Rotating machinery rubbing diagnosis method based on vibration signal deconstruction and anti-noise enhancement of frequency modulation characteristics

Abstract

The application discloses a rotating machinery rotor rub-impact diagnosis method based on vibration signal deconstruction and frequency modulation characteristic anti-noise enhancement, which comprises the following steps: based on second-order integral transformation and high-pass filtering, vibration acceleration signals collected from rotating machinery equipment are converted into vibration displacement signals; based on an improved variational mode decomposition method, a target extraction is performed on a rotating frequency component; using a normalized Hilbert transform based on orthogonal derivatives, the instantaneous change characteristics of the basic frequency of the rotating frequency component are calculated and estimated to obtain an instantaneous frequency; potential energy parameters and a step length of a Runge-Kutta calculation process are input into an optimal stochastic resonance system optimized by a particle swarm optimization algorithm to perform wave modulation characteristic anti-noise enhancement; after the wave modulation frequency characteristic is subjected to anti-noise enhancement, the instantaneous frequency is subjected to FFT transformation, based on the distribution characteristics of the harmonic amplitudes related to the rotor rotating frequency, the rotating machinery rotor rub-impact fault is diagnosed and identified. The application can accurately diagnose and identify the rotor rub-impact fault under complex noise interference.

Description

Technical Field

[0001] This invention belongs to the field of rotating machinery condition monitoring and fault diagnosis, and in particular relates to a rotating machinery rotor rubbing diagnosis method based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement. Background Technology

[0002] Various rotating machines, such as wind turbines, rotary compressors, centrifugal pumps, and generator sets, play a vital role in numerous modern applications, including mining, metallurgy, and energy development. The rotor is one of the most critical components of these machines. Due to prolonged operation under extremely harsh conditions, rotors may develop various potential faults, such as imbalance, cracks, misalignment, loosening, whirl, and warping. As a common fault exhibiting nonlinear dynamic behavior, rubbing failure occurs when the rotor quasi-periodically impacts the stator of the rotor system. The underlying mechanism is usually a slight initial imbalance and minor misalignment of the rotor, especially when the shaft is rotating at extremely high speeds and the radial clearance between the stator and rotor is very small, making rotor rubbing failure highly likely. In fact, with higher requirements for assembly precision and stricter specifications, the machine structures of most power equipment today are relatively more complex and compact, with components typically operating within extremely small clearances, increasing the likelihood of frictional impacts in the rotor system. Once a rubbing failure occurs, it leads to dynamic instability in the rotor system, which significantly reduces the machine's operating accuracy and adversely affects the structural properties and performance of related power equipment. It also causes additional defects and damage to components such as bearings and impellers, potentially resulting in a system-wide catastrophic failure. Therefore, condition monitoring and fault diagnosis of the rotor system, and timely detection of early rubbing failures in rotating machinery rotors, can effectively improve equipment prognosis and health management, thereby ensuring safe and stable industrial production. This has extremely important practical engineering significance.

[0003] Currently, the methods used for monitoring and diagnosing rotor rubbing faults in rotating machinery typically include the following:

[0004] (1) Based on conventional spectrum analysis techniques, the main approach is to observe the rotational frequency, half-frequency and higher harmonics corresponding to the fault characteristics that may exist in the Fourier spectrum of the collected vibration signal, including synchronous vibration, subsynchronous vibration and supersynchronous vibration. However, these frequency characteristics may also appear when other types of faults occur in the machine, such as rotor shaft cracks and mechanical loosening. Moreover, these characteristics are not significant in the early stage of rotor rubbing faults in rotating machinery and are difficult to detect in time.

[0005] (2) To observe the impact generated by the quasi-periodic contact and separation process between the rotor and stator in the time domain signal, it is usually necessary to use a signal decomposition method to effectively separate and extract the component, mainly including empirical mode decomposition, blind source separation and empirical wavelet transform, etc. However, when other parts of the rotating machinery fail, similar fault symptoms may also occur, such as gear defects and local damage to rolling bearings, and different noise distribution forms and noise levels tend to affect the effectiveness and usability of the signal decomposition results.

[0006] (3) Data-driven intelligent methods mainly include an end-to-end process of “data preprocessing - feature extraction - feature selection / dimensionality reduction - feature learning - diagnostic output”. The mainstream methods involved include support vector machines, random forests, deep belief networks, convolutional neural networks, etc. However, on the one hand, these methods usually require a large number of labeled, identically distributed, high-quality data samples for training and learning, which is difficult to meet under real-world conditions. On the other hand, the non-transparency and uninterpretability of those deep intelligent diagnostic models make them “black boxes”, and their output diagnostic conclusions are difficult to fully trust in practical applications.

[0007] (4) Rotor rubbing fault diagnosis technology for rotating machinery based on nonlinear stochastic dynamic behavior analysis mainly includes Volterra series identification and nonlinear output frequency response equation estimation methods; however, the computational complexity of such methods is usually high, and a certain amount of computational analysis time is required to obtain reliable rotor rubbing fault diagnosis results, which usually makes it difficult to meet the timeliness of online monitoring and diagnosis in practical engineering applications.

[0008] (5) Rotor rubbing diagnosis technology based on amplitude modulation and frequency modulation feature extraction of vibration signal. This type of method often involves time and frequency domain analysis of the signal, mainly including short-time Fourier transform, continuous wavelet transform, Hilbert-Huang transform and adaptive linear frequency modulation mode decomposition, etc. However, this type of method usually has poor robustness under noise interference, and it often fails when analyzing and processing vibration data with low signal-to-noise ratio, which greatly restricts the practical application of these methods.

[0009] It is evident that existing methods for monitoring and diagnosing rotor rubbing faults in rotating machinery all have their own shortcomings and deficiencies. Therefore, there is a need to research efficient rotor rubbing diagnostic methods capable of timely detecting early-stage rotor rubbing faults in complex environments. This would effectively improve equipment prognosis and health management, possessing significant practical engineering implications. Furthermore, since displacement signals typically have a high signal-to-noise ratio and accurately reflect the rotor's vibration state, most of the aforementioned methods are based on vibration displacement signals for rotor system diagnosis and analysis. However, it should be noted that while displacement sensors offer advantages such as a large linear range, zero-frequency response, strong anti-interference capability, and ease of calibration, their practical applications are not as widespread as accelerometers. They are generally only suitable for monitoring rotor systems in large rotating machinery, primarily limited by complex installation requirements, economic considerations, and other factors. Therefore, rotor rubbing diagnostic methods based on vibration displacement signal analysis may be impractical and unfeasible in certain situations. Consequently, it is necessary to research and apply more universal and advanced diagnostic strategies, as rotor rubbing faults can occur in various types of rotating machinery based on rotor systems. Summary of the Invention

[0010] To address the aforementioned problems in the prior art, this invention provides a method for diagnosing rotor rubbing faults in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement. This method can diagnose and identify rotor rubbing faults in rotating power equipment under complex noise interference based on vibration acceleration signal analysis.

[0011] The objective of this invention is achieved through the following technical solution:

[0012] A method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement includes:

[0013] Step 1: Acquire signal sequences from rotating machinery using vibration acceleration sensors;

[0014] Step 2: Based on second-order integral transform and high-pass filtering, the acquired vibration acceleration signal is converted into a vibration displacement signal;

[0015] Step 3: Targeted extraction of the frequency conversion component in the vibration displacement signal based on the improved variational mode decomposition method;

[0016] Step 4: For the extracted frequency components, the instantaneous variation characteristics of their fundamental frequencies are calculated and estimated using the normalized Hilbert transform based on orthogonal derivatives to obtain the instantaneous frequencies;

[0017] Step 5: Input the calculated instantaneous frequency sequence into the optimal stochastic resonance system for noise reduction enhancement of intra-wave modulation features. The potential energy parameters of the stochastic resonance system and the step size of the Runge-Kutta calculation process are optimized by the particle swarm optimization algorithm to obtain the optimal stochastic resonance system.

[0018] Step 6: Perform Fast Fourier Transform on the instantaneous frequency of the intra-wave frequency modulation feature after noise reduction enhancement. Based on the distribution characteristics of the harmonic amplitude related to the rotor frequency, diagnose and identify rotor rubbing faults in rotating machinery.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] This invention addresses the shortcomings and deficiencies of existing rotor rubbing fault diagnosis and analysis technologies for rotating machinery in power equipment. It proposes a method primarily based on vibration acceleration signal analysis and processing. This involves converting the vibration acceleration signal into a vibration displacement signal through a second-order integral transform, and then targeting and extracting the rotational frequency component using a signal deconstruction method. The instantaneous frequency of this component is then calculated and estimated. The resulting instantaneous frequency sequence is then input into an optimal stochastic resonance system for intra-wave modulation feature noise enhancement. The output frequency-enhanced instantaneous frequency is then processed using a Fast Fourier Transform. Finally, the distribution characteristics of harmonic amplitudes related to the rotor rotational frequency are used to diagnose and identify rotor rubbing faults in rotating machinery. This method is based on the more widely used vibration acceleration signal processing technology and exhibits excellent robustness against the unavoidable complex noise interference during signal acquisition and conversion in industrial environments, making it significant for practical engineering applications. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement, according to the present invention.

[0022] Figure 2 This is a time-domain waveform diagram of a vibration acceleration signal sequence collected from a centrifugal pump by a vibration acceleration sensor in an embodiment of the present invention.

[0023] Figure 3 This is a time-domain waveform diagram of the vibration displacement signal converted from the vibration acceleration signal by second-order integral transform and high-pass filtering in an embodiment of the present invention.

[0024] Figure 4 This is a spectrum diagram of the vibration displacement signal converted from the vibration acceleration signal through second-order integral transform and high-pass filtering in an embodiment of the present invention.

[0025] Figure 5 This is a time-frequency diagram of the revolving frequency component in the vibration displacement signal extracted by the improved variational mode decomposition method in an embodiment of the present invention.

[0026] Figure 6 This is a spectrum of the instantaneous frequency components estimated using the normalized Hilbert transform based on orthogonal derivatives in an embodiment of the present invention.

[0027] Figure 7 This is the instantaneous frequency spectrum diagram of the enhanced frequency modulation characteristics of the output of the optimal random resonance system in this embodiment of the invention. Detailed Implementation

[0028] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0029] This invention is based on vibration acceleration signal analysis. First, the vibration acceleration signal is converted into a vibration displacement signal through a second-order integral transform. Then, the frequency component is extracted using a signal deconstruction method. Next, the instantaneous frequency of the frequency component is calculated and estimated. Then, the obtained instantaneous frequency sequence is input into an optimal random resonance system for intra-wave modulation feature noise reduction enhancement. Finally, the instantaneous frequency enhanced by the frequency modulation feature of the optimal random resonance output is processed by a fast Fourier transform. Based on the distribution characteristics of harmonic amplitudes related to rotor frequency, the invention diagnoses and identifies rotor rubbing faults in rotating machinery. This provides important technical support for ensuring timely monitoring and control, and safe and efficient operation and maintenance of rotor systems in power equipment rotating machinery.

[0030] like Figure 1 As shown, a method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement includes the following steps:

[0031] Step 1: Collect signal sequences from rotating machinery using a vibration acceleration sensor.

[0032] In step one, a signal sequence is collected from the rotating machinery using a vibration acceleration sensor, denoted as... N is the total number of sampling points, and the sampling frequency is denoted as f. s Furthermore, during signal acquisition, the frequency response parameter of the vibration acceleration sensor used should be no less than 2kHz, the sampling frequency of the vibration signal sequence should be no less than 5.12kHz and no more than 100kHz, and the sampling duration should be no less than 1s.

[0033] Step Two: Based on second-order integral transform and high-pass filtering, the acquired vibration acceleration signal is converted into a vibration displacement signal. Step Two includes the following sub-steps:

[0034] (2-1) First, perform an integration to convert the vibration acceleration signal into a single signal. Converted into a vibration velocity signal, as follows:

[0035]

[0036]

[0037] Where, ΔT=1 / f s S a It is a signal The signal is obtained after filtering with a high-pass filter; to avoid the adverse effects of extremely low-frequency components in the original vibration signal on the final integration result, the cutoff frequency W of the high-pass filter is set. BG1 Typically, a frequency of 10Hz to 25Hz is selected to obtain better signal integration results.

[0038] (2-2) Next, for the obtained vibration velocity signal Similarly, it is first filtered by a high-pass filter to remove its extremely low-frequency components. The cutoff frequency of the filter is W. BG2 Based on experience, a frequency range of 10Hz to 25Hz is selected to obtain the filtered signal S. v Then for S v The final vibration displacement signal S is obtained by performing integration. x :

[0039] S x (1)=(S v (1)+S v (2)) / 2×ΔT

[0040] S x (i)=S x (i-1)+(S v (i)+S v (i+1)) / 2×ΔT, i=2, 3,...,N

[0041] In step two, a high-pass filter with a stopband attenuation of 60dB is used to perform zero-phase filtering on both the vibration acceleration signal and the vibration velocity signal. The high-pass filter attenuation is lower than the specified passband frequency, which can compensate for the delay introduced by the digital filter.

[0042] Step 3: Targeted extraction of the frequency response component from the vibration displacement signal based on the improved variational mode decomposition method. Step 3 includes the following sub-steps:

[0043] (3-1) First, regarding signal S x A series of intrinsic mode equations u k (t), k∈(1,2,...,K), where K is the total number of modes, is defined as an amplitude-frequency modulated signal, and can be expressed as:

[0044]

[0045] Where A k (t) and These represent the instantaneous amplitude and instantaneous phase, respectively. It is a non-decreasing function, meaning the instantaneous frequency always satisfies and In comparison, A k (t) and ω k The rate of change of (t) is almost extremely slow, therefore these intrinsic mode functions can be regarded as a series of bandwidth-limited signals, which can be expressed by variational method as:

[0046]

[0047] Where δ(·) is the Dirac equation, and * is the convolution operator; BW k is u k (t) is the norm of the bandwidth in the frequency domain. Thus, the constrained variational problem can be reshaped as:

[0048]

[0049] Where, ω k Let K represent the center frequencies of the K eigenmode functions, k∈(1,2,...,K).

[0050] (3-2) The minimization problem is solved by introducing a quadratic penalty term and a Lagrange multiplier term, as follows:

[0051]

[0052] Here, α represents the bandwidth balance parameter, and λ(t) is the Lagrange multiplier factor. Solving this equation using the alternating direction Lagrange multiplier method yields the information about the vibration displacement signal S. x The K eigenmode functions u of (t) k (t), k∈(1, 2,...,K).

[0053] (3-3) Based on the improved variational mode decomposition method, the vibration displacement signal S x The frequency conversion component is extracted from the signal S, that is, in the process of analyzing the signal S... x Before performing variational mode decomposition and eigenmode function extraction, the mode number K is set to 1, the initial bandwidth balancing parameter α is set to 5000, and the initial center frequency of the mode is set to the rotational frequency f. roc The spectral clarity of the signal deconstruction results was evaluated using the target frequency index MTFI, where the optimal parameters of the deconstruction algorithm were optimized through alternating forward and reverse iterations.

[0054] The formula for calculating the Target Frequency Index (MTFI) is as follows:

[0055]

[0056] F * (f roc )=max[F(f roc -0.02f roc f roc +0.02f roc )]

[0057] Where F(f) j ) indicates that the output signal is at frequency f j The amplitude of the Fourier spectrum corresponding to the location; F * (f roc ) indicates at frequency f roc The maximum frequency amplitude within a range of ±2% of the value.

[0058] Ultimately, the intrinsic mode function u corresponding to the maximum MTFI value target This refers to the extracted frequency conversion component.

[0059] Step (3-3) includes the following sub-steps:

[0060] (3-3-1) Input vibration displacement signal S x Initialize the iteration count Iter = 0, set the maximum number of iterations MaxIter = 500, and set the search step size of the bandwidth balancing parameter to Δα = 500;

[0061] (3-3-2) Variational mode decomposition is used to decompose the vibration displacement signal S x Perform a destructuring operation to obtain the intrinsic mode function u0 and its target frequency index MTFI0;

[0062] (3-3-3) Start the loop, let Iter = Iter + 1, and change the bandwidth balancing parameters to α1 = α0 + Δα and α2 = α0 - Δα respectively; next, substitute the two obtained bandwidth balancing parameter values ​​α1 and α2 into the variational mode decomposition algorithm to process the vibration displacement signal S. x Destructuring yields u 1，Iter and u 2，Iter and the corresponding MTFI 1，Iter and MTFI 2，Iter ;

[0063] (3-3-4) If MTFI0 > MTFI 1，Iter And MTFI0 > MTFI 2，Iter The loop terminates, and the final target frequency-dependent component u is output. target =u0; If MTFI0 does not meet the above conditions, then update MTFI0 = max(MTFI1，Iter MTFI 2，Iter )as well as

[0064] (3-3-5) If Iter > MaxIter, then terminate the loop and output the vibration displacement signal S under the action of the latest bandwidth balancing parameter α0 using the variational mode decomposition method. x Deconstruct the structure to obtain the result u. target Conversely, return to step (3-3-3);

[0065] (3-3-6) End the loop and output the final target frequency component u. target .

[0066] In steps (3-3-2), (3-3-3), and (3-3-5), the variational mode decomposition method is used to analyze the vibration displacement signal S. x Deconstruction includes the following sub-steps:

[0067] (1) Let u k The Fourier transforms of f(t), f(t), and λ(t) are respectively and initialization and And the iteration count n = 0;

[0068] (2) Iteration count n = n + 1;

[0069] (3) For k = 1, 2, ..., K, update all... As shown in the following formula:

[0070]

[0071] Simultaneously update ω k :

[0072]

[0073] (4) Update

[0074]

[0075] Where τ is the noise tolerance factor. In order to avoid the influence of strong noise on the signal deconstruction result and because the complete reconstruction of the signal is not the ultimate goal of this invention, the noise tolerance factor is set to τ = 0, which is used to control the influence of noise on the deconstruction process.

[0076] (5) Repeat (2)-(4) until the termination condition is met:

[0077]

[0078] Where ε represents the convergence factor, and we take ε = 1 × 10⁻⁶. -6 .

[0079] Step 4: For the extracted frequency components, the instantaneous variation characteristics of their fundamental frequencies are calculated and estimated using the normalized Hilbert transform based on orthogonal derivatives to obtain the instantaneous frequencies. Step 4 includes the following sub-steps:

[0080] (4-1) For a single-component amplitude-frequency modulation signal g(t), if its amplitude is not normalized, search for all the maximum extreme values ​​in its absolute value form, and then fit these extreme values ​​with a cubic spline function to obtain the empirical envelope function B0(t).

[0081] (4-2) Normalize the amplitude of the signal g(t), that is:

[0082] g1(t) = g(t) / B0(t)

[0083] If the magnitude of g1(t) is still not fully normalized, return to step (4-1) until the Lth empirical envelope function B. L The magnitude of (t) deviates from 1 by less than 10. -5 At this time, the output signal g is obtained. L+1 (t) is considered a pure frequency modulation signal;

[0084] (4-3) Calculate the instantaneous amplitude signal B(t) of signal g(t) as B(t) = B0(t)·B1(t)·…·B L (t), recalculate the frequency-modulated signal F(t) of g(t):

[0085] F(t) = g(t) / B(t)

[0086] (4-4) Let The instantaneous phase φ(t) is then estimated as:

[0087] φ(t)=arctan[Q(t) / F(t)]

[0088] Based on this, the instantaneous frequency ω(t) is calculated and estimated as follows:

[0089]

[0090] And it always satisfies ω(t)>0.

[0091] Step 5: Input the calculated instantaneous frequency sequence into the optimal stochastic resonance system for noise reduction enhancement of intra-wave modulation features. The potential energy parameters of the stochastic resonance system and the step size of the Runge-Kutta calculation process are optimized by the particle swarm optimization algorithm to obtain the optimal stochastic resonance system.

[0092] Step five includes the following sub-steps:

[0093] (5-1) Construct a stochastic resonance system, represented as follows:

[0094]

[0095] Where x represents the trajectory of the particle, S(t) represents a weak periodic signal, N(t) represents Gaussian white noise, and U(x) represents the potential function;

[0096] (5-2) The reflection-symmetric fourth potential function is used as U(x).

[0097]

[0098] Where a and b represent the positive parameters of the control potential function U(x) barrier and potential well, respectively;

[0099] The stochastic resonance system is derived as follows:

[0100]

[0101] (5-3) Solving the stochastic resonance system based on the fourth-order Runge-Kutta method yields the following results.

[0102] k1 = ax(i) - bx(i) 3 +S(i)+N(i)

[0103] k2=a(x(i)+hk1 / 2)-b(x(i)+hk1 / 2) 3 +S(i)+N(i)

[0104] k3=a(x(i)+hk2 / 2)-b(x(i)+hk2 / 2) 3 +S(i+1)+N(i+1)

[0105] k4=a(x(i)+hk3)-b(x(i)+hk3) 3 +S(i+1)+N(i+1)

[0106] x(i+1)=x(i)+(k1+2k2+2k3+k4)h / 6

[0107] Where x(i), S(i) and N(i) represent the discrete forms of signals x(t), S(t) and N(t) respectively, and h represents the step size of the calculation;

[0108] (5-4) The output x(i) of the stochastic resonance system is evaluated using the signal-to-noise ratio (SNR) index.

[0109]

[0110] Where P(·) represents the power spectrum of signal x(i), and Δf = f s / N is the spectral resolution, f m The target frequency component is the rotation frequency, hence f m =f roc The round(·) operator represents the rounding operator.

[0111] (5-5) For a given instantaneous frequency signal ω(t), which encompasses components S(t) and N(t), it is necessary to appropriately adjust the values ​​of a, b, and h to achieve the optimal stochastic resonance effect, as follows:

[0112]

[0113] This invention employs a particle swarm optimization algorithm to search for the optimal (a) in the parameter space. * b * h * The optimal stochastic resonance system under this parameter combination is used to enhance the intra-wave modulation feature noise immunity of the input instantaneous frequency ω(t), and the output is obtained.

[0114] Step (5-5) includes the following sub-steps:

[0115] (5-5-1) Input the estimated instantaneous frequency ω(t) into the stochastic resonance system, let S(t)+N(t)=ω(t), and initialize the particle swarm number N. p =50, set the dimension of the parameter search space to 3, set the upper and lower bounds of the search for parameters (a, b, h) to LB = [0, 0, 0] and HB = [0.05, 5000, 0.05] respectively, set the particle's inertial weight to w = 0.6, set the acceleration constant to γ1 = γ2 = 1.5, and set the absolute value of the particle's maximum velocity to be less than 0.8; initialize the iteration count IterPSO = 0, set the maximum number of iterations MaxIterPSO = 500, and set the precision coefficient controlling the iteration process to ε. P =0.001;

[0116] (5-5-2) Randomly initialize each particle and substitute the parameters of these particles into the stochastic resonance system to calculate the corresponding output; find the current global optimal output. and its SNR0;

[0117] (5-5-3) Begin the loop, setting IterPSO = IterPSO + 1, update and adjust the position and velocity of each particle, and recalculate the corresponding output by substituting the parameters of these particles into the stochastic resonance system; find the current global optimal output. and its SNRIterPSO ;

[0118] (5-5-4) Calculate the signal-to-noise ratio variation ΔSNR=|SNR0-SNR IterPSO | and update SNR0 = max(SNR0, SNR IterPSO )and

[0119] (5-5-5) If IterPSO > MaxIterPSO or ΔSNR < ε P Then terminate the loop and output. Conversely, return to (5-5-3);

[0120] (5-5-6) End the loop and output the final result based on the optimal parameter combination (a) * b * h * The optimal stochastic resonance system under the given conditions performs intra-wave modulation on the input instantaneous frequency ω(t), enhancing its noise immunity output.

[0121] Step 6: Perform Fast Fourier Transform on the instantaneous frequency of the intra-wave frequency modulation feature after noise reduction enhancement. Based on the distribution characteristics of the harmonic amplitude related to the rotor frequency, diagnose and identify rotor rubbing faults in rotating machinery.

[0122] In step six, instantaneous frequency The Fourier spectrum of the rotor frequency f roc The rotor of rotating machinery can be determined to have a rubbing fault only when the distribution characteristics of the relevant harmonic amplitudes meet the following two conditions:

[0123] (1) There exists a clear rotor frequency f that can be clearly distinguished from noise and other mechanical vibration components. roc and its harmonics;

[0124] (2) One-time frequency f roc Its amplitude is the largest.

[0125] The effectiveness of the method of the present invention will be demonstrated below through a specific embodiment.

[0126] S01, a signal sequence is acquired from a centrifugal pump that may have a potential rotor rubbing fault using a vibration acceleration sensor. The time-domain waveform of the vibration acceleration signal is as follows: Figure 2 As shown. This centrifugal pump operates at a speed of 1500 rpm, therefore its rotor frequency f roc This can be calculated as 25Hz; the sampling frequency of the vibration acceleration sensor is 5120Hz, and the sampling duration is 2 seconds. From Figure 2 It is difficult to observe characteristic signs associated with centrifugal pump rotor rubbing faults on the time-domain waveform of the vibration acceleration signal sequence.

[0127] S02, the acquired vibration acceleration signal is converted into a vibration displacement signal based on second-order integral transform and high-pass filtering, wherein the cutoff frequency W of the high-pass filter during the two high-pass filtering processes is... BG1 and W BG2 The frequency was set to 24Hz, and the time-domain waveform of the converted vibration displacement signal is shown below. Figure 3 As shown, its spectrum is as follows Figure 4 As shown. From Figure 4 As can be seen above, besides the relatively significant rotational frequency f... roc In addition to (marked with ○), there are other frequency components.

[0128] S03, based on the improved variational mode decomposition method, the rotational frequency component in the vibration displacement signal is extracted in a targeted manner, and the instantaneous variation characteristics (i.e., frequency modulation characteristics) of its fundamental frequency are calculated and estimated using the normalized Hilbert transform based on orthogonal derivatives, resulting in the time-frequency diagram of the rotational frequency component as shown below. Figure 5 As shown, it can be observed that the vibration signal exhibits intra-wave modulation, indicating a potential fault in the centrifugal pump rotor, but the specific fault type still needs further clarification.

[0129] S04, plot the instantaneous frequency spectrum of the frequency-shifting components estimated using the normalized Hilbert transform based on orthogonal derivatives, such as... Figure 6 As shown. It can be seen that, due to interference from noise and other complex factors, the Fourier spectrum of the instantaneous frequency is related to the rotor rotational frequency f. roc The distribution characteristics of the relevant harmonic amplitudes do not meet the criteria for judging rotor rubbing faults, so it is still difficult to determine whether the centrifugal pump has experienced rotor rubbing faults.

[0130] S05, the calculated instantaneous frequency sequence is input into the optimal stochastic resonance system for intra-wave modulation feature noise reduction enhancement. The potential energy parameters of the stochastic resonance system and the step size of the Runge-Kutta calculation process are optimized using a particle swarm optimization algorithm. The optimal value obtained by the algorithm search is (a * b * h * The parameter combination is (2.9×10) -4 4.4×10 3 (0.0015). The instantaneous frequency at which the frequency modulation characteristic of the output of the optimal stochastic resonance system is enhanced. Spectrum diagram as follows Figure 7 As shown, it possesses a clear rotor frequency f that can be distinctly differentiated from noise and other mechanical vibration components. roc And its harmonics, and one-fold turnaround f roc The amplitude is at its maximum. Therefore, it can be determined that the centrifugal pump experienced a rotor rubbing failure during operation.

[0131] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement, characterized in that, include: Step 1: Acquire signal sequences from rotating machinery using vibration acceleration sensors; Step 2: Based on second-order integral transform and high-pass filtering, the acquired vibration acceleration signal is converted into a vibration displacement signal; Step 3: Targeted extraction of the frequency conversion component in the vibration displacement signal based on the improved variational mode decomposition method; Step 4: For the extracted frequency components, the instantaneous variation characteristics of their fundamental frequencies are calculated and estimated using the normalized Hilbert transform based on orthogonal derivatives to obtain the instantaneous frequencies; Step 5: Input the calculated instantaneous frequency sequence into the optimal stochastic resonance system for noise reduction enhancement of intra-wave modulation features. The potential energy parameters of the stochastic resonance system and the step size of the Runge-Kutta calculation process are optimized by the particle swarm optimization algorithm to obtain the optimal stochastic resonance system. Step 6: Perform Fast Fourier Transform on the instantaneous frequency of the intra-wave frequency modulation feature after noise reduction enhancement. Based on the distribution characteristics of the harmonic amplitude related to the rotor frequency, diagnose and identify rotor rubbing faults in rotating machinery.

2. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, In step one, a signal sequence is collected from the rotating mechanical equipment using a vibration acceleration sensor, denoted as... N is the total number of sampling points, and the sampling frequency is denoted as f. s Furthermore, during signal acquisition, the frequency response parameter of the vibration acceleration sensor used should be no less than 2kHz, the sampling frequency of the vibration signal sequence should be no less than 5.12kHz and no more than 100kHz, and the sampling duration should be no less than 1s.

3. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, Step two includes the following sub-steps: (2-1) First, perform an integration to convert the vibration acceleration signal into a single signal. Converted into a vibration velocity signal, as follows: Where, ΔT=1 / f s S a It is a signal The result is obtained after filtering with a high-pass filter; the cutoff frequency W of this high-pass filter is... BG1 Select 10Hz to 25Hz; (2-2) For the obtained vibration velocity signal Similarly, it is first filtered by a high-pass filter to remove its extremely low-frequency components. The cutoff frequency of the filter is W. BG2 Selecting a frequency range of 10Hz to 25Hz, we obtain the filtered signal S. v Then for S v The final vibration displacement signal S is obtained by performing integration. x : S x (1)=(S v (1)+S v (2)) / 2×ΔT S x (t)=S x (i-1)+(S v (i)+S v (t+1)) / 2×ΔT,i=2,3,...,N In step two, a high-pass filter with a stopband attenuation of 60dB is used to perform zero-phase filtering on both the vibration acceleration signal and the vibration velocity signal. The high-pass filter attenuation is lower than the specified passband frequency, which can compensate for the delay introduced by the digital filter.

4. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, Step three includes the following sub-steps: (3-1) Regarding signal S x A series of intrinsic mode equations u k (t), k∈(1,2,...,K), where K is the total number of modes, defined as a series of bandwidth-limited amplitude-frequency modulated signals, expressed as: Among them, A k (t) and These represent the instantaneous amplitude and instantaneous phase, respectively. It is a non-decreasing function, meaning the instantaneous frequency always satisfies Expressed using the variational method: Where δ(·) is the Dirac equation, and * is the convolution operator; BW k is u k (t) is the norm of the bandwidth in the frequency domain; Thus, the constrained variational problem is reshaped as follows: Where, ω k Let K represent the center frequencies of the K eigenmode functions, k∈(1,2,...,K); (3-2) The constrained variational problem is solved by introducing a quadratic penalty term and a Lagrange multiplier term, as follows: Where α represents the bandwidth balance parameter, and λ(t) is the Lagrange multiplier factor; by solving this equation using the alternating direction Lagrange multiplier method, the vibration displacement signal S is obtained. x The K eigenmode functions u of (t) k (t), k∈(1,2,...,K); (3-3) Based on the improved variational mode decomposition method, the vibration displacement signal S x The frequency conversion component is extracted from the signal S, that is, in the process of analyzing the signal S... x Before performing variational mode decomposition and eigenmode function extraction, the mode number K is set to 1, the initial bandwidth balancing parameter α is set to 5000, and the initial center frequency of the mode is set to the rotational frequency f. roc The spectral clarity of the signal deconstruction results was evaluated using the target frequency index MTFI, where the optimal parameters of the deconstruction algorithm were optimized through alternating forward and reverse iterations. The formula for calculating the Target Frequency Index (MTFI) is as follows: F * (f roc )=max[F(f roc -0.02f roc ,f roc +0.02f r o c )] Where F(f) j ) indicates that the output signal is at frequency f j The amplitude of the Fourier spectrum corresponding to the location; F * (f roc ) indicates at frequency f roc The maximum frequency amplitude within a range of ±2% of the value; Ultimately, the intrinsic mode function u corresponding to the maximum MTFI value target This refers to the extracted frequency conversion component.

5. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 4, characterized in that, Step (3-3) includes the following sub-steps: (3-3-1) Input vibration displacement signal S x Initialize the iteration count Iter = 0, set the maximum number of iterations MaxIter = 500, and set the search step size of the bandwidth balancing parameter to Δα = 500; (3-3-2) Variational mode decomposition is used to decompose the vibration displacement signal S x Perform a destructuring operation to obtain the intrinsic mode function u0 and its target frequency index MTFI0; (3-3-3) Start the loop, let Iter = Iter + 1, and change the bandwidth balancing parameters to α1 = α0 + Δα and α2 = α0 - Δα respectively; next, substitute the two obtained bandwidth balancing parameter values ​​α1 and α2 into the variational mode decomposition algorithm to process the vibration displacement signal S. x Destructuring yields u 1,Iter and u 2,Iter and the corresponding MTEI 1,Iter and MTEI 2,Iter ; (3-3-4) If MTEI0 > MTEI 1,Iter And MTEI0>MTEI 2,Iter The loop terminates, and the final target frequency-dependent component u is output. target =u0; If MTFI0 does not meet the above conditions, then update MTFI0 = max(MTFI 1,Iter MTFI 2,Iter )as well as (3-3-5) If Iter > MaxIter, then terminate the loop and output the vibration displacement signal S under the action of the latest bandwidth balancing parameter α0 using the variational mode decomposition method. x Deconstruct the structure to obtain the result u. target Conversely, return to step (3-3-3); (3-3-6) End the loop and output the final target frequency component u. target .

6. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 5, characterized in that, In steps (3-3-2), (3-3-3), and (3-3-5), the variational mode decomposition method is used to analyze the vibration displacement signal S. x Deconstruction includes the following sub-steps: (1) Let u k The Fourier transforms of f(t), f(t), and λ(t) are respectively and initialization and And the iteration count n = 0; (2) Iteration count n = n + 1; (3) For k = 1, 2, ..., K, update all... As shown in the following formula: Simultaneously update ω k : (4) Update Where τ is the noise tolerance factor, τ=0, which is used to control the influence of noise on the deconstruction process; (5) Repeat (2)-(4) until the termination condition is met: Where ε represents the convergence factor, and we take ε = 1 × 10⁻⁶. -6 .

7. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, Step four includes the following sub-steps: (4-1) For a single-component amplitude-frequency modulation signal g(t), if its amplitude is not normalized, search for all the maximum extreme values ​​in its absolute value form, and then fit these extreme values ​​with a cubic spline function to obtain the empirical envelope function B0(t). (4-2) Normalize the amplitude of the signal g(t), that is: g1(t) = g(t) / B0(t) If the magnitude of g1(t) is still not fully normalized, return to step (4-1) until the Lth empirical envelope function B. L The magnitude of (t) deviates from 1 by less than 10. -5 At this time, the output signal g is obtained. L+1 (t) is considered a pure frequency modulation signal; (4-3) Calculate the instantaneous amplitude signal B(t) of signal g(t) as B(t) = B0(t)·B1(t)·…·B L (t), recalculate the frequency-modulated signal F(t) of g(t): F(t) = g(t) / B(t) (4-4) Let The instantaneous phase φ(t) is then estimated as: φ(t)=arctan[Q(t) / F(t)] Based on this, the instantaneous frequency ω(t) is calculated and estimated as follows: And it always satisfies ω(t)>0.

8. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, Step five includes the following sub-steps: (5-1) Construct a stochastic resonance system, represented as follows: Where x represents the trajectory of the particle, A(t) represents a weak periodic signal, N(t) represents Gaussian white noise, and U(x) represents the potential function; (5-2) The reflection-symmetric fourth potential function is used as U(x). Where a and b represent the positive parameters of the control potential function U(x) barrier and potential well, respectively; The stochastic resonance system is derived as follows: (5-3) Solving the stochastic resonance system based on the fourth-order Runge-Kutta method yields the following results. k1=ax(i)-bx(i) 3 +S(i)+N(i) k2=a(x(i)+hk1 / 2)-b(x(i)+hk1 / 2) 3 +S(i)+N(i) k3=a(x(i)+hk2 / 2)-b(x(i)+hk2 / 2) 3 +S(i+1)+N(i+1) k4=a(x(i)+hk3)-b(x(i)+hk3) 3 +S(i+1)+N(i+1) x(i+1)=x(i)+(k1+2k2+2k3+k4)h / 6 Where x(i), S(i) and N(i) represent the discrete forms of signals x(t), S(t) and N(t) respectively, and h represents the step size of the calculation; (5-4) The output x(i) of the stochastic resonance system is evaluated using the signal-to-noise ratio (SNR) index. Where P(·) represents the power spectrum of signal x(i), and Δf = f s / N is the spectral resolution, f m The target frequency component is the rotation frequency, hence f m =f roc The round(·) operator represents the rounding operator. (5-5) The particle swarm optimization algorithm is used to search for the optimal (a) in the parameter space. * b * h * The optimal stochastic resonance system under this parameter combination is used to enhance the intra-wave modulation feature noise immunity of the input instantaneous frequency ω(t), and the output is obtained.

9. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 8, characterized in that, Step (5-5) includes the following sub-steps: (5-5-1) Input the estimated instantaneous frequency ω(t) into the stochastic resonance system, let S(t)+N(t)=ω(t), and initialize the particle swarm number N. p =50, set the dimension of the parameter search space to 3, set the upper and lower bounds of the search for parameters (a, b, h) to LB = [0, 0, 0] and HB = [0.05, 5000, 0.05] respectively, set the particle's inertial weight to w = 0.6, set the acceleration constant to γ1 = γ2 = 1.5, and set the absolute value of the particle's maximum velocity to be less than 0.8; initialize the iteration count IterPSO = 0, set the maximum number of iterations MaxIterPSO = 500, and set the precision coefficient controlling the iteration process to ε. P =0.001; (5-5-2) Randomly initialize each particle and substitute the parameters of these particles into the stochastic resonance system to calculate the corresponding output; find the current global optimal output. and its SNR0; (5-5-3) Begin the loop, setting IterPSO = IterPSO + 1, update and adjust the position and velocity of each particle, and recalculate the corresponding output by substituting the parameters of these particles into the stochastic resonance system; find the current global optimal output. and its SNR IterPSO ; (5-5-4) Calculate the signal-to-noise ratio variation ΔSNR=|SNR0-SNR IterPSo | and update SNR0 = max(SNR0, SNR ItrePSO )and (5-5-5) If IterPSO>MaxIterPSO or ΔSNR<ε P Then terminate the loop and output. Conversely, return to (5-5-3); (5-5-6) End the loop and output the final result based on the optimal parameter combination (a) * ,b * ,h * The optimal stochastic resonance system under the given conditions performs intra-wave modulation on the input instantaneous frequency ω(t), enhancing its noise immunity output.

10. The method for diagnosing rotor rubbing in rotating machinery based on vibration signal deconstruction and frequency modulation feature noise reduction enhancement according to claim 1, characterized in that, In step six, instantaneous frequency The Fourier spectrum of the rotor frequency f roc The rotor of rotating machinery can be determined to have a rubbing fault only when the distribution characteristics of the relevant harmonic amplitudes meet the following two conditions: (1) There exists a clear rotor frequency f that can be clearly distinguished from noise and other mechanical vibration components. roc and its harmonics; (2) One-time frequency f roc Its amplitude is the largest.

Images