De-noising method for early failure detection signal of broken shaft of elevator

Abstract

The invention relates to the cross technical field of elevator fault diagnosis and signal processing, in particular to an elevator shaft breakage early fault detection signal denoising method. The method comprises the following steps: acquiring an original low and medium frequency discrete signal of elevator shaft body vibration, and filtering by adopting a four-order Butterworth band-pass filter to obtain a filtered signal; performing complementary set empirical mode decomposition on the filtering signal to obtain a plurality of IMF components and a residual component, and screening effective IMF components from all the IMF components; the center frequency of each effective IMF component is calculated, and Gaussian weight distribution is carried out on each effective IMF component based on the center frequency of each effective IMF component; performing signal reconstruction on all the effective IMF components and the residual components according to the weight of each effective IMF component to obtain a reconstructed signal; and smoothing residual noise in the reconstructed signal by adopting self-adaptive moving smoothing filtering to obtain a de-noised signal. According to the method, the 30-60Hz low and medium frequency characteristic signals generated in the elevator spindle microcrack propagation stage can be accurately extracted, so that fault judgment is facilitated.

Description

Technical Field

[0001] This invention relates to the field of elevator fault diagnosis and signal processing, specifically to a method for noise reduction of early fault detection signals in elevator shaft breakage. Background Technology

[0002] The structural integrity of the elevator spindle is crucial for ensuring safe elevator operation. Spindle breakage is not a sudden occurrence but rather exhibits a progressive development: in the early microcrack propagation stage (crack depth typically less than 0.1mm), the fatigue of the shaft metal gradually intensifies with operation. This fatigue state, transmitted through the mechanical structure, excites mid-to-low frequency characteristic signals of 30-60Hz. As the fault progresses, the amplitude of these characteristic signals gradually increases until they ultimately lead to a sudden spindle fracture. However, the actual elevator operating environment is subject to multiple noise interferences, severely affecting the identification of characteristic signals—the 300-500Hz electromagnetic noise generated by the motor and the 200-300Hz friction noise generated by the friction between the car and guide rails. These two types of high-frequency noise easily mask the 30-60Hz target signal. Simultaneously, the 20-25Hz low-frequency interference generated by the vibration of the building structure itself further increases the difficulty of extracting early fault signals.

[0003] In existing noise reduction algorithms, wavelet transform has poor adaptability to non-stationary signals in the 30-60Hz range, which can easily lead to energy loss. Traditional EMD decomposition suffers from mode aliasing, and the signal-to-noise ratio improvement is less than 5dB. Moreover, existing technologies are mostly designed for mid-to-late stage fault signals above 100Hz, and cannot achieve early warning of microcracks.

[0004] Therefore, there is an urgent need to design a noise reduction method suitable for early fault detection signals of elevator shaft breakage. Summary of the Invention

[0005] This invention aims to solve three core technical problems in the existing early fault detection of elevator shaft breakage: First, in the elevator operating environment, the amplitude of the microcrack characteristic signal in the 30-60Hz range is extremely low (usually not exceeding 0.1V), easily masked by low-frequency interference from building vibrations in the 20-25Hz range, electromagnetic noise from motors in the 300-500Hz range, and guide rail friction noise in the 200-300Hz range, making the characteristic signal difficult to identify; second, existing wavelet transform denoising technology has poor adaptability to non-stationary signals in the 30-60Hz range, with a characteristic signal energy loss rate exceeding 15%, failing to retain key details of the microcrack propagation stage; third, traditional EMD decomposition technology suffers from mode aliasing problems, with a signal-to-noise ratio improvement of less than 5dB, resulting in a high false alarm rate for early faults, and existing technologies cannot achieve effective early warning for microcracks with a depth of less than 0.1mm. Therefore, this invention proposes a denoising method for early fault detection signals in elevator shaft breakage, achieving accurate extraction of the 30-60Hz target signal while efficiently suppressing multi-band noise, facilitating fault determination, and meeting the practical needs of early warning for elevator shaft breakage.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is: a method for noise reduction of early fault detection signals in elevator shaft breakage, comprising: Step S1: Obtain the original low-to-medium frequency discrete signal of elevator shaft vibration. The filtered signal was obtained by using a fourth-order Butterworth bandpass filter. ; Step S2, filter the signal Perform complementary set empirical mode decomposition to obtain several IMF components and a residual component, and then select the effective IMF components from all IMF components; Step S3: Calculate the center frequency of each effective IMF component. Based on the center frequency of each valid IMF component Apply Gaussian weights to it; Step S4: Based on the weight of each effective IMF component, reconstruct the signal from all effective IMF components and residual components to obtain the reconstructed signal. ; Step S5: Adaptive moving smoothing filter is used to smoothly reconstruct the signal. The residual noise in the signal is used to obtain the denoised signal. .

[0007] Furthermore, in step S1, a fourth-order Butterworth bandpass filter is used to filter the signal to obtain the filtered signal. Specifically, including, The filter's transfer function H(f) is converted into a discrete unit impulse response h(n) using the impulse response invariance method, and then compared with the original low-to-medium frequency discrete signal. Convolution yields the filtered signal. .

[0008] Furthermore, in step S2, valid IMF components are screened from all IMF components. The strategy includes: Filter IMF components with a main frequency in the range of 30-80Hz.

[0009] Furthermore, in step S2, the strategy for filtering valid IMF components from all IMF components also includes: Both condition one and condition two must be satisfied; among them... Condition 1: |PZ|≤1; where P and Z are the number of poles and zero crossings of the same IMF component, respectively. Condition 2: The mean of the upper and lower envelopes of the cubic spline interpolation fitting is ≤ 1 / 20 of the maximum amplitude of the IMF component.

[0010] Furthermore, in step S3, based on the center frequency of each valid IMF component... Assign Gaussian weights to them; including, Calculate the energy integral value for each frequency band and calculate the normalization coefficient. :

[0011] In the formula, These are the energy integral values ​​for the target frequency band (30-60Hz), the transition frequency band (60-70Hz), the high-frequency noise frequency band (70-80Hz), and the low-frequency interference frequency band (20-30Hz), respectively. Center frequency based on effective IMF components Design a piecewise Gaussian weighting function: 20Hz≤f i <30Hz:

[0012] 30Hz≤f i <60Hz:

[0013] 60Hz≤f i ≤70Hz:

[0014] 70Hz <f i <80Hz:

[0015] For each valid IMF component, its center frequency is... Substituting the corresponding Gaussian weight function, we obtain the weights. .

[0016] Furthermore, in step S3, after assigning Gaussian weights to each valid IMF component, a weight anomaly correction step is also included; the weight anomaly correction step includes, Wavelet thresholding is used to extract the center frequency. The noise component n(n) for each effective IMF component within the 30-60Hz range; Calculate the variance of the effective IMF component and its corresponding noise component n(n) respectively, and calculate the signal-to-noise ratio based on the variance of the two. For effective IMF components with a signal-to-noise ratio less than a preset threshold, force the weight to be 0.

[0017] Furthermore, in step S4, during the signal reconstruction process of all valid IMF components and residual components, the weight of the residual components is 0.8.

[0018] Furthermore, step S5 also includes: For noise reduction signal Simultaneous time-domain and frequency-domain verification is performed; among them... The criteria for passing time-domain verification are: The amplitude difference between adjacent sampling points in the time domain is ΔA = |x denoise (n+1)-x denoise (n)|≤0.01V; The criteria for passing frequency domain verification are: The amplitude fluctuation in the target frequency band of 30-60Hz is ≤3%, and the amplitude in non-target frequency bands other than the target frequency band of 30-60Hz is ≤0.03V.

[0019] Furthermore, the method also includes: Step S6, determine the noise reduction signal Does it support early fault detection of elevator shaft breakage? The method for determining this is as follows: If the following conditions are met simultaneously: energy loss rate <5% in the 30-60Hz band, signal-to-noise ratio improvement ≥10dB, power spectral density (PSD)(f) shows a significant peak at 30-60Hz, PSD ≤ 1 / 10 of the peak value in the 70-80Hz band, and PSD ≤ 1 / 5 of the peak value in the 20-30Hz band, then it is supported.

[0020] By adopting the above technical solution, the present invention has the following beneficial effects: The method in this invention significantly improves the extraction capability of fault feature signals in the 30-60Hz frequency range, with an energy loss rate of less than 5%, a signal-to-noise ratio improvement of over 10dB, and a significant effect on suppressing noise in non-target frequency bands. Compared with traditional methods, the method in this invention has significant advantages in fault feature preservation and noise suppression, enabling early warning of elevator shaft breakage 1-2 months in advance, and is highly practical. Attached Figure Description

[0021] Figure 1 This is a flowchart of the elevator shaft breakage early fault detection signal noise reduction method of the present invention; Figure 2 The time-domain waveform of the original low-to-medium frequency discrete signal x(n) of the elevator shaft vibration; Figure 3 For filtered signals picture; Figure 4 A diagram of the six IMF components; Figure 5 The graph shows the piecewise Gaussian weighting function curve based on the center frequency. Figure 6 A comparison of the frequency-sensing sensitivity characteristics between the low-frequency perception model for elephants and the piecewise Gaussian weighting function in this invention. Figure 7 For reconstructing signals ; Figure 8 For noise reduction signal picture; Figure 9 This is a comparison of the power spectral density of the original signal and the denoised signal in the frequency domain. Detailed Implementation

[0022] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0023] like Figure 1 As shown, a noise reduction method for early fault detection signals of elevator shaft breakage includes: Step S1, firstly, define the original low-to-mid frequency discrete signal of the elevator shaft vibration as: The value of n ranges from 0 to N-1 (N=10000 points, which is determined by a sampling frequency of 1000Hz and a single sampling duration of 10s), and the signal unit is V.

[0024] Then, to initially eliminate low-frequency interference below 30Hz and high-frequency noise above 80Hz, a fourth-order Butterworth bandpass filter was selected. The fourth order was chosen because it better balances the passband flatness and stopband attenuation rate, ensuring a stopband attenuation of over 40dB per decade, meeting noise suppression requirements. The low-frequency cutoff frequency of this filter is... Set to 30Hz, high-frequency cutoff frequency The frequency is set to 80Hz, and its transfer function expression is:

[0025] In the formula, f is the signal frequency (unit: Hz), and M is the filter order (unitless, here the value is 4). This is the low-frequency cutoff frequency (unit: Hz, value: 30). This is the high-frequency cutoff frequency (unit: Hz, value: 80).

[0026] In practical filtering operations, the continuous domain transfer function needs to be converted using the impulse response invariance method. Converting to a discrete unit impulse response h(n) (unit: V / V, dimensionless, reflecting the gain relationship between the input and output voltage signals), the specific process is as follows: First, ... A Fourier inverse transform is performed to obtain the continuous impulse response h(t) (unit: s). Then, h(t) is sampled at a sampling period of 0.001 s (determined by a sampling frequency of 1000 Hz) to obtain the discrete form h(n). The filtering process is implemented through discrete convolution, i.e., the filtered signal... (Unit: V) is equal to the convolution of the original signal x(n) and the discrete unit impulse response h(n), and its calculation formula is:

[0027] After this filtering process Only the 30-80Hz frequency band was retained to lay the foundation for subsequent decomposition.

[0028] To ensure filtering effectiveness, a Discrete Fourier Transform (DFT) must be used. The verification criteria for performing spectrum analysis are: the signal amplitude below 30Hz should not exceed 1 / 10 of the original signal amplitude at 30Hz, and the signal amplitude above 80Hz should not exceed 1 / 15 of the original signal amplitude at 80Hz. Only when this condition is met can it be confirmed that out-of-band noise has been effectively filtered out.

[0029] Step S2, the signal after bandpass filtering The complementary ensemble empirical mode decomposition (CEEMD) technique is used to decompose the signal into several intrinsic mode functions (IMFs) and residual components, thereby achieving separation of different frequency components. The first step of the decomposition is to construct a noisy signal, specifically by generating two sets of Gaussian white noise sequences, both with amplitudes of [missing value]. The standard deviation is 0.2 times, one group is the original noise and the other is the inverted noise, and the length of both noise groups is the same as the standard deviation. Keeping consistent (unit: V), compare these two sets of noise with... By superimposing the signals, two noisy signals are obtained. and .

[0030] Next to and EMD decomposition is performed separately, following the classic EMD screening process: First, the local maxima and minima of each noisy signal are identified, and the upper and lower envelopes of the signal are fitted using cubic spline interpolation. Then, the mean of the upper and lower envelopes is calculated, and this mean is subtracted from the original noisy signal to obtain the intermediate signal. The process of "fitting the envelope - calculating the mean - subtracting the mean" is repeated until the intermediate signal meets the criteria for IMF components, i.e., the number of local maxima and minima is equal or differs by 1, and the mean of the envelope approaches 0. The intermediate signal obtained at this point is an IMF component. The screening continues until no new IMF components can be obtained. Finally, each noisy signal will be decomposed into multiple IMF components and a residual component r(n) (unit: V).

[0031] To ensure decomposition stability and reduce noise interference, the number of ensemble iterations for CEEMD was set to 100, and the standard deviation of the IMF components (SD) ≤ 0.3 was used as the decomposition termination condition. The formula for calculating SD is as follows:

[0032] In the formula, IMF i (n) represents the IMF component (unit: V) obtained from the i-th screening. i-1 (n) represents the IMF component obtained from the (i-1)th screening (unit: V), and SD is the standard deviation (unitless).

[0033] After decomposition, the validity of the obtained IMF components needs to be determined, and invalid components need to be removed to ensure the accuracy of subsequent processing. The determination criteria include three aspects: First, the number of poles P and the number of zero crossings Z of each IMF component are counted, requiring |PZ|≤1; Second, the upper and lower envelopes of the IMF components are fitted using cubic spline interpolation, and the mean of the envelopes is calculated, ensuring that the mean does not exceed 1 / 20 of the maximum amplitude of the IMF component; Third, the spectrum of the IMF components is analyzed by DFT, and components with a dominant frequency below 30Hz or above 80Hz are removed, retaining only the valid IMF components (denoted as IMF) with a dominant frequency in the range of 30-80Hz. h (n), h=1,2,…,H, where H is the number of effective IMF components).

[0034] Step S3: First, calculate the center frequency for the retained valid IMF components. The calculation process requires first analyzing the IMF components. h (n) Perform a Hilbert transform to construct an analytic signal z. h (n), whose expression is:

[0035] In the formula, j is the imaginary unit, HT[IMF h[n] is the IMF h The Hilbert transform result (in V) of (n) is calculated using numerical integration, and the formula is as follows:

[0036] Extracting analytical signals phase (Unit: rad) The instantaneous frequency f is calculated based on the phase. i (n) (unit: Hz), the calculation formula is:

[0037] In the formula, The sampling period is 0.001 seconds (unit: s). Finally, the center frequency of each effective IMF component is calculated using discrete integral form. The calculation error is controlled to not exceed 0.5Hz. The calculation formula is:

[0038] After the calculation is completed, all effective IMF components are sorted by center frequency. Sort by size from smallest to largest to prepare for subsequent weight allocation.

[0039] Then, the high sensitivity of elephants to specific low-frequency signals was simulated, based on the center frequency of the effective IMF component. Design a piecewise Gaussian weighting function to enhance the target frequency band signal and suppress interfering frequency band signals. Specifically: The energy integral value for each frequency band needs to be calculated. The specific operation is to... A Hanning window was used for windowing (the window length of 1024 was chosen because it can cover at least 30 cycles of the 30-60Hz target signal, effectively avoiding spectral leakage, while a 50% overlap rate was set to balance computational efficiency and spectral resolution). The power spectral density of the signal was calculated using the periodogram method (i.e., performing a DFT on the windowed signal, taking the square of the modulus, and dividing by the window function energy). Then, the power spectral density was integrated across different frequency bands to obtain the energy integral value for each band. Figure 3 Taking the signal as an example, the energy integral value W1 of the target frequency band of 30-60Hz is about 8.5V² / Hz, the energy integral value W2 of the transition frequency band of 60-70Hz is about 3.2V² / Hz, the energy integral value W3 of the high-frequency noise band of 70-80Hz is about 0.8V² / Hz, and the energy integral value W4 of the low-frequency interference band of 20-30Hz is about 1.2V² / Hz.

[0040] The normalization coefficient λ is calculated based on the above energy integral value, and the formula is as follows:

[0041] After substituting the values, we get Approximately 0.073 (unitless). Based on the normalization coefficient. Weights are assigned to IMF components in different center frequency ranges, and the specific segmented weighting formula is as follows: For center frequencies ≤ 20Hz The effective IMF component weighting formula for the low-frequency interference band (<30Hz) is as follows:

[0042] For center frequencies ≤ 30Hz For the target frequency band <60Hz, the effective IMF component weighting formula is:

[0043] For center frequencies ≤ 60Hz For the transition frequency band ≤70Hz, the effective IMF component weighting formula is as follows:

[0044] For center frequencies < 70Hz For the high-frequency noise band below 80Hz, the effective IMF component weighting formula is as follows: .

[0045] To avoid noise amplification caused by some low signal-to-noise ratio effective IMF components, anomaly correction of the weights is also required. The correction process uses the sym8 wavelet basis to correct the IMF in the 30-60Hz target frequency band. h (n) Wavelet threshold denoising is performed (the sym8 wavelet basis is chosen because it is highly adaptable to the decomposition of mechanical vibration signals and has low computational complexity; this step only corrects the weight anomaly of IMF components with center frequencies in the target frequency band of 30-60Hz because these components are given high weights and are the main contributors to signal reconstruction. To avoid excessive amplification of low signal-to-noise ratio components during reconstruction, thus introducing noise, a strict quality review is required. For non-target frequency band components, since their initial weights have been significantly attenuated, their impact on the final result is negligible, so this correction is not performed).

[0046] The specific correction process is as follows: IMF through multi-scale decomposition h(n) is decomposed into approximation coefficients and detail coefficients. Soft thresholding is applied to the detail coefficients (the threshold is set to 0.02V, determined based on three times the noise standard deviation, effectively filtering noise while preserving signal details). Detail coefficients with absolute values ​​less than the threshold are set to zero, while those with absolute values ​​greater than or equal to the threshold are subtracted from the threshold. The denoised IMF component, IMF', is then obtained through wavelet reconstruction. h (n). Extracting noise components n h (n) (unit: V), i.e., n h (n)=IMF h (n)-IMF' h (n), calculate IMF respectively h (n) and n h The variance of (n), in terms of signal For example, the formula for calculating variance is:

[0047] In the formula, For signal The mean (unit: V). Variance (unit: V²). The signal-to-noise ratio (SNR) for each IMF component is calculated based on the variance. h The formula is:

[0048] If the signal-to-noise ratio <5dB, forced weight = 0.

[0049] Step S4: Considering that the residual component r(n) obtained from CEEMD decomposition may carry low-frequency noise, it is necessary to assign a separate weight to it to balance signal trend preservation and noise suppression. In this embodiment, the weight ω of the residual component is set. r =0.8 (This value was chosen because a weight of 0.8 can retain 80% of the main trend signal in r(n) while reducing low-frequency noise interference to less than 20% of the original amplitude, thus avoiding affecting the target signal). Combined with the weights of each effective IMF component obtained in step S3. The signal is reconstructed according to the discrete reconstruction formula, which is as follows:

[0050] In the formula, The reconstructed signal (unit: V), H is the number of effective IMF components (unitless), ω r denoted as the weight of the residual component (unitless), and r(n) is the residual component obtained from CEEMD decomposition (unit: V).

[0051] After reconstruction, it needs to be performed using DFT. Spectrum analysis was performed to verify the reconstruction effect. The verification criteria were: the signal amplitude in the target frequency band of 30-60Hz should not be less than 80% of the original signal amplitude in that frequency band, and the signal amplitude in non-target frequency bands (specifically all frequency bands below 30Hz and above 60Hz, with particular attention to building vibration interference in 20-25Hz and high-frequency noise above 300Hz) should not exceed 0.05V, to ensure that the target signal characteristics are enhanced and the interference signal is effectively suppressed.

[0052] Step S5, to further smooth the reconstructed signal To eliminate residual noise while preserving the impact characteristics generated by metal fatigue, this embodiment employs an adaptive moving smoothing filter technique. This technique divides the signal into local windows centered on each sampling point n, with a window length set to L = 2k(n) + 1 (the sliding step is 1, and k(n) is the half-length of the window, dimensionless). First, the local mean of the signal within the window is calculated. (Unit: V) and local standard deviation (Unit: V), the calculation formulas are as follows:

[0053]

[0054] Based on local mean with standard deviation Calculate the local kurtosis K(n) (unitless). Kurtosis is a key indicator reflecting the impulse characteristics of a signal, and its calculation formula is as follows:

[0055] In the formula, The summation index variable represents the position offset within the window; The window half-length k(n) is dynamically adjusted based on the kurtosis value K(n) to achieve an adaptive balance between smoothing effect and impact feature preservation: when K(n) > 3, it indicates that the signal within the window contains impact features caused by metal fatigue. In this case, k(n) = 2 (corresponding to a window length of 5 points and a time span of 5ms), and the short window length can effectively avoid excessive smoothing of impact details; when K(n) ≤ 3, it indicates that the signal within the window is relatively smooth with no obvious impact features. In this case, k(n) = 5 (corresponding to a window length of 11 points and a time span of 11ms), and the long window length can enhance the smoothing effect on residual noise. Based on the adjusted window half-length k(n), the smoothed denoised signal is calculated. (Unit: V), the calculation formula is:

[0056] After smoothing, the effect needs to be verified in both the time and frequency domains: In the time domain, calculate the amplitude difference ΔA between adjacent sampling points (ΔA=|x denoise(n+1)-x denoise (n)|), requiring ΔA≤0.01V to ensure a smooth signal without abrupt noise; in the frequency domain, x is analyzed through DFT. denoise The spectrum of (n) requires that the amplitude fluctuation of the target frequency band of 30-60Hz not exceed 3%, and the amplitude of the non-target frequency band (other frequency ranges other than the target frequency band of 30-60Hz) not exceed 0.03V, so as to ensure that the target signal is stable and the noise suppression meets the standard.

[0057] Step S6, the final output noise-reduced signal Three core indicators must be met to ensure its effective support for early detection of elevator shaft breakage: The first indicator is energy loss rate, calculated by integrating the squares of the signal in the 30-60Hz frequency band. The energy loss rate (calculated as (original signal 30-60Hz energy - noise-reduced signal 30-60Hz energy) / original signal 30-60Hz energy × 100%) must be less than 5% to ensure that the energy of the target signal is not excessively lost. The second indicator is signal-to-noise ratio (SNR) improvement, based on the SNR calculation formula in step S3, requiring an SNR improvement of no less than 10dB after noise reduction to ensure that noise is effectively suppressed. The third indicator is power spectral density (PSD). Perform DFT to obtain X denoise (f), according to formula

[0058] Calculate the PSD (unit: V² / Hz, where T=10s is the sampling duration). It is required that there is a significant peak in the 30-60Hz frequency band, and the PSD in the 70-80Hz high frequency band should not exceed 1 / 10 of the peak value in the 30-60Hz frequency band, and the PSD in the 20-30Hz low frequency band should not exceed 1 / 5 of the peak value in the 30-60Hz frequency band, so as to ensure the identification of target feature signals and the effect of interference suppression.

[0059] The solutions involved in the above embodiments will be described in detail below with reference to specific examples.

[0060] like Figure 1 As shown, a noise reduction method for early fault detection signals of elevator shaft breakage includes: 1. Experimental parameters are as follows: The experiment uses an elevator spindle system, with steel bearings. The traction sheave rotation frequencies are 30Hz and 60Hz. Human interventions include lead breakage, impact, and friction. Infrasound sensors are used to collect vibration signals at a sampling frequency of 1000Hz and a sampling duration of 10s. The time-domain waveform of one of the original low-to-medium frequency discrete signals x(n) of the elevator shaft vibration is shown below. Figure 2As shown, the original signal contains multiple frequency components with a large amplitude fluctuation range; there is obvious high-frequency noise and low-frequency interference, and the target signal (30-60Hz) is submerged; this reflects the authenticity and complexity of the signal in the actual elevator operating environment.

[0061] 2. Figure 3 The original signal shown is obtained after being filtered by a 4th-order Butterworth bandpass filter. Its spectrum is as follows Figure 3 As shown, noise below 30Hz and above 80Hz is effectively suppressed, and useful components in the 30-80Hz frequency band are preserved, providing a preprocessing basis for subsequent CEEMD decomposition; the waveform is relatively smooth, but non-stationary characteristics still exist.

[0062] 3. Execute step S2; CEEMD decomposes into 6 IMF components, such as... Figure 4 As shown, IMF1 has the highest frequency and IMF6 has the lowest frequency, reflecting the decomposition order from high frequency to low frequency; the waveforms of each IMF component exhibit typical oscillation characteristics, which conforms to the IMF definition. Effective IMF components (IMF2, IMF3, IMF4) with a main frequency of 30-80Hz were selected.

[0063] 4. Execute step S3 to calculate the energy integral value for each frequency band (W1=8.52V² / Hz, W2=3.18V² / Hz, W3=0.79V² / Hz, W4=1.21V² / Hz), with a normalization coefficient λ=0.073. After weighting, remove the SNR. h Effective IMF component <5dB.

[0064] In this embodiment, the weight calculated for the IMF component with a center frequency of 25Hz is approximately 0.022, the weight calculated for the IMF component with a center frequency of 30Hz is approximately 0.072, the weight calculated for the IMF component with a center frequency of 45Hz is approximately 0.066, the weight calculated for the IMF component with a center frequency of 60Hz is approximately 0.072, and the weight calculated for the IMF component with a center frequency of 65Hz is approximately 0.044; the weight calculated for the IMF component with a center frequency of 70Hz is approximately 0.010 / 0.073 (a sharp drop in weight), the weight calculated for the IMF component with a center frequency of 75Hz is approximately 0.01 (a very low weight), and the weight calculated for the IMF component with a center frequency of 80Hz is approximately 0.00 (almost zero). In the step of abnormally correcting the weights, all effective components in the 30-60Hz target frequency band do not meet the requirement of being forced to have a weight of 0.

[0065] Figure 5The piecewise Gaussian weighting function curve based on the center frequency is shown, which is used to weight the IMF components in different frequency bands. The weight is the highest in the target frequency band of 30-60Hz, close to 0.07; the weight decays rapidly in the 20-30Hz and 70-80Hz frequency bands, reflecting the suppression of interference signals; the weighting curve is smooth and continuous, avoiding abrupt changes during frequency band switching.

[0066] Figure 6 This study compares the frequency-sensing sensitivity characteristics of an elephant low-frequency perception model (elephants possess a unique ability to perceive low-frequency signals; through the synergistic action of their inner ear hair cells and auditory nerves, they can selectively filter and amplify low-frequency signals in the 1-20Hz range) with the model designed in this embodiment (a piecewise Gaussian weighting function). The elephant model exhibits high sensitivity in the 1-20Hz range; while the model designed in this embodiment extends the high sensitivity range to 30-60Hz, adapting to the early fault characteristics of elevator shaft breakage; demonstrating the optimization and transfer of biomimetic mechanisms in the signal sensing frequency band.

[0067] 5. Perform step S4 to reconstruct the signal. Its spectrum shows that the amplitude in the 30-60Hz range is 85% of the original signal, and the amplitude in the non-target frequency band is ≤0.04V, such as... Figure 7 As shown.

[0068] 6. Execute step S5, and obtain the result after adaptive smoothing. ,like Figure 8 As shown. The time-domain waveform is smooth (ΔA≤0.008V), and the frequency domain target frequency band fluctuation is ≤2.5%.

[0069] 7. Execute step S6 to verify the noise-canceling signal: energy loss rate = 3.2%, SNR improvement = 12.5dB, PSD peaks significantly in the 30-60Hz range, and PSD is only 1 / 12 of the peak value in the 70-80Hz range. Figure 8 As shown.

[0070] The effectiveness of the method was verified by a multi-dimensional evaluation system. The energy loss rate (%), SNR improvement (dB), fault characteristic amplitude retention rate (%), and 70-80Hz PSD suppression ratio were calculated as the standard to evaluate the noise reduction effect. The results are shown in Table 1.

[0071] Table 1 Noise Reduction Evaluation Table method Energy loss rate (%) SNR improvement (dB) Fault feature amplitude retention rate (%) 70-80Hz PSD rejection ratio This method 3.2 12.5 96.8 1 / 12 Traditional wavelet transform 15.8 4.2 84.2 1 / 3 Traditional EMD 18.3 3.8 81.7 1 / 2.5

[0072] As shown in Table 1, the difference between the denoising method proposed in this embodiment and wavelet transform is that the denoising suppression ratio proposed in this embodiment reaches 1 / 12, while the denoising suppression ratio of wavelet transform is only 1 / 3, which greatly improves the signal denoising effect.

[0073] The difference between the noise reduction method proposed in this embodiment and the traditional EMD decomposition is that the method proposed in this patent calculates an energy loss rate of 3.2%, while the energy loss rate of the traditional EMD decomposition is 18.3%.

[0074] The method in this embodiment significantly improves the extraction capability of fault feature signals in the 30-60Hz frequency range, with an energy loss rate of less than 5%, a signal-to-noise ratio improvement of over 10dB, and a significant effect on suppressing noise in non-target frequency bands. Compared with traditional methods, the method in this embodiment has significant advantages in fault feature preservation and noise suppression, enabling early warning of elevator shaft breakage 1-2 months in advance, and is highly practical.

[0075] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for noise reduction of early fault detection signals in elevator shaft breakage, characterized in that, include: Step S1: Obtain the original low-to-medium frequency discrete signal of elevator shaft vibration. The filtered signal was obtained by using a fourth-order Butterworth bandpass filter. ; Step S2, filter the signal Perform complementary set empirical mode decomposition to obtain several IMF components and a residual component, and then select the effective IMF components from all IMF components; Step S3: Calculate the center frequency of each effective IMF component. Based on the center frequency of each valid IMF component Apply Gaussian weights to it; Step S4: Based on the weight of each effective IMF component, reconstruct the signal from all effective IMF components and residual components to obtain the reconstructed signal. ; Step S5: Adaptive moving smoothing filter is used to smoothly reconstruct the signal. The residual noise in the signal is used to obtain the denoised signal. .

2. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, In step S1, a fourth-order Butterworth bandpass filter is used to filter the signal to obtain the filtered signal. Specifically, including, The filter's transfer function H(f) is converted into a discrete unit impulse response h(n) using the impulse response invariance method, and then compared with the original low-to-medium frequency discrete signal. Convolution yields the filtered signal. .

3. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, In step S2, valid IMF components are selected from all IMF components. The strategy includes: Filter IMF components with a main frequency in the range of 30-80Hz.

4. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 4, characterized in that, In step S2, valid IMF components are selected from all IMF components. The strategy also includes... Both condition one and condition two must be satisfied; among them... Condition 1: |PZ|≤1; where P and Z are the number of poles and zero crossings of the same IMF component, respectively. Condition 2: The mean of the upper and lower envelopes of the cubic spline interpolation fitting is ≤ 1 / 20 of the maximum amplitude of the IMF component.

5. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, In step S3, based on the center frequency of each valid IMF component... Apply Gaussian weights to it; include, Calculate the energy integral value for each frequency band and calculate the normalization coefficient. : In the formula, These are the energy integral values ​​for the target frequency band (30-60Hz), the transition frequency band (60-70Hz), the high-frequency noise frequency band (70-80Hz), and the low-frequency interference frequency band (20-30Hz), respectively. Center frequency based on effective IMF components Design a piecewise Gaussian weighting function: 20Hz≤f i <30Hz: 30Hz≤f i <60Hz: 60Hz≤f i ≤70Hz: 70Hz<f i <80Hz: For each valid IMF component, its center frequency is... Substituting the corresponding Gaussian weight function, we obtain the weights. .

6. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, In step S3, after assigning Gaussian weights to each valid IMF component, a weight anomaly correction step is also included. The steps for correcting weight anomalies include: Wavelet thresholding is used to extract the center frequency. The noise component n(n) for each effective IMF component within the 30-60Hz range; Calculate the variance of the effective IMF component and its corresponding noise component n(n) respectively, and calculate the signal-to-noise ratio based on the variance of the two. For effective IMF components with a signal-to-noise ratio less than a preset threshold, force the weight to be 0.

7. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, In step S4, during the signal reconstruction of all valid IMF components and residual components, the weight of the residual components is 0.

8.

8. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, Step S5 also includes: For noise reduction signal Simultaneous time-domain and frequency-domain verification is performed; among them... The criteria for passing time-domain verification are: The amplitude difference between adjacent sampling points in the time domain is ΔA = |x denoise (n+1)-x denoise (n)|≤0.01V; The criteria for passing frequency domain verification are: The amplitude fluctuation in the target frequency band of 30-60Hz is ≤3%, and the amplitude in non-target frequency bands other than the target frequency band of 30-60Hz is ≤0.03V.

9. The method for noise reduction of early fault detection signals in elevator shaft breakage according to claim 1, characterized in that, The method also includes: Step S6, determine the noise reduction signal Does it support early fault detection of elevator shaft breakage? The method for determining this is as follows: If the following conditions are met simultaneously: energy loss rate <5% in the 30-60Hz band, signal-to-noise ratio improvement ≥10dB, power spectral density (PSD)(f) shows a significant peak at 30-60Hz, PSD ≤ 1 / 10 of the peak value in the 70-80Hz band, and PSD ≤ 1 / 5 of the peak value in the 20-30Hz band, then it is supported.

Images