Signal denoising method, elevator, and computer program product

Abstract

The application discloses a signal denoising method, an elevator and a computer program product, and belongs to the field of elevator fault diagnosis, and comprises the following steps: acquiring an original signal and a working condition signal, performing frequency spectrum analysis and filter processing on the original signal; performing complementary ensemble empirical mode decomposition on the filtered signal; screening out effective intrinsic mode function components; performing weight distribution on each effective intrinsic mode function component; determining the weight of a residual component, performing signal reconstruction on all effective intrinsic mode function components and the residual component to obtain a reconstructed signal; performing smoothing processing on residual noise in the reconstructed signal to obtain a denoised signal; and performing single core index verification and time sequence trend consistency verification on the denoised signal. The application can accurately capture weak characteristics in the micro-crack expansion stage, adapt to different working conditions and installation scenes, improve the safety guarantee capability of elevator operation, and provide reliable support for early fault warning of broken shafts.

Description

Technical Field

[0001] This invention relates to the field of elevator fault diagnosis technology, and in particular to a signal noise reduction method, an elevator, and a computer program product. Background Technology

[0002] As a core load-bearing component ensuring elevator operational safety, the structural integrity of the elevator spindle directly affects the elevator's operational stability and passenger safety. The development of spindle breakage faults exhibits a significant progressive characteristic. In the early microcrack propagation stage (crack depth typically less than 0.1 mm), the spindle will excite mid-to-low frequency characteristic signals of 30-60 Hz. However, the amplitude of these mid-to-low frequency characteristic signals is extremely weak and easily masked by the multiple noises generated during elevator operation. These noises mainly include low-frequency interference (20-25 Hz) caused by building structural vibration, friction noise generated by the friction between the car and guide rails (200-300 Hz), and electromagnetic noise (300-500 Hz) caused by motor operation, severely affecting the effective identification of early fault characteristic signals.

[0003] Existing signal denoising technologies have several significant shortcomings in detecting early-stage elevator shaft breakage faults. First, traditional methods generally employ fixed-parameter filtering and decomposition strategies, which cannot adapt to noise frequency fluctuations under different operating conditions such as elevator idling, full load, acceleration, and constant speed. This results in incomplete noise removal or excessive energy loss of target feature signals. Second, when selecting effective fault features from the decomposed signal components, existing methods often rely on single-dimensional criteria such as frequency range or the number of poles and zeros, failing to fully consider signal correlation, phase stability, and signal propagation differences caused by sensor installation locations. This leads to insufficient selection accuracy and the retention of interference components. Third, in the weighting and noise correction stages, existing technologies generally use fixed threshold modes, which are difficult to adapt to original signals with different signal-to-noise ratios and signal characteristics at different stages of fault development, easily introducing irrelevant interference components. Furthermore, the lack of dynamic adjustment mechanisms for smoothing filtering and verification indicators makes them poorly adaptable to the weak impact characteristics generated by early microcracks, and the absence of trend analysis of time-series signals leads to a high false alarm rate.

[0004] Therefore, there is an urgent need to design a signal denoising method with dual adaptive capabilities of working conditions and position, multi-dimensional fine-tuning function, and dynamic correction characteristics throughout the entire process. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide a signal noise reduction method, elevator and computer program product to achieve accurate extraction and efficient noise suppression of target signals of early failure of elevator shaft, so as to meet the actual needs of early warning of early failure of elevator shaft.

[0006] In a first aspect, this application provides a signal noise reduction method, comprising the following steps: The original signal is acquired, its spectrum is analyzed and the original signal-to-noise ratio is calculated. The operating condition signal is determined based on the original signal, and the filter coefficients are determined and filtered in combination with the operating condition signal to obtain the filtered signal. Complementary set empirical mode decomposition is performed on the filtered signal. The decomposition parameters are dynamically adjusted according to the original signal-to-noise ratio to obtain multiple intrinsic mode function components and one residual component. The effective intrinsic mode function components are selected from all intrinsic mode function components. The Hilbert transform is performed on each effective intrinsic mode function component to obtain the analytic signal. The instantaneous phase and instantaneous frequency are obtained based on the analytic signal. The center frequency of each effective intrinsic mode function component is obtained through discrete integration. The weights of each effective intrinsic mode function component are assigned based on the center frequencies. After the assignment, weight anomaly correction is performed. The weights of the residual components are determined based on the fault development stage and the original signal-to-noise ratio. The signals are then reconstructed by reconstructing all effective intrinsic mode function components and the residual components to obtain the reconstructed signal. Local feature calculations are performed on the reconstructed signal. Based on the calculation results, the half-length of the moving smoothing window is adaptively adjusted. An adaptive moving smoothing filtering method is used to smooth the residual noise in the reconstructed signal to obtain a denoised signal.

[0007] Optionally, complementary ensemble empirical mode decomposition is performed on the filtered signal. The decomposition parameters are dynamically adjusted according to the original signal-to-noise ratio to obtain multiple intrinsic mode function components and one residual component. Valid intrinsic mode function components are selected from all intrinsic mode function components, including: Set the decomposition parameters of the complementary set empirical mode decomposition based on the original signal-to-noise ratio; Complementary set empirical mode decomposition is performed on the filtered signal to obtain multiple intrinsic mode function components and a residual component. Based on the selection range, components that simultaneously meet the five-dimensional selection criteria are selected from all intrinsic mode function components to obtain effective intrinsic mode function components.

[0008] Optionally, the five-dimensional screening conditions include: the dominant frequency is within the effective intrinsic mode function component screening range; the number of poles and the number of zero crossings meet preset conditions; the mean of the upper and lower envelopes fitted by cubic spline interpolation is less than or equal to a preset proportion threshold of the maximum amplitude of the intrinsic mode function component; the cross-correlation coefficient with the target frequency band is greater than or equal to a preset correlation threshold; and the instantaneous phase fluctuation value is less than or equal to a preset phase stability threshold.

[0009] Optionally, local feature calculations are performed on the reconstructed signal, and the half-length of the moving smoothing window is adaptively adjusted based on the calculation results. An adaptive moving smoothing filtering method is then used to smooth the residual noise in the reconstructed signal to obtain a denoised signal, including: Calculate the local features of the reconstructed signal, including local mean, local standard deviation, instantaneous amplitude change rate, kurtosis, and local energy concentration. The half-length of the moving window is adaptively adjusted based on the triple features of kurtosis, instantaneous amplitude change rate, and local energy concentration. When the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, and the local energy concentration is greater than the local energy concentration threshold, the half-length of the window is set to N1; when the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, or the local energy concentration is greater than the local energy concentration threshold, the half-length of the window is set to N2; in other cases, the half-length of the window is set to N3, where N1, N2, and N3 are constants. An adaptive moving smoothing filter method is used to smooth the residual noise in the reconstructed signal to obtain a denoised signal.

[0010] Optionally, local feature calculations are performed on the reconstructed signal, and the half-length of the moving smoothing window is adaptively adjusted based on the calculation results. An adaptive moving smoothing filtering method is used to smooth the residual noise in the reconstructed signal to obtain a denoised signal. The method also includes verifying the denoised signal from both the time domain and the frequency domain.

[0011] Optionally, the signal denoising method also includes performing single-time core indicator verification and time-series trend consistency verification on the denoised signal, and performing adaptation correction in conjunction with the operating condition signal.

[0012] Optionally, the single core indicator verification includes energy loss rate verification, signal-to-noise ratio improvement verification, and power spectral density verification. The time-series trend consistency verification includes: the energy of the target frequency band shows an increasing trend, and the energy growth of two adjacent groups of signals is greater than or equal to the third threshold. At the same time, the least squares method is used to fit the energy trend line, and the goodness of fit of the energy trend line is greater than or equal to the fourth threshold.

[0013] Optionally, the operating condition signal includes the sensor installation location and the elevator operating condition.

[0014] Secondly, this application also provides an elevator, comprising: a body, a sensor, and a computing unit, wherein the sensor acquires the raw signals of the body and outputs them to the computing unit, and the computing unit performs the signal noise reduction method as described in any one of the first aspects.

[0015] Thirdly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the signal noise reduction method as described in any one of the first aspects.

[0016] This application provides a signal denoising method, an elevator, and a computer program product. By collecting data on sensor installation locations and elevator operating conditions, it effectively overcomes the poor adaptability of traditional fixed-parameter methods under varying operating conditions, avoiding incomplete noise suppression or target signal loss under different operating conditions and installation locations. Through an adaptive smoothing strategy, it strongly suppresses residual noise while maximally protecting the transient impact components characterizing microcrack propagation, enhancing residual noise suppression while preserving weak impact characteristics. By verifying the consistency between signal reconstruction and energy growth trends and temporal trends, it further reduces the risk of misjudgment, ensuring the consistency and reliability of denoising results. Furthermore, a differentiated adaptation strategy is designed for the sensor installation location, significantly improving the method's versatility and making it widely applicable to early shaft breakage fault detection in elevators under different installation scenarios.

[0017] To make the above-mentioned features and advantages of the invention more apparent and understandable, specific embodiments are described below, and detailed descriptions are provided in conjunction with the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a signal noise reduction method provided in one embodiment of this application.

[0020] Figure 2 This is a flowchart of step S2 in a signal noise reduction method provided in one embodiment of this application.

[0021] Figure 3 This is a flowchart of step S5 in a signal noise reduction method provided in one embodiment of this application.

[0022] Figure 4 The original signal is shown in its time domain and spectrum. Figure 4 (a) is the time-domain plot of the original signal; Figure 4 (b) is the spectrum of the original signal.

[0023] Figure 5 Here are the filtered time-domain and spectrum diagrams, where Figure 5 (a) is the time-domain plot after filtering; Figure 5 (b) is the filtered spectrum.

[0024] Figure 6 The graph shows the intrinsic mode function components and residual components obtained from the empirical mode decomposition of complementary sets.

[0025] Figure 7 This is a graph of the time-frequency joint weighting function.

[0026] Figure 8 The time-domain waveform of the reconstructed signal.

[0027] Figure 9 This is the time-domain waveform of the noise-reduced signal.

[0028] Figure 10 The graph shows a comparison of the target frequency band power spectral density of the original signal, the signal denoised by the traditional method, and the signal denoised by the method of this application.

[0029] Figure 11 This is a graph showing the energy trend of five consecutive signals in the 30-60Hz frequency band. Detailed Implementation

[0030] To make the objectives and technical solutions of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the described embodiments of this application without creative effort are within the scope of protection of this application.

[0031] In one embodiment, see Figure 1 This application provides a signal noise reduction method, which may include the following steps: steps S1 to S5.

[0032] Step S1: Acquire the original signal, perform spectral analysis on the original signal and calculate the original signal-to-noise ratio, determine the operating condition signal, determine the filter coefficients based on the operating condition signal and perform filtering processing to obtain the filtered signal.

[0033] Step S2: Perform complementary set empirical mode decomposition on the filtered signal, dynamically adjust the decomposition parameters according to the original signal-to-noise ratio, obtain multiple intrinsic mode function components and one residual component, and select the effective intrinsic mode function components from all intrinsic mode function components.

[0034] Step S3: Perform Hilbert transform on each effective intrinsic mode function component to obtain an analytical signal. Based on the analytical signal, obtain the instantaneous phase and instantaneous frequency. Obtain the center frequency of each effective intrinsic mode function component through discrete integration. Based on the center frequency, assign weights to each effective intrinsic mode function component. After assignment, perform weight anomaly correction.

[0035] Step S4: Determine the weight of the residual component based on the fault development stage and the original signal-to-noise ratio, and reconstruct the signal by performing signal reconstruction on all effective intrinsic mode function components and residual components to obtain the reconstructed signal.

[0036] Step S5: Perform local feature calculation on the reconstructed signal, adaptively adjust the half-length of the moving smoothing window based on the calculation results, and use the adaptive moving smoothing filtering method to smooth the residual noise in the reconstructed signal to obtain the denoised signal.

[0037] In this application's signal denoising method, a dual adaptive filtering mechanism for operating condition signals is employed to achieve fully dynamic matching of bandpass filter parameters with the actual elevator operating state and monitoring location. This ensures precise matching of core parameters such as filtering and decomposition with operating condition information, effectively solving the problem of noise residue or feature loss caused by traditional fixed-parameter filtering. It also avoids incomplete noise suppression or target signal loss under different operating conditions. Furthermore, an adaptive moving smoothing filter method dynamically adjusts the smoothing window, enhancing residual noise suppression while preserving weak impact characteristics. Adaptive reconstruction during the fault stage and consistency verification of temporal trends further reduce the risk of misjudgment. Simultaneously, a differentiated adaptation strategy is designed for the sensor installation location, significantly improving the method's versatility and making it widely applicable to early fault detection of elevator shaft breakage in various installation scenarios. This application's method can adaptively and accurately extract and enhance weak early fault features from a strong noise background, outputting rigorously verified high-quality signals and clear detection judgments, providing a stable and reliable technical means for early warning of elevator shaft breakage faults.

[0038] As an example, the signal noise reduction method of this application can be used to reduce the noise of weak vibration signals from early cracks or structural damage in elevator spindle systems, fan spindles, pump shafts, drive shafts, and other mechanical shafts.

[0039] The following section uses a broken shaft elevator as an example to explain steps S1 to S5 in detail.

[0040] In step S1, please refer to Figure 1 In step S1, the original signal is acquired, the spectrum of the original signal is analyzed and the original signal-to-noise ratio is calculated, the operating condition signal is determined, the filter coefficients are determined in combination with the operating condition signal and the filtering process is performed to obtain the filtered signal.

[0041] Specifically, the target frequency band is set to 30-60Hz. The raw vibration signals of the elevator shaft are collected using vibration sensors installed at key locations on the elevator shaft. x (n), synchronously determine the operating condition signal.

[0042] As an example, the operating condition signal may include the sensor installation location and the elevator operating condition. The sensor installation location may include three categories: shaft end, bearing housing, and traction machine base; the elevator operating condition can be divided into four types—no-load, full-load, acceleration, and constant speed—by identifying the amplitude variance and frequency distribution characteristics of the original signal.

[0043] As an example, a piezoelectric accelerometer or a high-precision microelectromechanical system (MEMS) vibration sensor can be used to acquire broadband vibration analog signals generated during shaft operation, which are then converted into discrete raw vibration signals of the elevator shaft via an analog-to-digital converter (ADC). x (n).

[0044] As an example, the original signal x (n) represents a low-to-medium frequency discrete signal.

[0045] Furthermore, regarding the original signal x (n) Perform spectrum analysis and calculate the original signal-to-noise ratio. The energy ratio method is used to accurately identify the upper limit of low-frequency interference. f L With high frequency noise lower limit f H Among them, the upper limit of low-frequency interference f L The upper limit of the lowest frequency band with an energy share of ≥5%; lower limit of high-frequency noise. f H The lower limit value corresponding to the highest frequency band with an energy share of ≥5%, based on the upper limit of low-frequency interference. f L With high frequency noise lower limit f H The core distribution range of the noise was determined as follows .

[0046] As an example, the raw signal-to-noise ratio The expression is: in, This is the variance calculation function. The effective vibration characteristic signal components in the original signal. denoted as the noise component in the original signal, and n as the discrete time series index.

[0047] Furthermore, the filter coefficients are determined based on the core distribution range of the noise, the elevator operating conditions, and the sensor installation location. These filter coefficients may include a low cutoff frequency. High cutoff frequency Filter order M.

[0048] As an example, low cutoff frequency The expression is: in, This is the upper limit for low-frequency interference. This is the correction amount for operating conditions.

[0049] As an example, operating condition correction amount The settings can be configured according to the elevator's operating conditions; for example, the correction amount can be set for no-load conditions. Correction amount under full load conditions Correction amount under acceleration conditions Correction amount under constant speed operating conditions .

[0050] As an example, high cutoff frequency The expression is: in, This is the lower limit of high-frequency noise. This is the position correction amount.

[0051] As an example, position correction amount The settings can be configured based on the sensor's installation location. For example, when the sensor is installed on the shaft end, a position correction amount can be set; when it is installed on a bearing housing, a position correction amount can be set. When the traction machine base is installed, a position correction amount can be set. .

[0052] As an example, the filter order M can be determined based on the original signal-to-noise ratio. Adjust flexibly according to elevator operating conditions, when the original signal-to-noise ratio At that time, the 6th order was uniformly used for all operating conditions; when the original signal-to-noise ratio At this time, the fourth order is used under no-load and constant speed conditions, and the fifth order is used under full-load and acceleration conditions, in order to balance the stopband attenuation capability and signal processing efficiency.

[0053] As an example, a Butterworth bandpass filter could be used as the filter.

[0054] Furthermore, the transfer function of the filter is obtained through the impulse response invariance method. H ( f Convert to discrete unit impulse responseh ( n ), and with the original signal x ( n Perform convolution operation to obtain the filtered signal. x bp ( n By employing dual adaptive filtering based on signal-to-noise ratio under different operating conditions, the filtering parameters are precisely matched with the actual operating state and installation scenario of the elevator. This solves the problem that existing filtering and decomposition methods using fixed parameters cannot adapt to various operating conditions such as elevator idling, full load, and acceleration, as well as signal changes caused by different sensor installation locations, resulting in incomplete noise suppression or loss of the target signal.

[0055] Furthermore, the filtering effect is verified by spectrum analysis to determine the filtered signal. x bp ( n If the first verification criterion is met, proceed to the next step; otherwise, reset the filter coefficients by adjusting the safety margin or correction amount.

[0056] As an example, the first verification criterion can be set to a low cutoff frequency. The signal amplitude in the following frequency bands does not exceed 1 / 15 of the original signal amplitude at 30Hz, high cutoff frequency The signal amplitude in the above frequency bands shall not exceed 1 / 20 of the amplitude at 80Hz of the original signal, to ensure that out-of-band noise is fully suppressed.

[0057] In step S2, please refer to Figure 1 In step S2, complementary set empirical mode decomposition is performed on the filtered signal. The decomposition parameters are dynamically adjusted according to the original signal-to-noise ratio to obtain multiple intrinsic mode function components and a residual component. Valid intrinsic mode function components are selected from all intrinsic mode function components.

[0058] For example, please refer to Figure 2 Step S2 may include the following steps: Step S21 to Step S23.

[0059] Step S21: Set the decomposition parameters of the complementary set empirical mode decomposition based on the original signal-to-noise ratio.

[0060] Step S22: Perform complementary set empirical mode decomposition on the filtered signal to obtain multiple intrinsic mode function components and a residual component.

[0061] Step S23: Based on the filtering range, select the components that simultaneously meet the five-dimensional filtering conditions from all intrinsic mode function components to obtain the effective intrinsic mode function components.

[0062] Specifically, in step S21, based on the original signal-to-noise ratio The decomposition parameters of the complementary set empirical mode decomposition are set and dynamically adjusted to adapt to the original signals of different qualities.

[0063] As an example, when At that time, the overall average number of times can be set. Next, decomposition termination condition Noise interference can be reduced by increasing the overall average ensemble number and increasing the severity of the decomposition termination condition SD; when At that time, the overall average number of times can be set. Next, decomposition termination condition To balance decomposition accuracy and computational efficiency; when At that time, the overall average number of times can be set. Next, decomposition termination condition Relaxed to It can improve processing speed while ensuring processing effectiveness.

[0064] Further, in step S22, the filtered signal... x bp ( n Perform complementary set empirical mode decomposition, specifically, generate two sets of filtered signals with amplitude values. x bp ( n Inverse white Gaussian noise with a standard deviation of 0.2 times was compared with the filtered signal. x bp ( n The two noisy signals are superimposed and then subjected to Empirical Mode Decomposition (EMD). The EMD process follows a classic screening procedure: first, the local maxima and minima of each noisy signal are identified; then, the upper and lower envelopes of the signals are fitted using cubic spline interpolation; the mean of the upper and lower envelopes is calculated and subtracted from the mean of the original noisy signal to obtain an intermediate signal; this process is repeated until the intermediate signal meets the screening range of intrinsic mode function components, ultimately yielding multiple intrinsic mode function components and one residual component. r ( n ).

[0065] Furthermore, in step S23, the selection range of effective intrinsic mode function components is determined based on the sensor installation position correction.

[0066] As an example, the filtering range can be set to 30-80Hz when the sensor is installed at the shaft end, 32-82Hz when it is installed at the bearing housing, and 31-81Hz when it is installed at the traction machine base, to accommodate the signal propagation differences caused by different sensor installation positions.

[0067] Furthermore, from all intrinsic mode function components, components that simultaneously satisfy the five-dimensional screening criteria are selected to obtain the effective intrinsic mode function components. IMF h ( n The effective intrinsic mode function components IMF h ( n It has a strong correlation with the target frequency band signal and its phase is stable. h This is the index of the effective intrinsic mode function components.

[0068] As an example, the five-dimensional screening criteria are: the dominant frequency is within the effective intrinsic mode function (IMF) component screening range; the number of poles P and the number of zero crossings Z meet preset conditions; the mean of the upper and lower envelopes fitted by cubic spline interpolation is less than or equal to a preset proportion threshold of the maximum amplitude of the IMF component; the cross-correlation coefficient with the target frequency band is greater than or equal to a preset correlation threshold; and the instantaneous phase fluctuation value is less than or equal to a preset phase stability threshold. These five-dimensional screening criteria make the screening more rigorous and refined, effectively eliminating interference components with weak correlation to fault characteristics or phase disturbances. This solves the problem of existing screening methods not considering the correlation and phase stability of components with fault characteristics, thus retaining interference components or losing useful components.

[0069] As an example, the preset conditions can be: Where P represents the number of poles and Z represents the number of zero crossings.

[0070] As an example, the preset ratio threshold can be 1 / 20.

[0071] As an example, the preset correlation threshold can be set to 0.6, which is the cross-correlation coefficient with the target frequency band. .

[0072] As an example, the preset phase stability threshold can be 0.1 rad.

[0073] In step S3, please refer to Figure 1 In step S3, a Hilbert transform is performed on each effective intrinsic mode function component to obtain an analytic signal. The instantaneous phase and instantaneous frequency are obtained based on the analytic signal. The center frequency of each effective intrinsic mode function component is obtained through discrete integration. The weights of each effective intrinsic mode function component are assigned based on the center frequencies. After the assignment, weight anomaly correction is performed.

[0074] Specifically, for each effective eigenmode function component IMF h ( n Perform a Hilbert transform to obtain the analytic signal. The expression is: in, The imaginary unit, To effectively identify the modal function components The Hilbert transform result can be obtained by numerical integration.

[0075] Furthermore, extract the analytical signal. phase ,in, i Indicates the index of the valid IMF component. h This is the index of the effective intrinsic mode function components.

[0076] Furthermore, based on phase Calculate instantaneous frequency The expression is: in, , These are the phase values ​​at time n+1 and time n, respectively. The sampling period is Pi is a constant.

[0077] As an example, the sampling period can be set. .

[0078] Furthermore, the center frequency of each effective eigenmode function component is obtained through discrete integration. To control the calculation error to no more than 0.3Hz, after the calculation is completed, all effective intrinsic mode function components are included. By center frequency Sort by size from smallest to largest.

[0079] Furthermore, based on the center frequency For each effective intrinsic mode function component Weights are assigned, and weight anomaly correction is performed after assignment. Specifically, the energy integral value of each key frequency band is calculated. The key frequency bands include 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), resulting in the energy integral values ​​W1 for the target frequency band, W2 for the transition frequency band, W3 for the high-frequency noise frequency band, and W4 for the low-frequency interference frequency band. A normalization coefficient is then calculated based on the energy integral values ​​of each key frequency band. The expression is: in, The energy integral value for the target frequency band. This is the energy integral value for the transition frequency band. This represents the energy integral value in the high-frequency noise band. This represents the energy integral value for the low-frequency interference band.

[0080] Furthermore, based on the normalization coefficient Determine the time-frequency-phase joint weighting function The expression for weight allocation is: in, For the frequency weights of the piecewise Gaussian function, For the first i The center frequencies of the effective intrinsic mode function components Instantaneous amplitude weight, For the instantaneous amplitude of the effective intrinsic mode function components, For instantaneous signal-to-noise ratio, For the signal-to-noise ratio of the effective intrinsic mode function components, For phase consistency weight, The phase consistency coefficient between the effective intrinsic mode function components and the target frequency band.

[0081] As an example, instantaneous amplitude weighting The expression is: in, This represents the instantaneous amplitude of the effective intrinsic mode function components. This represents the maximum instantaneous value of the effective characteristic mode function component.

[0082] Furthermore, the sym8 wavelet basis is used to perform multi-scale decomposition of all effective intrinsic mode function components. Specifically, the wavelet threshold is dynamically set based on noise estimation. The expression is: in, The standard deviation of noise is an estimate of noise intensity. This is the wavelet threshold adjustment coefficient.

[0083] As an example, wavelet threshold adjustment coefficient Based on the original signal-to-noise ratio Setting, when the original signal-to-noise ratio At that time, wavelet threshold adjustment coefficient When the original signal-to-noise ratio At that time, wavelet threshold adjustment coefficient .

[0084] Furthermore, noise components are extracted through soft thresholding. Calculate the signal-to-noise ratio of each effective intrinsic mode function component. Perform weight anomaly correction, and improve the signal-to-noise ratio of the effective intrinsic mode function components. The expression is: in, For each effective intrinsic mode function component variance Let Variance be the variance of the noise component. It is a logarithmic function with base 10.

[0085] As an example, noise components The expression is: in, For effective intrinsic mode function components, The component after wavelet denoising.

[0086] As an example, a dynamic threshold can be set. If the signal-to-noise ratio of the effective intrinsic mode function components Then the time-frequency-phase joint weighting function is forcibly set. To avoid interference introduced by low signal-to-noise ratio components, among which, This represents the original signal-to-noise ratio.

[0087] In step S4, please refer to Figure 1 In step S4, the weights of the residual components are determined based on the fault development stage and the original signal-to-noise ratio. The signals of all effective intrinsic mode function components and residual components are reconstructed to obtain the reconstructed signal.

[0088] Specifically, considering the residual components obtained from the empirical mode decomposition of complementary sets... r ( n It contains both the low-frequency trend of the signal and may carry some low-frequency noise, so it can be used to determine the fault development stage and the original signal-to-noise ratio. Dynamically determine the weights of the residual components .

[0089] As an example, the stage of fault development can be determined by the energy growth trend of the first three sets of signals in the 30-60Hz frequency band, including the initial stage of microcrack, the middle stage of microcrack, and the later stage of microcrack. Among them, an increase of ≤5% indicates the initial stage of microcrack, an increase of 5%-10% indicates the middle stage of microcrack, and an increase of >10% indicates the later stage of microcrack.

[0090] As an example, residual component weights Based on the original signal-to-noise ratio It can be set to, when the original signal-to-noise ratio At that time, the weight of the residual component in the early stage of microcracks Weight of residual components in the middle stage of microcracks Weight of residual components in the later stage of microcracks When the original signal-to-noise ratio At that time, the weight of the residual component in the early stage of microcracks Weight of residual components in the middle stage of microcracks Weight of residual components in the later stage of microcracks It achieves a balance between low-frequency trend preservation and noise suppression through dynamic adjustment.

[0091] Furthermore, by combining the weights of each effective intrinsic mode function component, the signal is reconstructed according to the discrete reconstruction formula to obtain the reconstructed signal. The expression is: in, The number of effective intrinsic mode function components, The weight of the h-th effective intrinsic mode function component. For the h-th effective intrinsic mode function component, The weights of the residual components, The residual components are obtained from the empirical mode decomposition of complementary sets.

[0092] By using time-frequency phase joint weighting, the value of each component is evaluated more scientifically, which solves the problem that the existing method of assigning weights only based on the center frequency ignores the differences in the contributions of different components in terms of instantaneous amplitude, signal-to-noise ratio, etc., thus limiting the quality of the reconstructed signal.

[0093] As an example, step S4 involves determining the weight of the residual component based on the fault development stage and the original signal-to-noise ratio, reconstructing the signal from all effective intrinsic mode function components and the residual component, and then verifying the reconstructed signal.

[0094] Specifically, for the reconstructed signal Verification and signal reconstruction are performed. The verification standard can be set to a first set range where the amplitude of the target frequency band is greater than or equal to the amplitude of the original signal in that frequency band, and the amplitude of the non-target frequency band is less than or equal to a set first rated voltage, so as to ensure that the characteristics of the target signal are enhanced while the interference signal is effectively suppressed.

[0095] As an example, the first set range can be set to 88%.

[0096] As an example, the first rated voltage can be set to 0.035V.

[0097] In step S5, please refer to Figure 1In step S5, local feature calculation is performed on the reconstructed signal. Based on the calculation results, the half-length of the moving smoothing window is adaptively adjusted. The adaptive moving smoothing filtering method is used to smooth the residual noise in the reconstructed signal to obtain a denoised signal.

[0098] For example, please refer to Figure 3 Step S5 may include the following steps: Step S51 to Step S53.

[0099] Step S51: Calculate the local features of the reconstructed signal, including local mean, local standard deviation, instantaneous amplitude change rate, kurtosis, and local energy concentration.

[0100] Step S52: Adaptively adjust the half-length of the moving smoothing window based on the triple features of kurtosis, instantaneous amplitude change rate, and local energy concentration. When the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, and the local energy concentration is greater than the local energy concentration threshold, set the half-length of the window to N1; when the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, or the local energy concentration is greater than the local energy concentration threshold, set the half-length of the window to N2; in other cases, set the half-length of the window to N3, where N1, N2, and N3 are constants.

[0101] Step S53: Use an adaptive moving smoothing filter method to smooth the residual noise in the reconstructed signal to obtain a denoised signal.

[0102] Specifically, in step S51, the reconstruction signal is determined. Local features, including local mean Local standard deviation Instantaneous amplitude change rate , cliff and local energy concentration .

[0103] As an example, local mean The expression is: in, It is a local window of half length, usually small, used for local feature extraction; To reconstruct the signal in The amplitude at any given moment.

[0104] As an example, local standard deviation The expression is: in, It is a local mean. For a local window half-length, To reconstruct the signal in The amplitude at any given moment.

[0105] As an example, instantaneous amplitude change rate The expression is: in, To reconstruct the signal in Amplitude at time, To reconstruct the signal in Amplitude at time, The sampling period.

[0106] As an example, kurtosis based on the fourth central moment The expression is: in, To reconstruct the signal in Amplitude at time, It is a local mean. For a local window half-length, This represents the local standard deviation.

[0107] As an example, local energy concentration It can be set as the ratio of local target frequency band energy to local total energy, and comprehensively characterize the local features of the signal through multi-feature fusion.

[0108] Further, in step S52, based on kurtosis Instantaneous amplitude change rate With local energy concentration Triple feature adaptive adjustment of the window half-length for smooth movement When the steepness Greater than the kurtosis threshold ( And instantaneous amplitude change rate Greater than the instantaneous amplitude change rate threshold and local energy concentration When the value exceeds the local energy concentration threshold, it indicates that the signal within the window contains strong impulse characteristics. Therefore, a window half-length should be set. To avoid excessive smoothing of impact details; when kurtosis Greater than the kurtosis threshold or instantaneous amplitude change rate Greater than the instantaneous amplitude change rate threshold or local energy concentration When the local energy concentration exceeds the threshold, set the window half-length. In other cases, it indicates that the signal within the window is relatively smooth, without obvious impact characteristics, and a half-length window is set. This enhances the smoothing effect on residual noise.

[0109] As an example, , , It can be a specific value, which can be set according to the actual situation. In this specific embodiment, it is set as follows: The corresponding window length is 3 points; settings The corresponding window length is 5 points; settings The corresponding window length is 11 points.

[0110] As an example, kurtosis threshold Based on the original signal-to-noise ratio Dynamically set, the expression is: Where 2.5 is the kurtosis baseline value, 0.5 is the adjustment coefficient, and 10 is the signal-to-noise ratio normalization coefficient.

[0111] As an example, the instantaneous amplitude change rate threshold can be set to 0.005V / s.

[0112] As an example, the local energy concentration threshold can be set to 0.8.

[0113] Furthermore, in step S53, an adaptive moving smoothing filter method is used to smooth the residual noise in the reconstructed signal, based on the adjusted window half-length. Calculate the noise-reduced signal The expression is: in, Indicates half the length of the window. To reconstruct the signal in Amplitude at time, The sum of the reconstructed signal amplitudes within the window. This is the window length.

[0114] As an example, the essence of denoising a signal is to use a local mean. As noise reduction outputs, the two are two expressions of the same calculation process.

[0115] Furthermore, step S5 may also include step S54: verifying the denoised signal from both the time domain and the frequency domain.

[0116] Specifically, in step S54, the denoised signal is processed in both the time and frequency domains after smoothing. Verification is performed. The time-domain verification standard is: when the original signal-to-noise ratio... At that time, the instantaneous amplitude change rate of adjacent sampling points When the original signal-to-noise ratio At that time, the instantaneous amplitude change rate of adjacent sampling points The frequency domain verification standard is as follows: the amplitude fluctuation of the target frequency band shall not exceed the first amplitude, and the amplitude of the non-target frequency band shall not exceed the second amplitude, so as to ensure a balance between the smoothing effect and the preservation of the target signal.

[0117] As an example, the first amplitude can be based on the original signal-to-noise ratio. Dynamic adjustment can be set to .

[0118] As an example, the second amplitude can be set to 0.025V.

[0119] In another specific embodiment, the signal denoising method provided in this application may further include step S6: performing single core index verification and time-series trend consistency verification on the denoised signal, performing adaptation correction in combination with the sensor installation position, determining whether the denoised signal supports early fault detection of elevator shaft breakage, and finally outputting the denoised signal and the determination result.

[0120] Specifically, for the noise-reduced signal Perform single-test core performance indicator (SPI) verification, including energy loss rate verification, signal-to-noise ratio (SNR) improvement verification, and power spectral density (PSD) verification. The denoised signal must meet the requirements. The energy loss rate is less than or equal to the first threshold; the signal-to-noise ratio improvement is greater than or equal to the second threshold; the power spectral density of the target frequency band has a significant peak, and the power spectral density of the 70-80Hz band does not exceed 1 / 15 of the peak value of the target frequency band, and the power spectral density of the 20-30Hz band does not exceed 1 / 8 of the peak value of the target frequency band.

[0121] As an example, the expression for the energy loss rate is: Among them, the energy of the original signal in the target frequency band is the energy integral value of the original signal in the 30-60Hz frequency band, and the energy of the denoised signal in the 30-60Hz frequency band is the energy integral value of the denoised signal in the 30-60Hz frequency band.

[0122] As an example, the first threshold is the energy loss rate tolerance threshold, which can be set to 2.5%.

[0123] As an example, the second threshold can be set to 15dB.

[0124] For example, energy loss rate verification requirements To ensure that the energy of the target signal is not excessively lost; the signal-to-noise ratio (SNR) improvement verification requires a higher SNR after denoising. Compared with the original signal-to-noise ratio The difference is greater than or equal to 15 dB, that is To ensure that noise is effectively suppressed; power spectral density verification requires that the target frequency band exhibit a significant peak at 30-60Hz, the power spectral density verification at 70-80Hz should not exceed 1 / 15 of the peak at 30-60Hz, and the power spectral density verification at 20-30Hz should not exceed 1 / 8 of the peak at 30-60Hz, to ensure the recognizability of the target features.

[0125] Furthermore, multiple consecutive sets of denoised signals are collected for temporal trend consistency verification. This verification includes verifying that the energy of the target frequency band exhibits an increasing trend, and that the energy increase between two adjacent sets of signals is greater than or equal to a third threshold. Simultaneously, the least squares method is used to fit the energy trend line, and the goodness of fit R of the energy trend line is... 2 The value should be greater than or equal to the fourth threshold to conform to the pattern of gradual enhancement of characteristic signals during microcrack propagation.

[0126] As an example, you can set up to collect five consecutive sets of denoised signals to verify the consistency of the time series trend.

[0127] As an example, the third threshold can be set to 2%.

[0128] As an example, the fourth threshold can be set to 0.95.

[0129] Furthermore, the detection results are fully adapted and corrected based on the sensor's installation location and signal attenuation characteristics. The full adaptation correction includes: when the sensor is installed at the shaft end, the power spectral density peak verification standard remains unchanged, the energy loss rate tolerance threshold is less than or equal to X, and the signal attenuation coefficient is E1; when the sensor is installed at the bearing housing, the power spectral density peak verification standard is reduced by F1, the energy loss rate tolerance threshold is less than or equal to Y, and the signal attenuation coefficient is E2; when the sensor is installed at the traction machine base, the power spectral density peak verification standard is reduced by F2, the energy loss rate tolerance threshold is less than or equal to Z, and the signal attenuation coefficient is E3.

[0130] As an example, X, Y, Z, F1, F2, E1, E2, and E3 can be specific values ​​that can be set according to actual conditions. In this specific embodiment, X=2.5%, Y=4%, Z=6%, F1=15%, F2=20%, E1=0.95, E2=0.85, and E3=0.75.

[0131] Further, it is determined whether the noise reduction signal supports early fault detection of elevator shaft breakage. If the noise reduction signal meets all the requirements of single-indicator verification and time-series trend consistency verification, it is determined that the noise reduction signal supports early fault detection of elevator shaft breakage, and the noise reduction signal and judgment result are finally output. If it does not meet all the requirements, it is marked as invalid detection and a warning is issued, and manual review is initiated.

[0132] In one specific embodiment, to verify the actual effect and superiority of the method of the present invention, the steps described herein are specifically implemented and their effects are demonstrated through a complete embodiment, combined with specific sensor configurations, elevator operating conditions, and signal parameters. The experiment uses an elevator main shaft system with steel bearings, and the traction sheave rotation frequencies are 30Hz and 60Hz. Human interventions include lead breakage, knocking, and friction. Infrasound sensors are used to collect vibration signals at a sampling frequency of 1000Hz and a sampling duration of 10s.

[0133] Further, step S1 is executed to obtain the original low-to-medium frequency discrete signal of the elevator shaft vibration, i.e., the original signal. x (n), synchronously acquire sensor installation location, based on raw signal x (n) Determine the elevator operating condition data and analyze the original signals. x (n) Perform spectrum analysis and calculate the original signal-to-noise ratio. The energy percentage method is used to identify the upper limit of the lowest frequency band with an energy percentage ≥ 5%. f L With the lowest frequency band f H The parameters of a 4th-order Butterworth bandpass filter are dynamically adjusted based on elevator operating conditions and sensor positions to achieve a low cutoff frequency. according to Calculation of high cutoff frequency according to Calculate the operating condition corrections for no-load, full-load, acceleration, and constant-speed operating conditions. The corresponding positional corrections for the shaft end, bearing housing, and traction machine base are 0Hz, 1Hz, 1.5Hz, and 0.5Hz, respectively. The frequencies are 0Hz, 2Hz, and 3Hz, respectively. The filter transfer function is obtained using the impulse response invariance method. H ( f Convert to discrete unit impulse response h ( n ), with the original signal x ( n Convolution yields the filtered signal. x bp ( n The filtering effect needs to meet the low cutoff frequency requirement. The following frequency bands have amplitudes not exceeding 1 / 15 of the original signal amplitude at 30Hz, and are high cutoff frequencies. The amplitude of the above frequency bands shall not exceed 1 / 20 of the amplitude of the original signal at 80Hz. Figure 4 The original signal is shown in its time domain and spectrum. Figure 4 (a) is the time-domain plot of the original signal. Figure 4 (b) is the spectrum of the original signal; Figure 5The filtered signal after adaptive bandpass filtering x bp ( n The time domain and spectrum of ) Figure 5 (a) is the time-domain plot after filtering; Figure 5 (b) is the filtered spectrum. It can be seen that after adaptive bandpass filtering, the waveform smoothness is improved and noise redundancy is reduced in the time domain; in the spectrum, energy is concentrated in the target frequency band of 30-60Hz, which significantly improves the identification of fault characteristics.

[0134] Furthermore, after completing the filtering preprocessing in step S1, the filtered signal is... x bp ( n Perform complementary set empirical mode decomposition, i.e., step S2. Based on the original signal-to-noise ratio... Dynamically adjust decomposition parameters: when At that time, the overall average number of times Next, decomposition termination condition ;when At that time, the overall average number of times Next, decomposition termination condition ;when At that time, the overall average number of times Next, decomposition termination condition Relaxed to Before decomposition, two sets of filtered signals with amplitude values ​​are generated. x bp ( n Inverse white Gaussian noise with a standard deviation of 0.2 times was compared with the filtered signal. x bp ( n After superposition, empirical mode decomposition is performed, ultimately yielding multiple eigenmode function components and one residual component. r ( n The selection frequency range of effective intrinsic mode function components is corrected based on the sensor installation location. The selection range is 30-80Hz when mounted on the shaft end, 32-82Hz when mounted on the bearing housing, and 31-81Hz when mounted on the traction machine base. Further selection is made for components that simultaneously satisfy the five-dimensional selection criteria: the dominant frequency is within the effective intrinsic mode function component selection range; the number of poles P and the number of zero crossings Z satisfy... The mean of the upper and lower envelopes fitted by cubic spline interpolation is less than or equal to 1 / 20 of the maximum amplitude of the intrinsic mode function components; the cross-correlation coefficient with the target frequency band. The instantaneous phase fluctuation value is less than or equal to 0.1 rad. The eigenmode function components and residual components obtained from complementary set empirical mode decomposition are as follows: Figure 6As shown, IMF1 to IMF9 represent nine intrinsic mode function components decomposed sequentially from high to low frequency, and Resi represents the residual components. It can be seen that the frequency boundaries of IMF1 to IMF9 are clear, with no overlap. Mode aliasing, a common feature of traditional empirical mode decomposition, is effectively suppressed. The smoothing trend of Resi complements the fluctuation characteristics of the intrinsic mode function components, together covering the full frequency range of the filtered signal.

[0135] Further, step S3 is performed for each effective intrinsic mode function component. IMF h ( n Perform Hilbert transform to construct an analytic signal and extract the phase. Based on phase The instantaneous frequency is calculated, and then the center frequency of each effective eigenmode function component is obtained through discrete integration. (Calculation error controlled within 0.3Hz). Based on center frequency. For each effective intrinsic mode function component Perform time-frequency-phase joint Gaussian weighting, first calculate the energy integral value W1 of the target frequency band (30-60Hz), the energy integral value W2 of the transition frequency band (60-70Hz), the energy integral value W3 of the high-frequency noise band (70-80Hz), and the energy integral value W4 of the low-frequency interference band (20-30Hz). Then... Calculate the normalization coefficient Then substitute into the joint weight function The weights of each effective intrinsic mode function component are obtained. Weight anomaly correction is performed using sym8 wavelet basis multi-scale decomposition, and wavelet thresholding is applied. ,in, The standard deviation of noise is an estimate of noise intensity. This is the wavelet threshold adjustment coefficient. When the original signal-to-noise ratio... At that time, wavelet threshold adjustment coefficient When the original signal-to-noise ratio At that time, wavelet threshold adjustment coefficient .

[0136] Furthermore, the signal-to-noise ratio of each effective intrinsic mode function component is calculated. Set dynamic threshold If the signal-to-noise ratio of the effective intrinsic mode function components Then the time-frequency-phase joint weighting function is forcibly set. The time-frequency joint weighting function curve is as follows: Figure 7As shown, the blue area represents the target frequency band. It can be seen that the joint weighting function exhibits a clear peak range in the 30-60Hz range, with a weight value close to 1.0, which is the region with the highest weight in the entire frequency band. The weighting curve in the target frequency band does not fluctuate drastically and has a smooth Gaussian distribution characteristic, indicating that the weighting of the effective intrinsic mode function components at different center frequencies in this frequency band is balanced and stable, which can avoid the distortion of target features due to local weight abrupt changes. At the frequency band boundary of 30Hz and 60Hz, the weighting curve gradually decays rather than abruptly changes, which not only ensures the full preservation of the effective components in the target frequency band, but also smoothly connects to the transition frequency band.

[0137] Further, in step S4, the fault development stage is determined by the energy growth trend of the first three sets of signals in the 30-60Hz frequency band: growth ≤5% indicates the initial stage of microcracks, growth 5%-10% indicates the middle stage of microcracks, and growth >10% indicates the late stage of microcracks. This is then combined with the signal-to-noise ratio of the original signal. Dynamically set the weights of the residual components When the original signal-to-noise ratio At that time, the weight of the residual component in the early stage of microcracks Weight of residual components in the middle stage of microcracks Weight of residual components in the later stage of microcracks When the original signal-to-noise ratio At that time, the weight of the residual component in the early stage of microcracks Weight of residual components in the middle stage of microcracks Weight of residual components in the later stage of microcracks The signal is reconstructed using the discrete reconstruction formula to obtain the reconstructed signal. Reconstructed signal The verification standard must be met that the amplitude in the target frequency band (30-60Hz) is no less than 88% of the amplitude of the original signal in that frequency band, and the amplitude in the non-target frequency band does not exceed 0.035V. (Reconstructed signal) The time-domain waveform is as follows Figure 8 As shown, the reconstructed signal has a smoother time-domain waveform, free from the chaotic fluctuations of high-frequency noise, and retains the periodic fluctuation trend related to the 30-60Hz fault characteristics, demonstrating the advantages of the time-frequency-phase joint weight allocation.

[0138] Further, step S5 is performed to calculate the reconstructed signal. Local mean Local standard deviation Instantaneous amplitude change rate , cliff and local energy concentration Local features. The half-length of the moving window is adaptively adjusted based on triple features: when kurtosis... and instantaneous amplitude change rate And local energy concentration When setting window half length When the ravine or instantaneous amplitude change rate or local energy concentration When setting window half length In other cases, set the window half-length. .

[0139] Further, calculate the denoised signal. For noise reduction signals Perform time-domain and frequency-domain verification. In the time domain, when the original signal-to-noise ratio... At that time, the instantaneous amplitude change rate of adjacent sampling points When the original signal-to-noise ratio At that time, the instantaneous amplitude change rate of adjacent sampling points In the frequency domain, the amplitude fluctuation of the target frequency band of 30-60Hz does not exceed [a certain value]. The amplitude in non-target frequency bands should not exceed 0.025V. Noise reduction signal. The time-domain waveform is as follows Figure 9 As shown, the time-domain waveform of the noise-cancelled signal is more regular, without any messy high-frequency fluctuations, and only retains the smooth periodic fluctuations related to the 30-60Hz fault characteristics.

[0140] Further, step S6 is performed to denoise the signal. Perform single-test core indicator verification and time-series trend consistency verification, requiring denoised signals. The energy loss rate is ≤2.5%; the signal-to-noise ratio improvement must meet the following requirements. The power spectral density must meet the following requirements: a clear peak in the 30-60Hz frequency band; power spectral density verification in the 70-80Hz frequency band must not exceed 1 / 15 of the peak value in the 30-60Hz frequency band; and power spectral density verification in the 20-30Hz frequency band must not exceed 1 / 8 of the peak value in the 30-60Hz frequency band. For time-series trend consistency verification, five consecutive sets of denoised signals must be collected to ensure that the energy in the 30-60Hz frequency band shows an increasing trend with an increase of ≥2% between adjacent sets, and the goodness of fit R of the energy trend line is [value missing]. 2 ≥0.95.

[0141] Furthermore, adaptation adjustments are made based on the sensor installation location: when installed at the shaft end, the power spectral density peak verification standard remains unchanged, the energy loss rate tolerance threshold is ≤2.5%, and the signal attenuation coefficient is 0.95; when installed in the bearing housing, the power spectral density peak verification standard is reduced by 15%, the energy loss rate tolerance threshold is ≤4%, and the signal attenuation coefficient is 0.85; when installed in the traction machine base, the power spectral density peak verification standard is reduced by 20%, the energy loss rate tolerance threshold is ≤6%, and the signal attenuation coefficient is 0.75. A comparison of the target frequency band power spectral density of the original signal, the signal denoised by the traditional method, and the signal denoised by the method of this application is provided. Figure 10 As shown, the energy trends of five consecutive signal groups in the 30-60Hz frequency band are as follows: Figure 11 As shown, if the noise-cancelled signal simultaneously meets all the above verification and adaptation requirements, it is determined that it supports early fault detection of elevator shaft breakage. Figure 10 It can be seen that the power spectral density of the denoised signal in the target frequency band of this application is much higher than that of the original signal and the denoised signal of the traditional method, clearly showing the energy concentration characteristics of the fault features. The power spectral density in the non-target frequency band is qualitatively improved compared with the original signal and the traditional method, and the target energy loss is minimal. Figure 11 It can be seen that the energy of the 30-60Hz frequency band of the five consecutive noise-canceling signals gradually increases, with the energy increase of each adjacent signal being ≥2%, without fluctuations or decreases. The energy growth pattern is significant and free from chaotic fluctuations. The method of this application satisfies both the single-test core indicator verification of low energy loss rate, thorough noise suppression, and prominent target characteristics, and the temporal consistency verification of increasing energy and stable trend.

[0142] In one instance, a multi-dimensional evaluation system was used to verify the effectiveness of the proposed method. The energy loss rate (%), signal-to-noise ratio improvement (dB), and fault feature amplitude retention rate (%) of the proposed method were compared with those of the traditional empirical mode decomposition method. Table 1 shows the noise reduction evaluation table obtained.

[0143] Table 1 Noise Reduction Evaluation Table As shown in Table 1, the energy loss rate of the method proposed in this application is 2.3%, while the energy loss rate of traditional EMD decomposition is 15.2%. The signal-to-noise ratio of the method proposed in this application is improved by 16.0 dB and the fault characteristic amplitude retention rate is 95.5%, which is a significant improvement over the traditional empirical mode decomposition method.

[0144] In this signal denoising method, a dual adaptive parameter adjustment mechanism based on operating conditions and location is employed to precisely match core parameters such as filtering and decomposition with the elevator's operating status and sensor installation location. This effectively solves the problem of poor adaptability of fixed parameters, reducing the target signal energy loss rate to below 2.5%, significantly lower than the 15% of traditional methods. This avoids incomplete noise suppression or target signal loss under different operating conditions and installation locations. Furthermore, by fusing a five-dimensional screening strategy based on frequency, number of poles and zero-crossings, envelope mean, target frequency band correlation, and phase stability, along with phase consistency weight allocation, the method achieves refined screening of effective components. This improves the identification accuracy of 30-60Hz characteristic signals by more than 35%, significantly outperforming existing single-dimensional screening methods and successfully eliminating phase disturbances. The method effectively eliminates chaotic interference components. By constructing a triple-feature adaptive moving smoothing filter mechanism based on kurtosis, amplitude change rate, and energy concentration, and dynamically adjusting the smoothing window, it enhances residual noise suppression while preserving weak impact characteristics. This results in a signal-to-noise ratio improvement of ≥15dB after denoising, significantly improving upon the less than 5dB improvement of traditional EMD decomposition. The timing misjudgment rate is reduced to below 1.2%, meeting the accuracy requirements for early fault warning. Through adaptive reconstruction of fault stages and consistency verification of timing trends, combined with the energy growth trend and goodness-of-fit test of five consecutive sets of signals, the risk of misjudgment is further reduced. At the same time, differentiated adaptation strategies are designed for the installation positions of three types of sensors, significantly improving the versatility of the method and making it widely applicable to early fault detection of elevator shaft breakage in different installation scenarios.

[0145] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0146] In another embodiment, this application also provides an elevator, including a body, a sensor, and a computing unit. The sensor collects the raw signals of the body and outputs them to the computing unit. The computing unit executes the elevator shaft breakage early fault detection signal noise reduction method described in the above embodiment.

[0147] In another embodiment, this application also provides a computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program, when executed by a processor, implements the various steps of the signal noise reduction method provided in the above embodiments.

[0148] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0149] Although this application has been disclosed above with reference to embodiments, it is not intended to limit this application. Anyone skilled in the art can make some modifications and refinements without departing from the spirit and scope of this application.

Claims

1. A signal denoising method, characterized in that, Includes the following steps: The original signal of the elevator shaft is acquired, the spectrum of the original signal is analyzed and the original signal-to-noise ratio is calculated. The elevator operating condition is identified based on the original signal, and the operating condition signal is obtained by combining the sensor installation position. The bandpass filter coefficient is adjusted based on the operating condition signal and filtered to obtain the filtered signal. The method of adjusting the bandpass filter coefficients based on operating condition signals includes: determining the upper limit of low-frequency interference and the lower limit of high-frequency noise using the energy ratio method; determining the core distribution range of noise based on the upper limit of low-frequency interference and the lower limit of high-frequency noise; and determining the bandpass filter coefficients based on the core distribution range of noise, elevator operating conditions, and sensor installation locations. The bandpass filter coefficients include a low cutoff frequency, a high cutoff frequency, and a filter order. The expression for the high cutoff frequency is: in, This is the lower limit of high-frequency noise. This is the position correction amount; Based on the original signal-to-noise ratio, the decomposition parameters of the complementary set empirical mode decomposition are dynamically set, and the complementary set empirical mode decomposition is performed on the filtered signal to obtain multiple intrinsic mode function components and a residual component. The effective intrinsic mode function components are then selected from all the intrinsic mode function components. The Hilbert transform is performed on each effective intrinsic mode function component to obtain the analytic signal. The instantaneous phase and instantaneous frequency are obtained based on the analytic signal. The center frequency of each effective intrinsic mode function component is obtained through discrete integration. The weights of each effective intrinsic mode function component are assigned based on the center frequencies. After the assignment, weight anomaly correction is performed. The weights of the residual components are determined based on the fault development stage and the original signal-to-noise ratio. The signals are then reconstructed by reconstructing all effective intrinsic mode function components and the residual components to obtain the reconstructed signal. Local feature calculations are performed on the reconstructed signal. Based on the calculation results, the half-length of the moving smoothing window is adaptively adjusted. An adaptive moving smoothing filtering method is used to smooth the residual noise in the reconstructed signal to obtain a denoised signal.

2. The signal noise reduction method according to claim 1, characterized in that, Complementary ensemble empirical mode decomposition is performed on the filtered signal. The decomposition parameters are dynamically adjusted based on the original signal-to-noise ratio to obtain multiple intrinsic mode function (IMF) components and one residual component. Valid IMF components are then selected from all IMF components, including: Set the decomposition parameters of the complementary set empirical mode decomposition based on the original signal-to-noise ratio; Complementary set empirical mode decomposition is performed on the filtered signal to obtain multiple intrinsic mode function components and a residual component. Based on the selection range, components that simultaneously meet the five-dimensional selection criteria are selected from all intrinsic mode function components to obtain effective intrinsic mode function components.

3. The signal noise reduction method according to claim 2, characterized in that, The five-dimensional screening conditions include: the main frequency is within the effective intrinsic mode function component screening range; the number of poles and the number of zero crossings meet preset conditions; the mean of the upper and lower envelopes fitted by cubic spline interpolation is less than or equal to a preset proportion threshold of the maximum amplitude of the intrinsic mode function component; the cross-correlation coefficient with the target frequency band is greater than or equal to a preset correlation threshold; and the instantaneous phase fluctuation value is less than or equal to a preset phase stability threshold.

4. The signal noise reduction method according to claim 1, characterized in that, Local feature calculations are performed on the reconstructed signal. Based on the calculation results, the half-length of the moving smoothing window is adaptively adjusted. An adaptive moving smoothing filtering method is used to smooth the residual noise in the reconstructed signal, resulting in a denoised signal, including: Calculate the local features of the reconstructed signal, including local mean, local standard deviation, instantaneous amplitude change rate, kurtosis, and local energy concentration. The half-length of the moving window is adaptively adjusted based on the triple features of kurtosis, instantaneous amplitude change rate, and local energy concentration. When the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, and the local energy concentration is greater than the local energy concentration threshold, the half-length of the window is set to N1; when the kurtosis is greater than the kurtosis threshold, the instantaneous amplitude change rate is greater than the instantaneous amplitude change rate threshold, or the local energy concentration is greater than the local energy concentration threshold, the half-length of the window is set to N2; in other cases, the half-length of the window is set to N3, where N1, N2, and N3 are constants. Based on the adjusted window half-length, the reconstructed signal is smoothed by motion to obtain a denoised signal.

5. The signal noise reduction method according to claim 4, characterized in that, The reconstructed signal is subjected to local feature calculation. Based on the calculation results, the half-length of the moving smoothing window is adaptively adjusted. The residual noise in the reconstructed signal is smoothed by an adaptive moving smoothing filtering method to obtain a denoised signal. The denoised signal is also verified from both the time domain and the frequency domain.

6. The signal noise reduction method according to claim 1, characterized in that, It also includes verifying the core indicators and the consistency of the time series trend of the noise-reduced signal in a single test, and making adaptation corrections in combination with the operating condition signal.

7. The signal noise reduction method according to claim 6, characterized in that, The single core indicator verification includes energy loss rate verification, signal-to-noise ratio improvement verification, and power spectral density verification. The temporal trend consistency verification includes: determining whether the energy of the target frequency band of multiple consecutive noise-canceling signals shows an increasing trend, and whether the energy growth of two adjacent signal groups is greater than or equal to the third threshold. At the same time, the least squares method is used to fit the energy trend line, and the goodness of fit of the energy trend line is greater than or equal to the fourth threshold.

8. The signal noise reduction method according to claim 1, characterized in that, The operating condition signals include the sensor installation location and the elevator operating conditions.

9. An elevator, characterized in that, The device includes a body, a sensor, and a computing unit. The sensor acquires the raw signal of the body and outputs it to the computing unit. The computing unit executes the signal noise reduction method as described in any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the signal noise reduction method according to any one of claims 1 to 8.

Images