Dual-objective optimization method for fault sign enhancement of same-direction linearly coupled biological neurons

Abstract

The present invention relates to a method for enhancing the fault symptom of a dual-target optimized co-directional linear coupled biological neuron. A co-directional linear coupled biological neuron module is constructed by a co-directional linear coupled hyperbolic tangent biological neuron and a Gaussian biological neuron. Then, a characteristic frequency amplitude ratio measurement index and a dwell time distribution index are designed based on the model response as the dual-target synchronous optimization function of the genetic algorithm to optimize the parameters of the co-directional linear coupled biological neuron model, realizing the extraction of weak symptom features of strong noise and strong coupling machinery under complex working conditions and timely diagnosing the early weak faults of mechanical equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and more particularly, to a method for enhancing mechanical fault symptom based on a double-objective optimized co-directional linear coupled biological neuron Background Art

[0002] With the accelerating advancement of intelligent manufacturing technology in China, mechanical equipment is constantly developing towards the directions of large-scale, precision, intelligence, digitization, and networking. Its operating conditions are becoming increasingly complex and harsh, accelerating the fatigue failure of key components of mechanical equipment. Coupled with poor lubrication, improper operation, etc., accidents are extremely likely to occur, endangering the safety of the equipment. Mechanical condition monitoring and fault diagnosis technology can timely predict equipment failure, ensure its safe and efficient operation, and avoid major accidents

[0003] Extracting weak symptom features is the core of mechanical condition monitoring and fault diagnosis. The weak symptom feature extraction technology based on eliminating or suppressing noise cannot eliminate or suppress strong background noise in the same frequency band, and the filter design depends on accurate parameter selection. If the filter parameters are selected improperly and the resonance frequency band excited by the fault is selected incorrectly, the weak symptom feature extraction fails. Moreover, even if the resonance frequency band is selected correctly, the weak symptom feature is aliased with strong background noise in the same frequency, and it is difficult for the filter to effectively filter out the weak symptom feature. On the contrary, stochastic resonance can use noise to enhance weak symptom features and is widely used in mechanical weak fault diagnosis. However, in the face of extracting strong noise and strongly coupled mechanical weak symptom features under complex working conditions, the existing stochastic resonance technology seems powerless. As is well known, tens of thousands of biological neurons can form a neural network through synaptic coupling, and continuously enhance weak symptom features by triggering resonance layer by layer using noise, so as to be able to sense extremely weak useful information Summary of the Invention

[0004] The problem solved by the present invention is how to improve the noise utilization ability of stochastic resonance, realize the extraction of strong noise and strongly coupled mechanical weak symptom features under complex conditions, and timely diagnose early weak faults of mechanical equipment

[0005] To solve the above problems, the present invention provides a method for enhancing mechanical fault symptom based on a double-objective optimized co-directional linear coupled biological neuron, including the following steps:

[0006] Step 1: Construct a co-directional linear coupled biological neuron model using hyperbolic tangent biological neurons and Gaussian biological neurons, and initialize the parameters of the co-directional linear coupled biological neuron model and the variable scale factor;

[0007] Step 2: Input the initialized parameters of the co-directional linear coupled biological neuron model into the co-directional linear coupled biological neuron model, and solve the model response and step size;

[0008] Step 3, establishing a characteristic frequency amplitude ratio measurement index and a residence time distribution ratio index according to the model response of the same-direction linearly coupled biological neuron model;

[0009] Step 4: Use the dual-objective synchronous optimization function to optimize the parameters and variable scale factors of the same-direction linear coupled biological neuron model

[0010] The beneficial effects of the present invention are as follows: a same-direction linear coupling biological neuron module is constructed by using same-direction linear coupling hyperbolic tangent biological neurons and Gaussian biological neurons, and then a characteristic frequency amplitude ratio measurement index and a residence time distribution index are designed according to the model response. The parameters of the same-direction linear coupling biological neuron model are used as a dual-objective synchronous optimization function of the genetic algorithm to optimize the same-direction linear coupling biological neuron model parameters, thereby realizing the extraction of weak sign features of strong noise and strong coupling machinery under complex working conditions, and timely diagnosing early weak faults of mechanical equipment.

[0011] Preferably, the same-direction linear coupling biological neuron model constructed in step 1 is:

[0012]

[0013] In the formula, ω f and λ is the hyperbolic tangent of biological neuron U1(x)=ω f x 2 / 2-λln(coshx) is the control parameter, and ω f >0,λ>0,control ω f The λ hyperbolic tangent biological neuron switches between monostable and bistable states to achieve the matching between the mechanical signal in(t) to be detected and the steady-state type; a, b and R are Gaussian biological neurons U2(x) = ax 2 / 2+bexp(-x 2 / R 2 ) and a>0, b>0, R>0. The control parameters can realize the switching between monostable and bistable states. The linear coupling strength in the same direction is δ and -1<δ<1. The sampling frequency of the mechanical signal in(t) to be detected is f s , the number of sampling points is N.

[0014] Preferably, the parameters and scaling factors of the same-direction linear coupling biological neuron model initialized in step 1 are:

[0015] ω f ,λ,a,b,R∈(0,10),δ∈(-1,1),G∈(0,f s / 2)

[0016] In the formula, ω f ,λ,a,b,R,δ are the parameters of the same-direction linearly coupled biological neuron model, and G is the variable scale factor.

[0017] Preferably, the solution model response in step 2 is y(t), and the solution step size is h = G / f s .

[0018] Preferably, the characteristic frequency amplitude ratio measurement index in step 3:

[0019]

[0020] In the formula, B i represents the spectral peak value at the i-th multiple spectral line of the fault characteristic frequency in the power spectrum of the same model response y(t), and A i represents the amplitude at the i-th spectral line in the power spectrum of the model response y(t);

[0021] The residence time distribution index in step 3 is:

[0022]

[0023] In the formula, τ represents the residence time distribution index of Brownian particles in the potential well. By mapping the model response y(t) into a random point process and calculating the time nodes of the zero-crossing rate of the point process, the residence time distribution index τ can be obtained.

[0024] Preferably, the double-objective synchronous optimization function in step 4 is:

[0025]

[0026] In the formula, λ opt , a opt , b opt , R opt , δ opt , G opt are the parameters and variable scale factors of the co-directional linear coupling biological neuron model after double-objective synchronous optimization. Description of the Drawings

[0027] Figure 1 is a flowchart of a method for enhancing the fault symptoms of a co-directional linear coupling biological neuron with double-objective optimization according to the present invention;

[0028] Figure 2 is the time-domain waveform, spectrum and its local enlarged envelope spectrum of the original vibration signal of the rolling bearing;

[0029] Figure 3 is the time-frequency characterization result of the weak fault symptoms of the bearing by the inventive method and the wavelet decomposition method;

[0030] Figure 4 is the extraction result of the weak fault symptoms of the bearing by the fast spectral kurtosis method Detailed Embodiments

[0031] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of specific embodiments of the present invention will be given in conjunction with the accompanying drawings.

[0032] The double-objective optimization co-directional linear coupling biological neuron fault symptom enhancement method, as Figure 1 shown, includes the following steps:

[0033] Step 1: Construct a co-directional linear coupling biological neuron model using a hyperbolic tangent biological neuron and a Gaussian biological neuron, and initialize the parameters of the co-directional linear coupling biological neuron model and the variable scale factor; the constructed co-directional linear coupling biological neuron model is:

[0034]

[0035] In the formula, ω f and λ are the regulation parameters of the hyperbolic tangent biological neuron U1(x)=ω f x 2 / 2 - λln(coshx), and ω f >0, λ>0, regulating ω f and λ to switch the hyperbolic tangent biological neuron between monostable and bistable states to achieve the matching between the mechanical signal to be detected in(t) and the steady-state type; a, b, and R are the regulation parameters of the Gaussian biological neuron U2(x)=ax 2 / 2 + bexp(-x 2 / R 2 ), and a>0, b>0, R>0, and the regulation parameters can achieve the switching between monostable and bistable states; the co-directional linear coupling strength is δ and -1<δ<1; the sampling frequency of the mechanical equipment sensor signal in(t) to be detected is f s , and the number of sampling points is N;

[0036] Initialize the parameters of the co-directional linear coupling biological neuron model and the variable scale factor as:

[0037] ω f ∈(0,10), λ∈(0,10), a∈(0,10), b∈(0,10), R∈(0,10), δ∈(-1,1), G∈(0, f s / 2)

[0038] In the formula, ω f , λ, a, b, R, δ are the parameters of the co-directional linear coupling biological neuron model, and G is the variable scale factor;

[0039] Step 2: Input the initialized parameters of the co-directional linear coupling biological neuron model into the co-directional linear coupling biological neuron model, and solve the model response y(t) and the step size h = G / fs ;

[0040] Step 3: Establish a characteristic frequency amplitude ratio measurement index and a dwell time distribution ratio index based on the model response of the co-directional linear coupled biological neuron model; among them, the characteristic frequency amplitude ratio measurement index:

[0041]

[0042] In the formula, B i represents the spectral peak value at the i-th multiple spectral line of the fault characteristic frequency in the power spectrum of the same model response y(t), and A i represents the amplitude at the i-th spectral line in the power spectrum of the model response y(t);

[0043] The dwell time distribution index in Step 3 is:

[0044]

[0045] In the formula, τ represents the dwell time distribution index of the Brownian particle in the potential well. By mapping the model response y(t) into a random point process and calculating the time nodes of the zero-crossing rate of the point process, the dwell time distribution index τ can be obtained;

[0046] Step 4: Use the characteristic comment amplitude ratio measurement index and the dwell time distribution index in the power spectrum of the model response y(t) as the double-objective synchronous optimization function of the genetic algorithm. Optimize the parameters of the co-directional linear coupled biological neuron model and the variable scale factor within the initial model parameter range in Step 1 to obtain the optimal parameter combination, solve the best model response y′(t), and identify weak fault symptom information; the double-objective synchronous optimization function is:

[0047]

[0048] In the formula, λ opt , a opt , b opt , R opt , δ opt , G opt are the parameters of the co-directional linear coupled biological neuron model and the variable scale factor after double-objective synchronous optimization;

[0049] Solving the best model response y′(t) is:

[0050]

[0051] Then the optimal step size is h′ = G opt / f s .

[0052] In this specific embodiment, a four-bearing coupling fault test bench is used to verify the method of the present invention. The test bench consists of a motor, a belt, and four bearing devices. The sampling frequency is set to 20 kHz, the number of sampling points is 20,480, the test bearing is a double-row cylindrical roller bearing, the radial load is 6000 pounds, the motor speed is 2000 revolutions per minute, and a PCB 353B33 acceleration sensor is installed on the bearing housing to pick up the bearing vibration signal;

[0053] Figure 2 are the time-domain waveform, spectrum, and its local amplified envelope spectrum of the vibration signal for the slight rubbing fault of the bearing outer ring. Among them, Figure 2 No obvious transient impact component can be observed in the time-domain vibration signal of (a), Figure 2 There is an obvious spectral peak at 1000 Hz in the spectrum of (b), but it is not the fault characteristic frequency of the bearing outer ring or the resonance frequency band excited by the local rubbing fault of the bearing outer ring; similarly, Figure 2 In the local amplified envelope spectrum of (c), a weak spectral peak can be seen at the fault characteristic frequency of the bearing outer ring, but it is completely submerged in the strong background noise and difficult to identify, and the spectral peaks at the fault characteristic frequency of the bearing outer ring and its multiples cannot be observed either.

[0054] After being processed by the dual-objective optimization co-directional linear coupling biological neuron fault symptom enhancement method of the present invention Figure 2 the original vibration signal of the bearing in (a), the results are as shown in Figure 3 (a). The optimal parameters of the co-directional linear coupling biological neuron model are respectively λ opt = 1.45, δ opt = -0.61, a opt = 0.59, b opt = 1.34, R opt = 0.72, G opt = 5000. From Figure 3 (a), obvious highlighted spectral lines can be clearly observed at the fault characteristic frequency of the bearing outer ring and its 3 times and 5 times frequency components, indicating that an initial rubbing fault has occurred on the outer ring of the rolling bearing, and the diagnosis result is consistent with the actual fault location.

[0055] For comparison, Figure 3 (b) shows the wavelet time-frequency characterization result. It can be seen that there is an obvious highlighted spectral line at 1000 Hz, but it is not the fault symptom information; in addition, no other obvious spectral lines can be observed. Similarly, when the fast spectral kurtosis method is used to process the original vibration signal of the bearing, the results are as shown in Figure 4 . It can be seen from the local amplified envelope spectrum of the filtered signal that the spectral peak of the outer ring fault characteristic frequency is still submerged in the strong background noise and difficult to clearly distinguish.

[0056] In summary, the dual-objective optimization co-directional linear coupling biological neuron fault symptom enhancement method of the invention can realize the automatic switching between monostability and bistability through co-directional linear coupling hyperbolic tangent biological neurons and Gaussian biological neurons, and consider the measurement indexes of characteristic frequency amplitude ratio and dwell time distribution ratio, and optimize the model parameters and variable scale factors with dual objectives, improving the weak symptom enhancement ability of a single steady state or a single model, overcoming the disadvantage of single-index optimization falling into local optimum, thus effectively using noise to enhance mechanical weak fault symptoms and realizing the early fault diagnosis of mechanical equipment.

[0057] Although the present disclosure is disclosed as above, the protection scope of the present disclosure is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will all fall within the protection scope of the present invention.

Claims

1. A method for enhancing fault symptom of a dual-objective optimization co-directional linear coupled biological neuron, characterized in that It includes the following steps: Step 1: Construct a co-directional linear coupling biological neuron model using hyperbolic tangent biological neurons and Gaussian biological neurons, and initialize the parameters of the co-directional linear coupling biological neuron model and the variable scale factor; the input of the co-directional linear coupling biological neuron model is the mechanical signal to be detected ; the co-directional linear coupling biological neuron model is as follows: ； In the formula, and are the regulation parameters of the hyperbolic tangent biological neuron , and , , regulating and the hyperbolic tangent biological neuron switches between monostability and bistability to achieve the matching between the mechanical signal to be detected and the steady state type; , and are the regulation parameters of the Gaussian biological neuron , and , , the regulation parameters can achieve the switching between monostability and bistability; The in-phase linear coupling strength is and ; the sampling frequency of the mechanical signal to be detected is and the number of sampling points is , and the number of sampling points is ; Step 2: Input the initialized parameters of the co-directional linear coupled biological neuron model into the co-directional linear coupled biological neuron model, and solve the model response and step size; Step 3: Establish the characteristic frequency amplitude ratio measurement index and the dwell time distribution ratio index based on the model response of the co-directional linear coupled biological neuron model; Characteristic frequency amplitude ratio measurement index: ； In the formula, represents the response of the same model the spectral peak value at the multiple spectral line of the fault characteristic frequency in the power spectrum, represents the model response at the amplitude of the root spectral line in the power spectrum; The dwell time distribution index is: ； In the formula, represents the residence time distribution index of Brownian particles in the potential well. By mapping the model response as a random point process, calculate the zero-crossing rate time nodes of the point process, so as to obtain the residence time distribution index ; Step 4: Use a dual-objective synchronous optimization function to synchronously optimize the parameters and variable scale factors of the co-directional linear coupled biological neuron model, obtain the optimal parameter combination, and solve for the best model response , where the best model response is the power spectrum of the co-directional linear coupled biological neuron model output that contains multiple harmonic spectral lines of the fault characteristic frequency. Identify weak fault symptom information from the output best model response .

2. The method for enhancing fault symptom of dual - target optimization co - directional linear coupled biological neurons according to claim 1, characterized in that, In the said Step 1, the initialized parameters of the co-directional linear coupled biological neuron model and the variable scale factor are: ； In the formula, are the parameters of the same-direction linear coupled biological neuron model, is the variable scale factor.

3. The method for enhancing the fault symptom of a dual-objective optimized co-directional linearly coupled biological neuron according to claim 2, wherein The model response solved in Step 2 is , and the solution step size is .

4. The method for enhancing fault symptom of double - target optimized co - directional linearly - coupled biological neurons according to claim 3, wherein The bi-objective synchronization optimization function in the said Step 4 is: ; In the formula, are the parameters of the co-directional linear coupling biological neuron model and the variable scale factor after dual-objective synchronous optimization.

Images