a fixed multiple of the rotational speed, that is, the horizontal axis range is 0-nr, where n is a constant value and r is the rotational speed. Divide the three fault characteristic frequencies by nr respectively and represent them as the horizontal axis. That is, the position of the fault point in the spectrum graph does not change with the rotational speed. S4. Build a convolutional neural network model. Train the convolutional neural network model using the training set, test the trained convolutional neural network model using the test set, and obtain the test results. Determine the parameters of the convolutional neural network model based on the test results and represent it as a recognition model. S5. Obtain the bearing vibration signal to be identified, and perform fault identification on the obtained bearing vibration signal based on the identification model to obtain the fault type.

2. The rolling bearing fault identification method based on stochastic resonance and convolutional neural network according to claim 1, characterized in that, The stochastic resonance system is a bistable stochastic resonance system.

Images