Rolling bearing fault diagnosis model establishment method, diagnosis method, device and equipment

Abstract

The application relates to a rolling bearing fault diagnosis model establishment method, a diagnosis method, a device and equipment, and belongs to the technical field of bearing fault diagnosis. The rolling bearing fault diagnosis model establishment method comprises the following steps: acquiring measured sample data under various fault states of a rolling bearing, wherein the measured sample data comprises training sample data and test sample data; a BP neural network model is established; a particle swarm optimization algorithm is used to optimize connection weights and thresholds of the BP neural network model, so that global errors of the BP neural network model reach preset accuracy; the training sample data is input into the optimized BP neural network model for training, and test sample data is used to verify the training result, so that a rolling bearing fault diagnosis model is obtained. The application has the effect of improving the accuracy of a rolling bearing fault diagnosis result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bearing fault diagnosis, and in particular to a rolling bearing fault diagnosis model establishing method, a diagnosis method, a device and equipment. BACKGROUND

[0002] The rapid development of modern industrial technology makes its equipment increasingly large and complex. If defects occur in machines and equipment, the defects cannot be found and repaired in time, which will affect the production efficiency of the factory and will bring safety risks and significant economic losses. Therefore, how to diagnose mechanical defects in time and accurately has become an important research topic. Rolling bearings are widely used and are an important component of mechanical equipment. The failure of rolling bearings will affect the movement of machines and damage the machines. Therefore, it is necessary to monitor the running status of rolling bearings in real time and perform fault diagnosis.

[0003] At present, the fault prediction methods of rolling bearings based on machine learning include neural networks, fuzzy clustering, random forests, k-nearest neighbors, support vector machines, convolutional neural networks, etc. Among them, the traditional BP neural network is prone to local minimum value and slow convergence speed during training, which leads to unsatisfactory classification effect and inaccurate fault diagnosis results of rolling bearings. SUMMARY

[0004] In order to improve the accuracy of the fault diagnosis results of rolling bearings, the present application provides a rolling bearing fault diagnosis model establishing method, a diagnosis method, a device and equipment.

[0005] In the first aspect, the present application provides a rolling bearing fault diagnosis model establishing method, which adopts the following technical scheme:

[0006] Obtain measured sample data under various fault states of the rolling bearing, wherein the measured sample data includes training sample data and test sample data;

[0007] Establish a BP neural network model;

[0008] Optimize the connection weights and thresholds of the BP neural network model by using a particle swarm optimization algorithm, so that the global error of the BP neural network model reaches a preset precision;

[0009] Input the training sample data into the optimized BP neural network model for training, and use the test sample data to verify the training result to obtain a rolling bearing fault diagnosis model.

[0010] By adopting the technical scheme, the electronic device obtains measured sample data in various states of the rolling bearing, optimizes the BP neural network model by using a particle swarm optimization algorithm, trains the optimized model according to the training sample data, and verifies the training result by using test sample data, so as to complete the fault diagnosis model of the rolling bearing. Therefore, when training the fault diagnosis model, the particle swarm optimization algorithm can optimize the BP neural network model, improve the problem of local minimum, and improve the convergence speed, thereby improving the classification effect and the accuracy of the trained fault diagnosis model.

[0011] Further, the particle swarm optimization algorithm is used to optimize the connection weights and thresholds of the BP neural network model, including:

[0012] A population size, a preset accuracy value, and a maximum iteration number of the particle swarm are obtained.

[0013] The particle swarm is randomly obtained according to the population size, and the speed and position of the particle are initialized.

[0014] The connection weights and thresholds of a group of BP neural networks are determined according to the speed and position of each particle, the BP neural network is adjusted according to each group of connection weights and thresholds, the measured sample data is input into the BP neural network for training, a training error is calculated, an initial fitness of each particle is calculated according to the fitness function and the training error, an individual optimal solution of each particle and a group optimal solution of the particle swarm are determined according to the initial fitness, and it is judged whether the training error corresponding to the group optimal solution is less than the preset accuracy value. If yes, the group optimal solution is output.

[0015] Otherwise, the iteration step is repeatedly executed until the iteration number reaches the maximum iteration number or the training error corresponding to the group optimal solution reaches the preset accuracy value, and the group optimal solution is output.

[0016] The connection weights and thresholds of the optimized BP neural network model are determined according to the group optimal solution.

[0017] The iteration step includes:

[0018] The position and speed of the particle are updated according to the position and speed of the particle and the inertia weight in the last iteration.

[0019] The fitness of each particle is evaluated according to the fitness function, and the individual optimal solution of each particle and the group optimal solution of the particle swarm are determined according to the fitness.

[0020] The inertia weight is linearly reduced according to the increase of the iteration number.

[0021] By adopting the technical solutions, the electronic device obtains particles representing the connection weights and the threshold values of the BP neural network model, sets the population size, the preset accuracy and the maximum iteration number of the particle swarm, evaluates the fitness of each particle according to an adaptive function, evaluates the individual optimal solution of each particle and the group optimal solution of the particle swarm according to the fitness, changes the position and the speed of the particle through multiple iterations, simultaneously updates the individual optimal solution of the particle and the group optimal solution of the particle swarm, and ends the iteration until the ending condition is reached, so as to find the most suitable individual optimal solution and the group optimal solution of the particle swarm, and find the connection weights and the threshold values with the highest accuracy, thereby accelerating the convergence speed.

[0022] Further, after the fitness of each particle is evaluated according to the adaptive function, and the individual optimal solution of each particle and the group optimal solution of the particle swarm are determined according to the fitness, the method further includes:

[0023] determining whether the absolute value of the difference between the group optimal solutions of two adjacent iterations for N consecutive times is less than a minimum constant;

[0024] if the absolute value of the difference between the group optimal solutions of two adjacent iterations for N consecutive times is less than a minimum constant, performing local deep search for a plurality of steps by using the L-M algorithm to obtain a first search solution;

[0025] determining whether the fitness of the first search solution is less than the fitness of the group optimal solution, and if the fitness of the first search solution is less than the fitness of the group optimal solution, updating the group optimal solution according to the first search solution.

[0026] By adopting the technical solutions, the electronic device monitors each particle, and when the change of the particle is small for multiple times, it is possible that the particle is trapped in a local minimum value, so local deep search is performed near the group optimal solution to obtain a first search solution, and the first search solution is compared with the group optimal solution, and when the fitness of the first search solution is better than the fitness of the group optimal solution, the first search solution is replaced. Therefore, the diversity of the particle is improved, and the particle is prevented from being trapped in a local minimum value.

[0027] Further, if the absolute value of the difference between the group optimal solutions of two adjacent iterations for N consecutive times is not less than a minimum constant, the method further includes:

[0028] selecting any particle to perform L-M search to obtain a second search solution;

[0029] replacing the particle corresponding to the second search solution with a particle with the lowest fitness in the particle swarm.

[0030] By adopting the technical scheme, when the difference between the group optimal solutions corresponding to two adjacent iterations is not less than a preset minimum value, the electronic device selects any particle to perform deep search, obtains a second search solution, replaces the worst particle with the second search solution, and performs overall optimization of the particle group.

[0031] Further, if the fitness of the first search solution is not less than the fitness of the group optimal solution, the method further comprises:

[0032] selecting any particle to perform L-M search to obtain a second search solution;

[0033] replacing the particle corresponding to the second search solution with a particle having the lowest fitness in the particle group.

[0034] By adopting the technical scheme, if the first search solution is not suitable to replace the group optimal solution, any particle is selected to perform deep search to obtain a second search solution, and the particle corresponding to the second search solution replaces the worst particle to perform overall optimization of the particle group.

[0035] Further, if the absolute value of the difference between the group optimal solutions of two adjacent iterations does not continuously appear for N times and is less than a minimum value constant, the method further comprises:

[0036] determining whether the number of times that the absolute value of the difference between the group optimal solutions of two adjacent iterations continuously appears is less than a minimum value constant reaches a first preset value;

[0037] if yes, obtaining the absolute value of the difference corresponding to the most recent N times of continuous iterations;

[0038] determining a stable value of search according to each of the absolute values of the differences;

[0039] determining whether the stable value is less than a second preset value;

[0040] if the stable value is not less than the second preset value, obtaining a first number of times that the absolute value of the difference is less than the second preset value and a second number of times that the absolute value of the difference is not less than the second preset value;

[0041] calculating a ratio of the first number of times to the second number of times;

[0042] determining a number of particles corresponding to deep search according to the ratio;

[0043] replacing each particle after deep search with a bad particle;

[0044] otherwise, selecting any particle to perform L-M search to obtain a second search solution;

[0045] replacing the particle corresponding to the second search solution with a particle having the lowest fitness in the particle group.

[0046] By adopting the technical scheme, if the absolute value of the difference between the optimal solutions of two adjacent iterations of the population continuously appears less than the minimum value constant for a number of times reaching a first preset value, then the stable value of the search is calculated according to the number of times of the small difference in N iterations, the stable value is used for judgment to determine the number of particles for deep search, so that deep search is performed according to the trend of search into a local minimum value, and the overall diversity is improved.

[0047] Further, the acquisition of the measured sample data under various fault states of the rolling bearing includes:

[0048] The measured sample signal under various fault states of the rolling bearing is acquired.

[0049] The measured sample signal is decomposed by wavelet packets to obtain a frequency component energy spectrum of the signal.

[0050] 70% of the frequency component energy spectrum is taken as training sample data.

[0051] 30% of the frequency component energy spectrum is taken as test sample data.

[0052] In a second aspect, the application provides a fault diagnosis method of a rolling bearing, and the method includes:

[0053] A real-time vibration signal of a measured rolling bearing is acquired.

[0054] The real-time vibration signal is processed to obtain real-time data.

[0055] The real-time data is input into the fault diagnosis model of the rolling bearing in any one of the first aspect to perform identification, and a fault identification result of the rolling bearing is output.

[0056] By adopting the technical scheme, the electronic device acquires the real-time vibration signal of the measured rolling bearing, processes the real-time vibration signal to obtain real-time data, and inputs the real-time data into the trained fault diagnosis model of the rolling bearing to perform identification, so that the fault diagnosis model of the rolling bearing has high diagnosis accuracy and can output a more accurate fault identification result.

[0057] In a third aspect, the application provides a fault diagnosis device of a rolling bearing, and the following technical scheme is adopted:

[0058] A real-time vibration signal acquisition module is configured to acquire a real-time vibration signal of a rolling bearing.

[0059] A processing module is configured to process the real-time vibration signal to obtain real-time data.

[0060] a fault identification module configured to input the real-time data into the fault diagnosis model of the rolling bearing according to any one of claims 1-7 to perform identification, and output a fault identification result of the rolling bearing.

[0061] By using the above technical solution, the real-time vibration signal acquisition module acquires the real-time vibration signal of the measured rolling bearing, the processing module processes to obtain real-time data, and the fault identification module inputs the real-time data into the trained fault diagnosis model of the rolling bearing to perform identification. The fault diagnosis model of the rolling bearing has high diagnosis accuracy and can output more accurate fault identification results.

[0062] In a fourth aspect, the present application provides an electronic device, which adopts the following technical solution:

[0063] An electronic device comprises:

[0064] at least one processor;

[0065] a memory;

[0066] at least one computer program, wherein the at least one computer program is stored in the memory and configured to be executed by the at least one processor, and the at least one computer program is configured to perform the rolling bearing fault diagnosis model establishment method according to any one of the first aspect and the rolling bearing fault diagnosis method according to the second aspect.

[0067] By using the above technical solution, the processor loads and executes the computer program in the memory, acquires real measured sample data of the rolling bearing in various states, applies a particle swarm optimization algorithm to optimize the BP neural network model, trains the optimized model according to these training sample data, and verifies the training result with test sample data. The rolling bearing fault diagnosis model is trained, and then the real-time vibration signal of the measured rolling bearing is acquired, the real-time data is obtained after processing, the real-time data is input into the trained fault diagnosis model of the rolling bearing to perform identification, and more accurate fault identification results are output.

[0068] In summary, the present application has at least one of the following beneficial technical effects:

[0069] 1. Real measured sample data of the rolling bearing in various states is acquired, a particle swarm optimization algorithm is applied to optimize the BP neural network model, the optimized model is trained according to these training sample data, and the training result is verified with test sample data. The rolling bearing fault diagnosis model is trained, so that when the fault diagnosis model is trained, the particle swarm optimization algorithm can optimize the BP neural network model, improve the problem of local minimum value, improve the convergence speed, thereby improve the classification effect, and further improve the accuracy of the trained fault diagnosis model.

[0070] 2. The electronic device monitors each particle, when the particle changes less for continuous multiple times, performs a local deep search near the group optimal solution, obtains a first search solution, when the fitness of the first search solution is better than the fitness of the group optimal solution, performs replacement, improves the diversity of the particle, and avoids falling into a local minimum value;

[0071] 3. When the difference between the group optimal solutions corresponding to adjacent two iterations is not less than a preset threshold, the electronic device selects any particle to perform a deep search, obtains a second search solution, replaces the worst particle with the second search solution, and performs overall optimization of the particle group. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 FIG. 1 is a flowchart of a method for establishing a fault diagnosis model of a rolling bearing in an embodiment of the present application.

[0073] Figure 2 FIG. 2 is a flowchart of steps S11-S17 in an embodiment of the present application.

[0074] Figure 3 FIG. 3 is a flowchart of steps Sa-Sf in an embodiment of the present application.

[0075] Figure 4 FIG. 4 is a flowchart of a fault diagnosis method of a rolling bearing in an embodiment of the present application.

[0076] Figure 5 FIG. 5 is a flowchart of a fault diagnosis device of a rolling bearing in an embodiment of the present application.

[0077] Figure 6 FIG. 6 is a structural block diagram of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0078] In order to make the objectives, 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 some but not all of 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 scope of protection of the present application.

[0079] In addition, the term “and / or” in this paper is only to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the character “ / ” in this paper generally represents an “or” relationship between the associated objects unless otherwise specified.

[0080] The embodiment of the application discloses a fault diagnosis model establishment method of a rolling bearing. Figure 1 The electronic device can be a server or a terminal device. The server can be a physical server, a server cluster composed of multiple physical servers, a distributed system, or a cloud server providing cloud computing services. The terminal device can be a smartphone, a tablet computer, a desktop computer, or the like, but is not limited thereto. Figure 1 The method comprises the following steps:

[0081] Step S101: acquiring measured sample data of the rolling bearing in various fault states, the measured sample data comprising training sample data and test sample data, comprising (steps S1011-S1014):

[0082] Step S1011: acquiring measured sample signals of the rolling bearing in various fault states.

[0083] Specifically, the electronic device acquires the measured sample signals through a rolling bearing test system. The rolling bearing test system comprises a test device, and the application is not limited to the type and structure of the test device. Only the test device capable of acquiring the vibration signals of the rolling bearing is required.

[0084] In the application, the test device has three loading mechanisms, namely, a pre-tightening mechanism, an axial loading mechanism, and a radial loading mechanism.

[0085] The pre-tightening mechanism is used to change the opposite pre-tightening force of the two measured rolling bearings, so as to achieve the pre-tightening effect. The pre-tightening mechanism measures the pre-tightening force signals of the rolling bearings.

[0086] The axial loading mechanism is used to apply a load force to the axial direction of the rolling bearing. The axial loading mechanism measures the axial displacement signals of the rolling bearing.

[0087] The radial loading mechanism is used to apply a load force to the radial direction of the rolling bearing. The radial loading mechanism measures the radial load force signals of the rolling bearing.

[0088] Further, an exciter is installed on the test device, and the exciter is used to generate excitation. The above loading mechanisms acquire various detection signals under the condition that the exciter is started. The rolling bearing test system also measures the acceleration signals of the rolling bearing under the applied load.

[0089] The test device further comprises a signal conditioner and a data acquisition instrument. The acceleration signals, the pre-tightening force signals, the axial displacement signals, and the radial load force signals are transmitted to the signal conditioner. The signals processed by the signal conditioner are transmitted to the data acquisition instrument. Then, the electronic device acquires the measured sample signals of the rolling bearing through the data acquisition instrument.

[0090] The fault states of the rolling bearing in the application include four states of normal bearing, outer ring fault, inner ring fault and rolling element fault, a plurality of rolling bearings with faults are selected for each state, and a plurality of rolling bearings are tested multiple times, so that the bearing vibration signals of the rolling bearings in each state under different radial force conditions, the bearing vibration signals under different axial force conditions, the bearing vibration information under different pre-tightening force conditions, the bearing vibration signals under different fault frequency conditions and the bearing vibration signals under different fault frequency conditions can be obtained.

[0091] Step S1012: wavelet packet decomposition is performed on the measured sample signal by the wavelet packet to obtain a frequency energy spectrum of the signal.

[0092] Specifically, when the rolling bearing has faults such as fatigue, indentation or local corrosion during operation, impact vibration will be generated, and these vibrations repeatedly excite each other and cause greater vibration. The bearing vibration acceleration signal is obtained by using an acceleration sensor. In the energy of each frequency component of the signal, rich fault information is contained, and the change of the energy of one or several frequency components represents a fault, thereby forming a bearing fault feature vector.

[0093] If a bearing vibration acceleration signal x(t) is decomposed by a wavelet packet to the ith layer, the energy of each sub-band signal can be calculated according to formula (1).

[0094] The total energy calculation formula is as follows:

[0095]

[0096] Taking the percentage of the energy of each sub-band as the feature vector of the signal, the feature vector can be expressed as:

[0097]

[0098] In the application, i is set to 3. Therefore, the electronic device performs 3-layer wavelet packet decomposition on the bearing signal by the wavelet packet to obtain 8 groups of frequency bands corresponding to the frequency energy spectrum of the signal.

[0099] Step S1013: 70% of the frequency energy spectrum is taken as the training sample data.

[0100] Step S1014: 30% of the frequency energy spectrum is taken as the test sample data.

[0101] Specifically, the electronic device randomly selects 70% of the 8 frequency energy spectrums of the bearing as the training sample data, and the remaining 30% as the test sample data.

[0102] Step S102: a BP neural network model is established.

[0103] Specifically, the BP neural network model comprises an input layer, a hidden layer and an output layer, the input layer has i nodes, the hidden layer has j nodes, and the output layer has t nodes. Each neuron between the upper and lower layers is fully connected, that is, each unit of the lower layer is connected with each unit of the upper layer (i.e. weight value and threshold value vector).

[0104] The first layer of the BP neural network model is the input layer, and 8 groups of frequency component energy spectrums corresponding to frequency bands of the rolling bearing in various fault states are taken as input vectors. The nodes of the output layer are respectively four fault states of the bearing: normal bearing, outer ring fault, inner ring fault and rolling element fault. Corresponding output vectors are respectively (1, 0, 0, 0), (0, 1, 0, 0), (0, 0, 1, 0) and (0, 0, 0, 1).

[0105] According to the number of neurons in the hidden layer of the BP neural network, the empirical formula can determine the data of the hidden layer nodes:

[0106] In formula (3), h is the number of hidden layer nodes; m is the data of input layer nodes; n is the number of output layer nodes. In the present application, m = 8, n = 4, and a is an adjusting constant between 1 and 10. Therefore, the structure of the established neural network diagnosis model is 8-H-4.

[0107] Further, the algorithm steps of the BP neural network include:

[0108] (1) Set the initial weight coefficient w(0) to a small random non-zero value.

[0109] (2) Calculate the output of the network for a given input / output sample pair:

[0110] Let the p-th sample input and output be

[0111] u p =(u 1p ,u 2p ,…,u np )

[0112] d p =(d 1p ,d 2p ,…,d np )

[0113] wherein p = 1, 2, …, L, and L = 2 i .

[0114] The output of node i at the p-th sample input is

[0115] y ipf[x ip (t)]=f[∑ j w ij (t)I jp ] (4)

[0116] In formula (4), I jp is the jth input of node i when the pth group of samples is input; f is an excitation function, and a Sigmoid function is commonly used, that is:

[0117]

[0118] The electronic device substitutes x=x ip (t) into formula (5), so as to calculate y ip . Therefore, the output of the network output layer node can be obtained from the input layer through the hidden layer to the output layer.

[0119] (3) Calculate the target function J

[0120] Let E p be the target function of the network when the pth group of samples is input, and take the L2 norm, where the L2 norm is the square root of the sum of squares of all elements of a vector, so that

[0121]

[0122] In formula (6), y kp (t) is the output of the network after t times of weight value adjustment when the pth group of samples is input, and k is the kth node of the output layer. E p (t) is the output error of the network after t times of weight value adjustment after the pth group of samples is input. The total target function of the network is J(t)=∑ p E p (t), and J(t) is used as an evaluation of the learning condition of the network.

[0123] The electronic device gives an error ε, and ε≥0. If J≤ε, the algorithm ends, otherwise step (4) is performed.

[0124] (4) Back propagation calculation is performed from the output layer, the weight value is adjusted layer by layer according to J by the “gradient descent method”, and the neural network weight value is updated to reduce the error until J≤ε, and the algorithm ends.

[0125] In formula (7), η is a step size or a learning rate.

[0126] Step S103: The connection weight and the threshold value of the BP neural network model are optimized by using the particle swarm optimization algorithm, so that the global error of the BP neural network model reaches a preset precision.

[0127] In particular, the basic idea of the particle swarm optimization algorithm (PSO) is that each potential solution of an optimization problem corresponds to a bird in the search space, called a particle, and all the particles have a fitness value determined by an optimization function. Each particle also has a velocity vector that determines their direction and distance of flight, and then the particles search in the solution space following the current optimal particle.

[0128] PSO first initializes a swarm of random particles (random solutions), and then finds the optimal solution by iteration. In each iteration, the particle updates itself by tracking two "extremes". One of the extremes is the optimal solution found by the particle itself up to the current time, called the individual extreme value Pi; the other is the optimal solution found by the entire population up to the current time, called the global extreme value Pg. When the two optimal values are found, the particle updates its own speed and new position using its own information, individual extreme value and global extreme value.

[0129] Suppose there are N particles in a D-dimensional search space, where the position of the i-th particle (i = 1, 2,..., N) is X i = (x i1 , x i2 ,..., x iD ), and the velocity is V i = (v i1 , v i2 ,..., v iD ). Bring X i into the objective function J to calculate its fitness value, and measure the pros and cons of X i . Denote the optimal position searched by the i-th particle as the vector P i = (p i1 , p i2 ,..., p iD ), and the optimal position searched by the entire particle swarm as the vector P g = (p g1 , p g2 ,..., p gD ). The particle state update operation is as follows:

[0130]

[0131]

[0132] The above formulas (8) and (9) constitute the standard particle swarm optimization algorithm. In the formula, d = 1, 2,..., D; k is the current iteration number; learning factors c1 and c2 are non-negative constants, taking values between 0 and 2; r1 and r2 are random numbers between 0 and 1, and are independent of each other, used to maintain the diversity of the population; v id∈ [-Vmax, Vmax], Vmax is a constant set by the user, the velocity of the particle is limited in a maximum velocity Vmax range; the inertia weight ω is non-negative, which describes the influence of the last generation velocity on the current generation velocity, makes the particle keep the motion inertia, has the tendency to expand the search space, and has the ability to explore new areas; selecting a larger value of the inertia weight is conducive to improving the global search ability of the algorithm, while selecting a smaller value will enhance the local search ability of the algorithm; since different problems have different requirements for the global or local search ability of the algorithm, the balance between the global search ability and the local search ability of the algorithm can be adjusted, that is, the inertia weight can be automatically adjusted according to different problems. In order to make the algorithm have strong exploration ability at the beginning of iteration, new areas can be searched, and then the development ability is gradually enhanced, so that the algorithm can search around the possible optimal solution. Here, the linearly decreasing weight (LDW) strategy of linearly reducing the value of ω with iteration is adopted, that is:

[0133]

[0134] In formula (10), ω max is the maximum weight of ω, ω min is the minimum weight of ω, and ω max = 0.9-1.4, ω min = 0.4; k is the current evolution generation, and k max is the maximum evolution generation. When the set maximum evolution generation is reached, or the optimal position searched by the particle swarm meets the set minimum fitness threshold, the iteration is stopped.

[0135] Further, with reference to Figure 2 , step S103 includes (step S11-step S19):

[0136] Step S11: Obtain the population size of the particle swarm, the preset precision value, and the maximum iteration number.

[0137] Specifically, the population size is N, the preset precision value is e, and e can be equal to the given error ε or smaller than ε, e≥0; the maximum iteration number is k max .

[0138] Step S12: Randomly obtain the particle swarm according to the population size, and initialize the velocity and position of the particle.

[0139] Specifically, the electronic device obtains the particle swarm, each particle has a position and a direction, and the size of the vector can represent the position and direction of the particle. Each particle represents a set of connection weights and threshold values of the BP neural network model.

[0140] Step S13: Determine the connection weights and thresholds of each group of BP neural networks according to the speed and position of each particle, respectively adjust the BP neural networks according to each group of connection weights and thresholds, input the measured sample data into the BP neural networks for training, calculate the training error, calculate the initial fitness of each particle according to the fitness function and the training error, and determine the individual optimal solution of each particle and the group optimal solution of the particle group according to the initial fitness.

[0141] Specifically, after the measured sample data is input into the BP neural network for training, the electronic device obtains the error corresponding to each output, and then calculates the training error through the formula J.

[0142] The electronic device uses the mean square error MSE as the fitness function of PSO to evaluate the initial fitness of each particle, and the current position of each particle is taken as the optimal solution P i of the particle itself. g The position of the best particle in the initialized particle is set as the position of P

[0143] The mean square error is a statistical index used to measure the deviation of data from the true value in statistics and machine learning. Its formula is:

[0144]

[0145] The electronic device takes the mean square error of the training error obtained after training the BP neural network composed of particles as the fitness value.

[0146] Step S14: Determine whether the training error corresponding to the group optimal solution is less than the preset precision value. If yes, execute step S15 and step S19. Otherwise, repeat the iteration steps S16-S18 until the iteration number reaches the maximum iteration number or the training error corresponding to the group optimal solution reaches the preset precision value, and execute step S15 and step S19.

[0147] Step S15: Output the group optimal solution.

[0148] Step S19: Determine the connection weights and thresholds of the optimized BP neural network model according to the group optimal solution.

[0149] Specifically, when the iteration number reaches the maximum iteration number k max , or the training error of the group optimal solution is less than the preset precision value e, the optimization step is stopped, and it can be determined that the accuracy of the current model has reached the requirement, and the algorithm is ended.

[0150] The preset precision value e is set according to the actual situation and can be 10 -3 . The maximum iteration number k max may be 10 -6 .

[0151] Step S16: updating the position and velocity of the particle according to the position and velocity of the particle at the last iteration and the inertia weight.

[0152] Specifically, the electronic device makes all particles move again at the position after the last iteration, and updates the position and velocity of all particles.

[0153] Step S17: evaluating the fitness of each particle according to the fitness function, and determining the individual optimal solution of each particle and the group optimal solution of the particle group according to the fitness.

[0154] Specifically, the electronic device re-evaluates the fitness of each particle according to the fitness function, and determines the position corresponding to the smaller fitness of each particle as the individual optimal solution P i , and determines the position with the smallest fitness among all particles as the group optimal solution P g .

[0155] The smaller the fitness is, the smaller the mean square error of the representative vector is, and the more accurate the connection weight and threshold of the corresponding BP neural network are.

[0156] Step S18: linearly reducing the inertia weight according to the increase of the iteration number.

[0157] Specifically, the inertia weight is reduced as shown in the above formula (10).

[0158] Further, in order to avoid the PSO search process from falling into a local minimum value and failing to adjust the local extreme value, losing the particle diversity, with reference to Figure 3 , after step S17, the method further includes (steps Sa to step Sf):

[0159] Step Sa: determining whether the absolute value of the difference between the group optimal solutions of the adjacent two iterations is less than a minimum constant for consecutive N times; if yes, executing step Sb; otherwise, continuing to execute step S18.

[0160] Specifically, the minimum constant δ can be 10 -6 , and the consecutive number of times that the absolute value of the difference is less than the minimum constant and the consecutive number of times are pre-set by the electronic device. For example, if the absolute value of the difference between the adjacent two generations P g is less than the given minimum constant δ for 10 consecutive times, it means that the group optimal solution changes very little, and the current search falls into a local minimum value.

[0161] Step Sb: performing local deep search for a number of steps by using the L-M algorithm to obtain a first search solution.

[0162] In the derivative-based BP algorithm, the L-M (Levenberg-Marquardt) optimization algorithm is one of the most successful typical algorithms. The L-M algorithm is developed from the classical Newton algorithm, which uses the derivative deduced by the nonlinear least square method. The iteration formula of the L-M algorithm is as follows:

[0163]

[0164] In formula (12), I is a unit matrix, and λ is a non-negative value. By changing the amplitude of λ, the method smoothly changes between two extreme cases: the Newton method (λ→0) and the standard gradient method (when λ→∞). Therefore, the L-M algorithm actually combines the advantages of the Newton method and the standard gradient descent method.

[0165] Through the above algorithm, the electronic device performs P g The corresponding position and velocity are centered, and a local depth search is performed for several steps in a region with a preset length as a radius to obtain a first search solution. The maximum number of times of L-M local depth search can be 20 times.

[0166] Step Sc: It is judged whether the fitness of the first search solution is less than the fitness of the group optimal solution. If yes, step Sd is executed; otherwise, steps Se to Sd are executed.

[0167] Step Sd: The group optimal solution is updated according to the first search solution.

[0168] Specifically, when the fitness of the first search solution is less than the fitness of the group optimal solution, the first search solution is more accurate, and therefore the first search solution is determined as a new group optimal solution P g .

[0169] Step Se: Any particle is selected for L-M search to obtain a second search solution.

[0170] Step Sf: The particle corresponding to the second search solution is replaced with the particle with the lowest fitness in the particle swarm.

[0171] Specifically, the electronic device selects any particle as a center, performs depth search in a range with a preset length as a radius to obtain a second search solution, replaces the particle corresponding to the second search solution with the particle with the lowest fitness in the particle swarm, and performs local optimization on the particle.

[0172] In another possible implementation, if the absolute value of the difference between two adjacent iteration group optimal solutions is less than a minimum constant for N consecutive times, it is possible that the number of consecutive occurrences of the above phenomenon does not reach N times, that is, the current PSO search has a tendency to fall into a local minimum value. Therefore, the method further includes (steps Sg to Sn):

[0173] Step Sg: determining whether the number of times that the absolute value of the difference between the optimal solutions of two adjacent iterations in succession is less than the minimum constant value reaches a first preset value.

[0174] If yes, steps Sh-Sn are executed; otherwise, steps Se-Sf are executed.

[0175] Specifically, the first preset value is less than the continuous number of times.

[0176] Step Sh: obtaining the absolute values of the differences corresponding to the last N iterations in succession.

[0177] Step Si: determining the stability value searched according to the absolute values of the differences.

[0178] Specifically, the electronic device determines that the initial value of the calculation value is 0, first determines whether the second absolute value of the difference is less than the first absolute value of the difference, if yes, maintains the calculation value as 0, if not, increases the calculation value by 1; then repeatedly compares the next absolute value of the difference with the second absolute value of the difference, and updates the calculation value, until the comparison of all absolute values of the differences is completed, and the final calculation value is obtained.

[0179] The electronic device divides the calculation value by the number of absolute values of the differences to obtain the stability.

[0180] If the absolute values of the differences are successively reduced, the process of particle search is relatively stable, and the stability is relatively small. If the absolute values of the differences are not successively reduced, but change irregularly, the stability is relatively large.

[0181] Step Sj: determining whether the stability value is less than a second preset value. If not, step Sj is executed.

[0182] Specifically, the electronic device sets the second preset value, when the stability value is less than the second preset value, the deep search process is basically stable. If not, the deep search process is not stable, and can be optimized.

[0183] Step Sk: obtaining a first number of times that the absolute values of the differences are less than the second preset value, and a second number of times that the absolute values of the differences are not less than the second preset value.

[0184] Step Sl: calculating the ratio of the first number of times to the second number of times.

[0185] Step Sm: determining the number of particles corresponding to the deep search according to the ratio.

[0186] Specifically, the electronic device multiplies the ratio by a third preset value to obtain the number of particles. The third preset value is pre-stored by the electronic device, for example, 0.5N is set as the third preset value in the present application.

[0187] Step Sn: replace each particle searched by deep search with a bad particle.

[0188] Specifically, the electronic device acquires a plurality of bad particles with high fitness, the number of the bad particles being equal to the number of the particles corresponding to the deep search, and replaces each particle corresponding to the deep search with a bad particle one by one.

[0189] The application also provides a fault diagnosis method of a rolling bearing, referring to Figure 4 , comprising:

[0190] Step S201: acquiring a real-time vibration signal of a measured rolling bearing.

[0191] Specifically, the electronic device installs the measured rolling bearing on a bearing detection system, and acquires the real-time vibration signal of the measured rolling bearing by the bearing detection system.

[0192] Step S202: processing the real-time vibration signal to obtain real-time data.

[0193] Specifically, the electronic device performs wavelet packet decomposition on the acquired real-time vibration signal to obtain an energy spectrum of a frequency component of the signal as the real-time data.

[0194] Step S203: inputting the real-time data into a fault diagnosis model of the rolling bearing for recognition, and outputting a fault recognition result of the rolling bearing.

[0195] Specifically, the electronic device inputs the real-time data into the fault diagnosis model of the rolling bearing, and the fault diagnosis model of the rolling bearing obtains the output fault recognition result of the rolling bearing according to the input real-time data.

[0196] The application introduces a particle swarm optimization algorithm and an L-M algorithm to perform hybrid optimization on a neural network, forming a PSOLM hybrid particle swarm algorithm. The application has the following results after testing:

[0197] (1) The minimum training error of the PSOLM algorithm reaches 1 / 9 of the BP algorithm. It can be seen that the PSOLM algorithm has higher convergence accuracy than the BP algorithm, that is, the PSOLM algorithm has better learning ability than the BP algorithm and can obtain higher convergence accuracy.

[0198] (2) The fault recognition rate of the PSOLM algorithm for each single fault type is greater than 86%, which indicates that the PSOLM algorithm has superiority in optimizing the parameters of the neural network and shows strong generalization ability.

[0199] (3) The total positive judgment rate of the PSOLM algorithm is 91.67%, and compared with the BP algorithm (the total positive judgment rate is 78.67%), the method indeed improves the accuracy of fault diagnosis.

[0200] Table 1 Comparison of training and diagnosis results of two algorithms

[0201]

[0202] In summary, the PSOLM hybrid particle swarm algorithm proposed in the application fully gives play to the characteristics of PSO global search and L-M local fast optimization, makes up for the respective shortcomings, and realizes the complementary advantages of the algorithm. The PSOLM algorithm is used to optimize each parameter of the neural network, the modeling process is simple, has good convergence performance, avoids falling into a local minimum value, effectively finds a global optimal solution of the problem, has fast optimization performance, and has better efficiency and robustness than the BP algorithm. Diagnosis examples show that the PSOLM algorithm can well distinguish various fault modes, and improves the accuracy and reliability of bearing fault diagnosis.

[0203] In order to better implement the above method, the embodiment of the application further provides a fault diagnosis device of a rolling bearing, referring to Figure 5 , the fault diagnosis device of the rolling bearing 300 comprises:

[0204] A real-time vibration signal acquisition module 301 is configured to acquire a real-time vibration signal of the rolling bearing.

[0205] A processing module 302 is configured to process the real-time vibration signal to obtain real-time data.

[0206] A fault recognition module 303 is configured to input the real-time data into a fault diagnosis model of the rolling bearing for recognition, and output a fault recognition result of the rolling bearing.

[0207] The various changes and specific examples in the method in the foregoing embodiments are also applicable to the fault diagnosis device of the rolling bearing in the embodiment, and the implementation method of the fault diagnosis device of the rolling bearing in the embodiment can be clearly understood by those skilled in the art through the foregoing detailed description of the fault diagnosis method of the rolling bearing. Therefore, in order to make the description brief, the fault diagnosis device of the rolling bearing in the embodiment will not be described in detail here.

[0208] In order to better implement the above method, the embodiment of the application provides an electronic device, referring to Figure 6 , the electronic device 400 comprises a processor 401, a memory 403 and a display screen 405. The memory 403 and the display screen 405 are connected to the processor 401, such as being connected through a bus 402. Optionally, the electronic device 400 can further comprise a transceiver 404. It should be noted that the transceiver 404 is not limited to one in actual application, and the structure of the electronic device 400 does not constitute a limitation on the embodiments of the application.

[0209] The processor 401 can be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in combination with the disclosure. The processor 401 can also be a combination of computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0210] The bus 402 can include a path for transmitting information between the above-mentioned components. The bus 402 can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 402 can be divided into an address bus, a data bus, a control bus, etc.

[0211] The memory 403 can be a ROM (Read Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0212] The memory 403 is used to store application program code for implementing the scheme of the present application, and is controlled by the processor 401 to execute. The processor 401 is used to execute the application program code stored in the memory 403 to realize the content shown in the foregoing method embodiments.

[0213] Figure 6 The electronic device 400 shown is merely one example and should not be taken as limiting the scope of functionality or use of embodiments of the application.

[0214] The above are only preferred embodiments of the present application and are not intended to limit the protection scope of the present application, so: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

[0215] In addition, it should be understood that relative terms such as first and second are used only to distinguish one entity or action from another, and do not necessarily require or imply that any such entity or action be present or in any particular sequence. The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Claims

1. A method of establishing a failure diagnosis model of a rolling bearing, characterized by, The method comprises: Obtaining measured sample data of various fault states of the rolling bearing, the measured sample data comprising training sample data and test sample data; Establishing a BP neural network model; Inputting the training sample data into the BP neural network model for training; Optimizing the connection weights and thresholds of the BP neural network model by using a particle swarm optimization algorithm, so that the global error of the BP neural network model reaches a preset precision; Inputting the training sample data into the optimized BP neural network model for training, and verifying the training result by using the test sample data, to obtain a fault diagnosis model of the rolling bearing; The optimization of the connection weights and thresholds of the BP neural network model by using the particle swarm optimization algorithm comprises: Obtaining the population size, the preset precision value and the maximum iteration number of the particle swarm; Randomly obtaining the particle swarm according to the population size, and initializing the speed and position of the particles; Determining the connection weights and thresholds of a group of BP neural networks according to the speed and position of each particle, adjusting the BP neural networks according to the connection weights and thresholds of each group respectively, inputting the measured sample data into the BP neural networks for training, calculating the training error, calculating the initial fitness of each particle according to the fitness function and the training error, and determining the individual optimal solution of each particle and the group optimal solution of the particle swarm according to the initial fitness; Judging whether the training error corresponding to the group optimal solution is less than the preset precision value, and outputting the group optimal solution if yes; Otherwise, repeatedly performing the iteration step until the iteration number reaches the maximum iteration number or the training error corresponding to the group optimal solution reaches the preset precision value, and outputting the group optimal solution; Determining the connection weights and thresholds of the optimized BP neural network model according to the group optimal solution; The iteration step comprises: Updating the position and speed of the particle according to the position and speed of the particle and the inertia weight at the last iteration; Calculating the fitness of each particle, and determining the individual optimal solution of each particle and the group optimal solution of the particle swarm according to the fitness; Linearly reducing the inertia weight according to the increase of the iteration number; After determining the individual optimal solution of each particle and the group optimal solution of the particle swarm according to the initial fitness, the method further comprises: Judging whether the absolute value of the difference between two adjacent iteration group optimal solutions is less than a minimum value constant for N consecutive times; If the absolute value of the difference between two adjacent iteration group optimal solutions is less than a minimum value constant for N consecutive times, performing local deep search for several steps by using the L-M algorithm to obtain a first search solution; Judging whether the fitness of the first search solution is less than the fitness of the group optimal solution, and updating the group optimal solution according to the first search solution if the fitness of the first search solution is less than the fitness of the group optimal solution.

2. The method of claim 1, wherein, If the absolute value of the difference between two adjacent iteration group optimal solutions is not less than a minimum value constant for N consecutive times, the method further comprises: Selecting any particle to perform L-M search to obtain a second search solution; The particle corresponding to the second search solution is used to replace the particle with the lowest fitness in the particle swarm.

3. The method according to claim 1 or 2, characterized in that, If the fitness of the first search solution is not less than the fitness of the optimal solution of the population, the method further comprises: selecting any particle to perform L-M search to obtain a second search solution; The particle corresponding to the second search solution is used to replace the particle with the lowest fitness in the particle swarm.

4. The method of claim 1, wherein, If the absolute value of the difference between the optimal solutions of the population in two adjacent iterations does not reach a minimum value for N consecutive times, the method further comprises: determining whether the absolute value of the difference between the optimal solutions of the population in two adjacent iterations reaches a first preset value for consecutive times; if yes, obtaining the absolute values of the differences corresponding to the last N consecutive iterations; determining a stable value of the search according to the absolute values of the differences; determining whether the stable value is less than a second preset value; if the stable value is not less than the second preset value, obtaining a first number of times when the absolute value of the difference is less than the second preset value and a second number of times when the absolute value of the difference is not less than the second preset value; calculating a ratio of the first number of times to the second number of times; determining the number of particles corresponding to the deep search according to the ratio; replacing each particle with a bad particle after the deep search; otherwise, selecting any particle to perform L-M search to obtain a second search solution; The particle corresponding to the second search solution is used to replace the particle with the lowest fitness in the particle swarm.

5. The method of claim 1, wherein, The method comprises: obtaining measured sample data of the rolling bearing in various fault states; obtaining measured sample signals of the rolling bearing in various fault states; performing wavelet packet decomposition on the measured sample signals to obtain frequency component energy spectrum of the signals; taking 70% of the frequency component energy spectrum as training sample data; 6. A method of diagnosing a failure of a rolling bearing, characterized in that, taking 30% of the frequency component energy spectrum as test sample data. The method comprises: obtaining a real-time vibration signal of the rolling bearing; processing the real-time vibration signal to obtain real-time data; 7. A fault diagnosis device for rolling bearings, characterized in that, inputting the real-time data into the fault diagnosis model of the rolling bearing to perform identification, and outputting a fault identification result of the rolling bearing. The method comprises: a real-time vibration signal obtaining module, configured to obtain a real-time vibration signal of the rolling bearing; a processing module, configured to process the real-time vibration signal to obtain real-time data; 8. An electronic device, comprising: a fault identification module, configured to input the real-time data into the fault diagnosis model of the rolling bearing to perform identification, and output a fault identification result of the rolling bearing. The method comprises: at least one processor; a memory; at least one computer program, wherein the at least one computer program is stored in the memory and configured to be executed by the at least one processor, and the at least one computer program is configured to: perform the fault diagnosis model establishment method of the rolling bearing according to any one of claims 1 to 5 and the fault diagnosis method of the rolling bearing according to claim 6.

Images