Elevator polyurethane buffer anomaly detection method based on millimeter wave radar

Abstract

This invention relates to the field of elevator technology, and more particularly to an anomaly detection method for elevator polyurethane buffers based on millimeter-wave radar. The method includes: acquiring echo signals within a preset time window after the polyurethane buffer's compression and rebound by millimeter-wave radar; determining the target distance unit and generating a vibration displacement signal in the radar's line-of-sight direction; identifying intervals where the overall vibration envelope shows a decaying trend and extracting the damping response signal; implementing bistable stochastic resonance enhancement and variational mode decomposition to identify the target modal signal; calculating damping characteristic indices using the target envelope amplitude and stiffness characteristic indices using the target phase to construct a health feature vector; and determining whether the obtained health feature vector lies within the constructed health reference space. Using non-contact millimeter-wave radar to acquire echo signals during the polyurethane buffer's rebound process can accurately capture phase changes caused by minute movements or vibrations of the target, resulting in high detection efficiency and high safety.

Description

Technical Field

[0001] This invention relates to the field of elevator technology, and in particular to an abnormal detection method for elevator polyurethane buffers based on millimeter-wave radar. Background Technology

[0002] Elevator buffers are used to absorb impact energy through deformation and damping when an elevator unexpectedly bottoms out or overtravels, protecting passengers and equipment. Currently, widely used polyurethane buffers offer advantages such as light weight, easy installation, and maintenance-free operation. However, their performance gradually deteriorates with use and environmental factors, such as material aging, elasticity loss, or internal damage. If buffer performance deteriorates, it may fail to effectively buffer in emergencies, posing a safety hazard. Therefore, regular performance testing of elevator polyurethane buffers is crucial.

[0003] Current buffer inspection methods rely primarily on manual labor and simple instruments, which have significant limitations. Conventional methods typically include two approaches: First, inspectors visually assess buffer deformation at the elevator pit and estimate recovery time using a stopwatch. This purely manual visual inspection method is highly susceptible to subjective factors and has low measurement accuracy. Second, it employs sensors or rangefinders for auxiliary inspection, such as installing pressure sensors on the buffer to measure pressure, or using ultrasonic / laser rangefinders to monitor buffer displacement and time. However, these methods often present engineering implementation difficulties and blind spots: for example, adding pressure sensors complicates the system structure and can only monitor a single buffer state; a displacement measuring device is also needed to determine the buffer reset time, as illustrated by the elevator buffer reset time detection device disclosed in utility model patent CN216038014U. Ultrasonic rangefinders are cumbersome to install, and the ultrasonic beam is easily blocked by obstacles in the pit, making measurement impossible. Laser rangefinders have high environmental requirements, requiring a reflector plate on the top of the buffer and a special clamp for fixation; different buffer models require customized clamps, making installation and debugging complex. It is evident that current methods for inspecting polyurethane buffers either rely on human experience, which is highly subjective and has blind spots, or require additional equipment, making operation and maintenance difficult and posing potential safety risks (such as requiring inspectors to operate at close range in the elevator pit, posing a risk of being crushed by the elevator car). These factors result in low efficiency and insufficient reliability in routine buffer inspections, making it difficult to detect subtle anomalies inside the buffer in a timely manner. Summary of the Invention

[0004] The purpose of this invention is to propose an abnormal detection method for elevator polyurethane buffers based on millimeter-wave radar. This method uses non-contact millimeter-wave radar to acquire the echo signal during the rebound process of the polyurethane buffer. It has high measurement accuracy, can accurately capture phase changes caused by minute movements or vibrations of the target, and has high detection efficiency and high safety.

[0005] To achieve this objective, the present invention adopts the following technical solution:

[0006] A method for detecting anomalies in elevator polyurethane buffers based on millimeter-wave radar includes the following steps:

[0007] S1: Aim the main beam of the millimeter-wave radar at the side wall of the elevator polyurethane buffer and collect the echo signal within a preset time window after the polyurethane buffer has finished compressing and rebounding.

[0008] S2: Determine the target range unit through the echo signal and extract the corresponding phase change information, and generate the vibration displacement signal in the radar line-of-sight direction according to the phase-displacement mapping relationship;

[0009] S3: Identify the intervals in the vibration displacement signal where the overall vibration envelope shows a decaying trend, and extract the vibration displacement signal corresponding to the interval as the damping response signal;

[0010] S4: After preprocessing the damping response signal, bistable stochastic resonance enhancement is performed. The enhanced damping response signal is then subjected to variational mode decomposition to obtain multiple mode signals. The envelope entropy of each mode signal is calculated, and the mode signal with the smallest envelope entropy is selected as the target mode signal.

[0011] S5: After analyzing the target modal signal, extract the target envelope amplitude information and the target phase information. Calculate the damping characteristic index with the target envelope amplitude and the stiffness characteristic index with the target phase. Construct a health feature vector through the damping characteristic index and the stiffness characteristic index.

[0012] S6: Construct a health benchmark space based on the health benchmark samples, and determine whether the health feature vector obtained in step S5 is located in the health benchmark space. If it is, it is judged to be normal; otherwise, it is judged to be abnormal.

[0013] Furthermore, step S2 includes:

[0014] S21: Perform a range-direction fast Fourier transform on the discrete sampled data of the echo signal to obtain complex echo data distributed along the range direction, and then divide the complex echo data into multiple range cells according to the range resolution of the millimeter-wave radar.

[0015] S22: Within a preset distance range, calculate the modulus square of the complex echo data of the distance unit at continuous sampling time, accumulate or average the modulus square data to obtain the echo energy of each distance unit, and select the distance unit with the largest echo energy as the target distance unit corresponding to the polyurethane buffer.

[0016] S23: Extract the complex echo data of the target distance unit at continuous sampling time, calculate the instantaneous phase at the sampling time of the complex echo data, and obtain a continuous phase sequence through phase unwinding processing;

[0017] The continuous phase of the target distance unit at the first sampling time within the preset time window As a reference phase, for any sampling time continuous phase By performing differential processing, the phase change is obtained: ;

[0018] S24: Based on the phase-displacement mapping relationship, the phase change is converted into a vibration displacement signal. The formula for the phase-displacement mapping relationship is as follows:

[0019] ,

[0020] in, Represents the sampling time The output vibration displacement signal reflects the real-time minute vibration displacement of the polyurethane buffer along the line of sight of the millimeter-wave radar. For millimeter wave wavelengths; It is a time variable.

[0021] Furthermore, step S3 includes:

[0022] S31: Perform Hilbert transform on the vibration displacement signal to obtain an analytical signal, and calculate the vibration envelope amplitude sequence by taking the modulus of the values ​​of the analytical signal at each time point in the continuous sampling time.

[0023] S32: Perform natural logarithmic operation on the vibration envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence; perform least squares linear fitting on the logarithmic envelope amplitude sequence within a sliding time window to obtain the fitting slope and goodness of fit; determine the sliding time window where the fitting slope, goodness of fit, and vibration envelope amplitude meet the preset conditions as a candidate interval with a local decay trend; merge temporally adjacent candidate intervals to obtain an interval where the overall vibration envelope has a decay trend;

[0024] S33: Extract the vibration displacement signal corresponding to the interval where the overall vibration envelope shows a decaying trend from the vibration displacement signal, and use it as the damping response signal.

[0025] Furthermore, in step S32, the preset conditions for the fitting slope, goodness of fit, and vibration envelope amplitude are as follows:

[0026] Within the sliding window, the fitting slope is less than zero, the goodness of fit is not less than the preset threshold, and the average value of the vibration envelope amplitude is higher than the preset noise threshold.

[0027] Furthermore, step S4 includes:

[0028] S41: Perform frequency shift scaling on the damping response signal, and then perform mean removal and amplitude normalization on the frequency shift scaling signal to obtain a preprocessed signal;

[0029] S42: The preprocessed signal is used as an external driving input to the bistable stochastic resonance system. In the bistable stochastic resonance system, random noise is superimposed on the preprocessed signal to drive the state of the bistable stochastic resonance system to transition between the two potential wells, and the enhanced damped response signal is output.

[0030] S43: Perform variational mode decomposition on the enhanced damping response signal: Set the constraint that the decomposed modal signals can reconstruct the enhanced damping response signal, and take the center frequency corresponding to each modal signal as the parameter to be estimated. Under the premise of satisfying the constraint, construct a variational optimization problem that minimizes the bandwidth of each modal signal near the center frequency; use the augmented Lagrange method to process the variational optimization problem, and perform modal signal update, center frequency update and Lagrange multiplier update in an alternating iterative manner until the change of the modal signal in two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, and output all decomposed modal signals;

[0031] S44: Perform Hilbert transform on each modal signal to obtain modal analytical signals; calculate the modulus of each modal analytical signal at each time step in continuous sampling to obtain the envelope amplitude sequence of each modal signal; normalize each envelope amplitude sequence to obtain the probability distribution of the envelope energy of each modal signal along the time axis; calculate the envelope entropy of each modal signal according to the probability distribution using the information entropy calculation method; select the modal signal with the smallest envelope entropy as the target modal signal.

[0032] Furthermore, the method for outputting the enhanced damped response signal of the bistable stochastic resonance system in S42 includes:

[0033] The potential function of the bistable stochastic resonance system is:

[0034] ,

[0035] in, For the state variables of a bistable stochastic resonance system; and For the bistable state trap parameters, and and ;

[0036] The dynamic equations of the bistable stochastic resonance system are as follows:

[0037] ,

[0038] in, The state output signal of the bistable stochastic resonance system serves as the enhanced damped response signal. The noise intensity coefficient, Zero-mean, unit-intensity Gaussian white noise For preprocessed signals;

[0039] The enhanced damping response signal is obtained by iteratively solving the dynamic equations point by point using the Euler-Maruyama numerical discretization method. .

[0040] Furthermore, S43 includes:

[0041] S431: The constraint condition is set as the ability of the decomposed modal signals to reconstruct the enhanced damped response signal. The functional expression of the constraint condition is as follows:

[0042] ,

[0043] in, Indicates the first One modal signal; For modal index, ; To decompose the total number of modes;

[0044] S432: The center frequency corresponding to each modal signal As parameters to be estimated, under the premise of satisfying the aforementioned constraints, a mechanism is constructed that ensures each modal signal is at its corresponding center frequency. The variational optimization problem with the minimum nearby bandwidth is addressed by using the augmented Lagrangian method.

[0045] The augmented Lagrange function is expressed as:

[0046] ,

[0047] in, Represents the set of all modal signals; Represents the set of center frequencies of all modal signals. For Lagrange multipliers, This represents the time partial derivative operator. Represents the Dirac function, Represents the imaginary unit. This represents the convolution operation. It is a complex exponential modulation factor. This represents the L2 norm square operator. This represents the inner product operation. These are the modal narrowband constraint weight parameters;

[0048] S433: Within the augmented Lagrange framework, modal signal updates, center frequency updates, and Lagrange multiplier updates are performed sequentially using an alternating iterative approach.

[0049] Under the condition of fixed center frequency and Lagrange multipliers, each modal signal is updated. Based on the updated modal signal, the corresponding center frequency is updated according to the spectral energy distribution. And based on the current modal signal and the enhanced damped response signal... The reconstruction error between them updates the Lagrange multipliers;

[0050] S434: When the change in modal signal between two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, the variational mode decomposition is completed and all modal signals are output.

[0051] Furthermore, step S5 includes:

[0052] S51: Perform Hilbert transform on the target modal signal to obtain the target modal analytical signal; calculate the modulus of the target modal analytical signal at each time step in the continuous sampling time to obtain the target envelope amplitude sequence; calculate the argument of the target modal analytical signal at each time step in the continuous sampling time to obtain the target phase sequence.

[0053] S52: Perform a natural logarithmic operation on the target envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence; based on the decay trend of the logarithmic envelope amplitude sequence, select a preset linear fitting interval corresponding to the amplitude decaying from the initial peak to a preset proportion range, and perform linear regression using the least squares method within the linear fitting interval to obtain the regression slope; take the absolute value of the regression slope to obtain the damping characteristic index. ;

[0054] S53: The target angular frequency is defined as the first derivative of the target phase with respect to time. A discrete target angular frequency sequence is obtained by calculating the phase difference between adjacent sampling times of the target phase sequence. The discrete target angular frequency sequence is then statistically averaged over the time interval corresponding to the linear fitting interval to obtain an estimated principal frequency. The estimated principal frequency is then squared to obtain the stiffness characteristic index. ;

[0055] S54: Through the aforementioned damping characteristic index and stiffness characteristic index Constructing a health feature vector:

[0056] ,

[0057] in, This indicates the transpose operation.

[0058] Furthermore, step S6 includes:

[0059] S61: Collect multiple sets of health benchmark samples, and obtain health feature vectors of multiple sets of health benchmark samples by calculating damping characteristic index and stiffness characteristic index; calculate the average of each health feature vector according to the feature dimension to obtain the health sample benchmark mean; subtract each health feature vector from the health sample benchmark mean vector to obtain the corresponding deviation vector, and calculate the average of the outer product of each deviation vector and its transpose vector to obtain the health benchmark covariance matrix; perform regularization processing on the health benchmark covariance matrix to obtain the invertible covariance matrix; construct the health benchmark space using the binary tuple composed of the health benchmark mean vector and the invertible covariance matrix.

[0060] S62: Calculate the squared Mahalanobis distance of the health feature vector obtained in step S5 relative to the health reference space, and use the squared Mahalanobis distance as the anomaly score;

[0061] S63: Compare the abnormal score with a preset abnormal judgment threshold. When the abnormal score is not greater than the abnormal judgment threshold, determine that the current health feature vector is within the health reference space and output a normal result; when the abnormal score is greater than the abnormal judgment threshold, determine that the current health feature vector is outside the health reference space and output an abnormal result.

[0062] Furthermore, in step S61:

[0063] Multiple sets of health baseline samples are denoted as , Indicates the first The health feature vector obtained from this test is used as the health baseline feature vector, and ; Index for health benchmark samples; Indicates the number of healthy baseline samples;

[0064] The health baseline center is defined as the sample mean, and the health baseline sample mean is:

[0065] ,

[0066] Define the health baseline covariance matrix as follows:

[0067] ,

[0068] Among them, the health baseline covariance matrix It is a symmetric positive semi-definite matrix; Indicates the transpose operation;

[0069] Invertible covariance matrix:

[0070] ,

[0071] in, This is the invertible covariance matrix after regularization; The inverse matrix can be represented as , The regularization coefficient is . for An identity matrix, where 2 represents the number of health benchmark characteristic indicators, and satisfies... ;

[0072] The health baseline space is:

[0073] .

[0074] The technical solution provided by this invention may include the following beneficial effects:

[0075] This invention utilizes millimeter-wave radar to non-contactly acquire the minute vibrations of an elevator polyurethane buffer after compression and rebound. It extracts the vibration displacement signal of the buffer along the radar line of sight by analyzing echo phase changes, and identifies intervals where the overall vibration envelope exhibits a decaying trend as the damping response signal. Based on this, by combining stochastic resonance enhancement, variational mode decomposition, and minimum envelope entropy mode selection, the target mode signal corresponding to the buffer's free damping decay process can be stably separated from the background noise, thereby improving the targeting and effectiveness of subsequent feature extraction.

[0076] This invention extracts damping characteristic indices representing energy dissipation capacity and stiffness characteristic indices representing mechanical response and structural state changes from the target modal signals, respectively, and constructs a health feature vector. Combined with a health reference space and Mahalanobis distance, it identifies abnormal states of the buffer. Because this feature system is highly compatible in physical mechanism with the displacement time-series signals acquired by millimeter-wave radar, it eliminates the need for contact sensors, complex structural parameters, or direct inversion of absolute stiffness. It can quantitatively characterize changes in the key safety performance of polyurethane buffers, offering advantages such as non-contact operation, clear physical meaning, suitability for field applications, and good stability in anomaly detection. Correspondingly, the non-contact millimeter-wave radar data sampling eliminates the need for inspection personnel to operate at close range in the elevator pit for extended periods, significantly improving inspection safety. Attached Figure Description

[0077] Figure 1 This is a schematic diagram showing the positional relationship between the millimeter-wave radar and the polyurethane buffer;

[0078] Figure 2 A flowchart of an abnormal detection method for elevator polyurethane buffer based on millimeter-wave radar according to one embodiment of the present invention.

[0079] Figure 3 This is a schematic diagram illustrating the principle of determining the target distance unit through echo signals in steps S21-S22;

[0080] Figure 4 This is a logarithmic attenuation rate analysis diagram of the vibration envelope amplitude and logarithmic envelope amplitude calculated from the vibration displacement signal in step S3;

[0081] Figure 5 This is a flowchart illustrating the process of inputting the preprocessed signal into the bistable stochastic resonance system in step S42, processing the signal, and outputting the enhanced damped response signal.

[0082] Figure 6 This is a flowchart illustrating the process of obtaining the target modal signal by calculating the envelope entropy after modal decomposition of the damped response signal in steps S43-S44.

[0083] Figure 7 This is a flowchart illustrating the process of constructing a health feature vector from the target modal signal in step S4.

[0084] Figure 8 This is a schematic diagram showing the positional relationship between the health baseline space, health baseline samples, and abnormal score samples in step S6 using Mahalanobis distance. Detailed Implementation

[0085] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the present invention.

[0086] As a viscoelastic buffer element, the compression test of a polyurethane buffer typically includes a vertical compression deformation stage, a rapid rebound stage, and a horizontal free decay vibration stage after the compression rebound ends. This free decay vibration stage refers to the process by which the polyurethane buffer, after the vertical compression load is released and the rebound is complete, generates minute reciprocating vibrations in the horizontal direction under its own equivalent stiffness recovery, and these vibrations gradually decay over time under its own damping. Since the vibration response in the free decay vibration stage is related to the damping characteristics, equivalent stiffness, and structural integrity of the polyurethane buffer, it can reflect its energy dissipation capacity, elastic recovery capacity, and overall health status. Therefore, this invention uses millimeter-wave radar to non-contactly collect the minute vibrations of the elevator polyurethane buffer after the compression rebound ends.

[0087] Reference Figure 1 and Figure 2 An embodiment of the present invention provides an abnormal detection method for elevator polyurethane buffers based on millimeter-wave radar, characterized by comprising the following steps:

[0088] S1: Aim the main beam of the millimeter-wave radar at the side wall of the elevator polyurethane buffer and collect the echo signal within a preset time window after the polyurethane buffer has finished compressing and rebounding.

[0089] S2: Determine the target range unit through the echo signal and extract the corresponding phase change information, and generate the vibration displacement signal in the radar line-of-sight direction according to the phase-displacement mapping relationship;

[0090] S3: Identify the intervals in the vibration displacement signal where the overall vibration envelope shows a decaying trend, and extract the vibration displacement signal corresponding to the interval as the damping response signal;

[0091] S4: After preprocessing the damping response signal, bistable stochastic resonance enhancement is performed. The enhanced damping response signal is then subjected to variational mode decomposition to obtain multiple mode signals. The envelope entropy of each mode signal is calculated, and the mode signal with the smallest envelope entropy is selected as the target mode signal.

[0092] S5: After analyzing the target modal signal, extract the target envelope amplitude information and the target phase information. Calculate the damping characteristic index with the target envelope amplitude and the stiffness characteristic index with the target phase. Construct a health feature vector through the damping characteristic index and the stiffness characteristic index.

[0093] S6: Construct a health benchmark space based on the health benchmark samples, and determine whether the health feature vector obtained in step S5 is located in the health benchmark space. If it is, it is judged to be normal; otherwise, it is judged to be abnormal.

[0094] When testing polyurethane buffers using the method of this invention, a millimeter-wave radar with a bracket is simply placed in the elevator pit with the radar facing the side of the polyurethane buffer. The operation is simple and quick, eliminating the need for testing personnel to enter the pit or operate at close range for extended periods, thus greatly improving testing safety. As an example, such as... Figure 1 As shown, the millimeter-wave radar is arranged on the side of the polyurethane buffer and positioned in the pit by a bracket, so that the main beam of the radar is aligned with the upper middle part of the side wall of the polyurethane buffer. The horizontal spacing between the millimeter-wave radar and the side wall of the polyurethane buffer is preferably set to 0.4m to 0.6m to take into account the echo signal strength, measurement stability and the on-site layout space requirements of the pit.

[0095] In step S1, the echo signal of the polyurethane buffer during the free damped vibration stage after it is removed from the external load is acquired. That is, after the polyurethane buffer completes its compression rebound and enters the free decay vibration stage, the millimeter-wave radar is triggered by an external control signal to start acquisition. Preferably, the echo signal is acquired within a preset time window of 2 seconds, starting 50ms after the external load is removed, and is used as the processing object. The millimeter-wave radar of the present invention is preferably a frequency-modulated continuous wave millimeter-wave radar, whose output includes complex echo data containing in-phase component I and quadrature component Q, which serves as the raw input for target range cell determination and vibration displacement signal generation.

[0096] Step S2 of this invention can accurately determine the target range cell corresponding to the polyurethane buffer from the millimeter-wave radar echo signal, extract the phase change information of the target range cell, and then convert it into a vibration displacement signal along the radar line of sight. This effectively suppresses environmental clutter interference, improves the accuracy and stability of vibration response acquisition, and provides a reliable data foundation for subsequent health feature extraction and anomaly identification. In an embodiment of this invention, step S2 includes:

[0097] S21: Perform a range-direction fast Fourier transform on the discrete sampled data of the echo signal to obtain complex echo data distributed along the range direction, and then divide the complex echo data into multiple range cells according to the range resolution of the millimeter-wave radar.

[0098] S22: Within a preset distance range, calculate the squared modulus of the complex echo data of the distance unit at continuous sampling times, accumulate or average the squared modulus data to obtain the echo energy of each distance unit, and select the distance unit with the largest echo energy as the target distance unit corresponding to the polyurethane buffer (refer to...). Figure 3 );

[0099] S23: Extract the complex echo data of the target distance unit at continuous sampling time, calculate the instantaneous phase at the sampling time of the complex echo data, and obtain a continuous phase sequence through phase unwinding processing;

[0100] The continuous phase of the target distance unit at the first sampling time within the preset time window As a reference phase, for any sampling time continuous phase By performing differential processing, the phase change is obtained: ;

[0101] S24: Based on the phase-displacement mapping relationship, the phase change is converted into a vibration displacement signal. The formula for the phase-displacement mapping relationship is as follows:

[0102] ,

[0103] in, Represents the sampling time The output vibration displacement signal reflects the real-time minute vibration displacement of the polyurethane buffer along the line of sight of the millimeter-wave radar. For millimeter wave wavelengths; It is a time variable.

[0104] In this scheme, after the millimeter-wave radar receives the echo signal reflected from the sidewall of the polyurethane buffer, it performs frequency mixing, low-pass filtering, and analog-to-digital conversion in the receiving channel to obtain discrete sampled data within a single transmission cycle. The discrete sampled data corresponding to a single transmission cycle is used as the processing object.

[0105] As an example, the millimeter-wave radar preferably operates at a frequency of 77 GHz with a modulation bandwidth of 4 GHz. Since the range resolution is determined by dividing the electromagnetic wave propagation speed by twice the modulation bandwidth, the range resolution at a modulation bandwidth of 4 GHz is approximately 0.0375 m. The horizontal spacing between the millimeter-wave radar and the sidewall of the polyurethane buffer is preferably set to 0.4 m to 0.6 m. Considering on-site installation errors, equivalent distance changes caused by vibration, and range resolution discrepancies, the target distance range where the polyurethane buffer is located is pre-set to 0.35 m to 0.65 m to ensure that the target echo signal stably falls within the preset distance range, improving the accuracy and robustness of target distance unit determination. When the range resolution is approximately 0.0375 m, the target distance range can correspond to the 10th to 17th distance units. As an example, if the echo energies calculated for the 10th to 17th distance cells are 12.4, 18.1, 24.7, 95.6, 22.3, 16.8, 13.5 and 10.9 respectively, then the 13th distance cell with the largest echo energy is selected as the target distance cell.

[0106] In step S23 of this embodiment, for the complex echo data at each sampling time, the in-phase component corresponding to its real part and the quadrature component corresponding to its imaginary part are read respectively, and the instantaneous phase corresponding to that sampling time is calculated using the arctangent angle method. The value range is limited to (-π, π], thereby obtaining a wrapped phase sequence arranged in chronological order. The wrapped phase difference between the current sampling time and the previous sampling time in the slow time dimension is calculated. When the wrapped phase difference is greater than +π, 2π is subtracted from the wrapped phase at the current sampling time; when the wrapped phase difference is less than -π, 2π is added to the wrapped phase at the current sampling time; otherwise, the wrapped phase at the current sampling time remains unchanged, completing the phase unwrapping and obtaining a continuous phase sequence.

[0107] Step S3 of this invention can effectively separate the effective response range corresponding to the free decay vibration stage of the polyurethane buffer from the vibration displacement signal generated in step S2, avoiding the introduction of transition signals during compression rebound or low signal-to-noise ratio noise signals in the later stage of decay into subsequent analysis, thereby improving the accuracy and stability of damping response extraction and providing reliable input for subsequent stochastic resonance enhancement, variational mode decomposition, and health feature extraction. In one embodiment of this invention, referring to… Figure 4 Step S3 includes:

[0108] S31: Regarding the vibration displacement signal The analytic signal is obtained by performing a Hilbert transform, and the analytic signal can be expressed as:

[0109] ,

[0110] in, Represents the imaginary unit; Represents the Hilbert transform operator;

[0111] The vibration envelope amplitude sequence is obtained by taking the modulus of the values ​​of the analytical signal at each time point in the continuous sampling time. The vibration envelope amplitude can be expressed as:

[0112] ;

[0113] S32: Perform natural logarithmic operation on the vibration envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence; perform least squares linear fitting on the logarithmic envelope amplitude sequence within a sliding time window to obtain a fitting straight line that best reflects the overall trend of data change within the time window, and use its slope as the fitting slope to characterize the direction of change of the vibration envelope with time; calculate the magnitude of the deviation between the fitting straight line and the actual data, and compare the deviation with the total deviation of the data itself to obtain the goodness of fit; determine the sliding time window where the fitting slope, goodness of fit, and vibration envelope amplitude meet the preset conditions as a candidate interval with a local decay trend, and merge the temporally adjacent candidate intervals to obtain an interval where the overall vibration envelope has a decay trend;

[0114] S33: Extract the vibration displacement signal corresponding to the interval where the overall vibration envelope shows a decaying trend from the vibration displacement signal, and use it as the damping response signal.

[0115] The method for obtaining the goodness of fit in step S32 is as follows: First, calculate the sum of squares of the differences between the logarithmic envelope amplitude of each sampling point within the sliding time window and the predicted value of the fitted line at the corresponding sampling time, as the total fitting error; then calculate the sum of squares of the differences between the logarithmic envelope amplitude of each sampling point and its average value, as the total sum of squares of deviations; finally, divide the total fitting error by the total sum of squares of deviations to obtain the proportion of fitting error to total deviations, and obtain the goodness of fit by subtracting this proportion from 1. The goodness of fit is preferably expressed using the coefficient of determination. The smaller the fitting error is relative to the total deviation, the closer the data points are to the fitted line, and the closer the goodness of fit is to 1; conversely, when the fitting error is large, the goodness of fit is close to 0.

[0116] Preferably, in step S32, the preset conditions for the fitting slope, goodness of fit, and vibration envelope amplitude are as follows:

[0117] Within the sliding window, the fitting slope is less than zero, the goodness of fit is not less than a preset threshold, and the average value of the vibration envelope amplitude is higher than a preset noise threshold. In this embodiment of the invention, the preset threshold is 0.85 to 0.95, preferably 0.90; the preset noise threshold is 0.05 to 0.15, preferably 0.10.

[0118] Understandably, when the fitting slope within a certain sliding time window is less than zero, it indicates that the overall vibration envelope within that time window shows a decreasing trend. When the goodness of fit is not less than a preset threshold, it indicates that the logarithmic envelope within that time window has a high consistency with the linear model, meaning that the corresponding vibration envelope approximately satisfies the exponential decay law. Simultaneously, when the amplitude of the vibration envelope within that time window is higher than a preset noise threshold, it indicates that the signal segment still has effective vibration components. Based on these criteria, the time intervals corresponding to the sliding time windows that meet the conditions are determined as candidate intervals showing a local decay trend. Subsequently, multiple temporally adjacent candidate intervals are merged to obtain intervals where the overall vibration envelope shows a decay trend.

[0119] As an example, in a set of detection data, the vibration envelope amplitudes within a certain sliding time window are 1.20, 0.98, 0.81, 0.67, and 0.54, respectively. Performing natural logarithmic calculations on these vibration envelope amplitudes point by point yields the corresponding logarithmic envelope amplitudes of 0.1823, -0.0202, -0.2107, -0.4005, and -0.6162, respectively. Then, using the sampling time corresponding to each logarithmic envelope amplitude as the independent variable and the logarithmic envelope value as the dependent variable, a linear fit is performed using the least squares method. The resulting fit has a negative slope and a goodness-of-fit of 0.95. If the preset goodness-of-fit threshold is 0.90 and the preset noise threshold is 0.10, then this sliding time window meets the attenuation trend determination condition, and its corresponding time interval can be identified as a candidate interval showing a local attenuation trend. As time progresses, when the vibration envelope amplitude within subsequent sliding time windows decreases to 0.09, 0.07, 0.10, 0.06, and 0.08, the fitting results become unstable because the average vibration envelope amplitude within these time windows is below the preset noise threshold of 0.10. Therefore, these intervals are no longer considered candidate intervals. Through the above processing, the intervals where the overall vibration envelope of the polyurethane buffer exhibits a decaying trend during the free decay vibration stage can be effectively identified, and the vibration displacement signal of these intervals can be extracted as the damping response signal.

[0120] This invention, through step S4, first performs frequency-shift scaling, mean-reduction, and amplitude normalization on the damped response signal to match the signal's frequency scale and amplitude range with the input requirements of the bistable stochastic resonance system, reducing the impact of DC components, amplitude differences, and scale mismatches on subsequent processing. Then, the preprocessed signal is input into the bistable stochastic resonance system, and appropriate random noise is superimposed. The noise-induced potential well transition mechanism enhances the weak attenuation characteristics, improving signal detectability and signal-to-noise ratio. Based on this, variational mode decomposition is used to decompose the enhanced damped response signal into multiple narrowband modes, reducing mode aliasing and broadband noise interference. Finally, combining Hilbert transform, envelope extraction, and envelope entropy calculation, target mode signals with more concentrated envelope distribution and clearer attenuation patterns are selected. Thus, effective components characterizing the free attenuation properties of polyurethane buffers can be extracted more accurately from complex vibration responses, providing a stable and reliable signal foundation for subsequent damping feature extraction, stiffness feature extraction, and anomaly identification.

[0121] In one embodiment of the present invention, step S4 includes:

[0122] S41: Perform frequency shift scaling on the damping response signal, and then perform mean removal and amplitude normalization on the frequency shift scaling signal to obtain a preprocessed signal;

[0123] S42: The preprocessed signal is used as an external driving input to the bistable stochastic resonance system. In the bistable stochastic resonance system, random noise is superimposed on the preprocessed signal to drive the state of the bistable stochastic resonance system to transition between the two potential wells, and the enhanced damped response signal is output.

[0124] S43: Perform variational mode decomposition on the enhanced damping response signal: Set the constraint that the decomposed modal signals can reconstruct the enhanced damping response signal, and take the center frequency corresponding to each modal signal as the parameter to be estimated. Under the premise of satisfying the constraint, construct a variational optimization problem that minimizes the bandwidth of each modal signal near the center frequency; use the augmented Lagrange method to process the variational optimization problem, and perform modal signal update, center frequency update and Lagrange multiplier update in an alternating iterative manner until the change of the modal signal in two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, and output all decomposed modal signals;

[0125] S44: Perform Hilbert transform on each modal signal to obtain modal analytical signals; calculate the modulus of each modal analytical signal at each time step in continuous sampling to obtain the envelope amplitude sequence of each modal signal; normalize each envelope amplitude sequence to obtain the probability distribution of the envelope energy of each modal signal along the time axis; calculate the envelope entropy of each modal signal according to the probability distribution using the information entropy calculation method; select the modal signal with the smallest envelope entropy as the target modal signal.

[0126] Considering that the classical bistable stochastic resonance theory is based on the adiabatic approximation assumption, it requires the input signal to be a low-frequency, small-parameter signal, i.e., the driving frequency is much less than 1 Hz and the excitation amplitude is much less than 1. However, the free decay vibration in the horizontal direction of a polyurethane buffer after compression and rebound is usually affected by its geometric dimensions and the dynamic elastic modulus of the material, generally ranging from a few Hz to tens of Hz. If the damping response signal is directly input into the bistable stochastic resonance system, the transition process between the two potential wells will be difficult to follow the rhythm of the external driving signal, easily leading to the system falling into a local small oscillation state within a single potential well, thus weakening or even eliminating the enhancement effect of stochastic resonance on weak features. Therefore, in step S4, the damping response signal is first subjected to frequency shift scaling transformation to equivalently map the original signal to a low-frequency signal that satisfies the input conditions of bistable stochastic resonance. As an example, when the main vibration frequency of the damping response signal is approximately 10 Hz, a scaling factor is set. , time variable Linear scaling, mapped to a virtual time variable At this time, the damping response signal It is equivalently stretched on the time axis, and its frequency components are synchronously compressed proportionally. After frequency shift scaling transformation, its equivalent main frequency is about 0.1Hz, thus falling into the relatively ideal operating frequency response range of the bistable stochastic resonance system.

[0127] The mean processing and amplitude normalization processing in step S41 are as follows: First, the average value of the signal after frequency shift scaling is calculated over the entire sampling time range. Then, the value of each sampling point in the signal is subtracted from this average value to obtain the mean-reduced signal. After the mean-reducing processing is completed, the sampling point with the largest absolute value in the mean-reduced signal is found and used as the amplitude reference. Then, the value of each sampling point in the signal is divided by the amplitude reference, thereby uniformly adjusting the amplitude of the entire signal to a smaller and fixed range.

[0128] Step S41 involves processing the damping response signal. By sequentially performing frequency shift scaling, mean removal, and amplitude normalization, a preprocessed signal suitable for input bistable stochastic resonance systems can be obtained. The preprocessed signal It retains the main dynamic characteristics related to the damping vibration of the original damping response, and adjusts its frequency and amplitude scales to a range that the stochastic resonance system can easily respond to and enhance, thus laying the foundation for subsequent weak damping characteristic enhancement processing.

[0129] Reference Figure 5 Specifically, the method for outputting the enhanced damped response signal of the bistable stochastic resonance system in S42 includes:

[0130] The potential function of the bistable stochastic resonance system is:

[0131] ,

[0132] in, For the state variables of a bistable stochastic resonance system; and For the bistable state trap parameters, and and These parameters are used to adjust the potential well depth and the potential barrier height, thereby determining the bistable structural characteristics of the system. and It is preferable to set it in the range of 0.5 to 2;

[0133] By coupling the negative gradient term of the potential function, the external driving input, and the random noise excitation, the dynamic equation of the bistable stochastic resonance system is established as follows:

[0134] ,

[0135] in, The state output signal of the bistable stochastic resonance system serves as the enhanced damped response signal. The noise intensity coefficient is used to adjust the energy level of injected random noise; it is a noise intensity parameter. The preferred setting is 0.1 to 0.3, which can usually achieve a good balance between promoting effective transitions and suppressing excessive noise disturbances, thereby improving the enhancement effect of weak damping characteristics and the reliability of subsequent processing; Zero-mean, unit-intensity Gaussian white noise represents the time-varying random noise process in a stochastic resonance system, used to characterize the random disturbances experienced by the system at various moments. For preprocessed signals;

[0136] The enhanced damping response signal is obtained by iteratively solving the dynamic equations point by point using the Euler-Maruyama numerical discretization method. .

[0137] The point-by-point iterative solution specifically involves: first, preprocessing the signal... The system sequentially uses the sampling time sequence as the external driving input and determines the numerical integration step size based on the actual sampling frequency, so that each sampling point corresponds to a discrete iteration time. Then, taking the initial state of the system as the starting point of the iteration, at each discrete time, the combined effects of the negative gradient term of the potential function, the external driving input term, and the random noise excitation term are considered simultaneously to calculate the system state increment corresponding to the current time, and update the system state value at the next sampling time accordingly. After that, the process is repeated point by point along the entire sampling sequence until the numerical iteration of all sampling points is completed, resulting in a system state output sequence corresponding to the length of the input signal. This system state output sequence is then used as the enhanced damped response signal. .

[0138] Reference Figure 6 Specifically, S43 includes:

[0139] S431: The constraint condition is set as the ability of the decomposed modal signals to reconstruct the enhanced damped response signal. The functional expression of the constraint condition is as follows:

[0140] ,

[0141] in, Indicates the first One modal signal; For modal index, ; The total number of decomposed modes; the total number of decomposed modes. The number of modes is preferably set to 3 to 6 to balance the effective feature separation capability and modal physical meaning. The total number of modes is further preferably set to 4 to achieve effective separation of high-frequency disturbance components, main damping vibration components and low-frequency background components in the enhanced damped response signal.

[0142] S432: The center frequency corresponding to each modal signal As parameters to be estimated, under the premise of satisfying the aforementioned constraints, a mechanism is constructed that ensures each modal signal is at its corresponding center frequency. The variational optimization problem with the minimum nearby bandwidth is addressed by using the augmented Lagrangian method.

[0143] The augmented Lagrange function is expressed as:

[0144] ,

[0145] in, Represents the set of all modal signals; It represents the set of center frequencies of all modal signals, used to characterize the main energy concentration location of each modal signal in the frequency domain; For Lagrange multipliers, This represents the time partial derivative operator. The Dirac function is represented by the expression "dirac function" in specific operations. The form of participating in temporal convolution; Represents the imaginary unit. This represents the convolution operation. It is a complex exponential modulation factor, used to modulate the frequency that originally revolved around the center frequency. The modal spectrum of the distribution is shifted to the baseband centered at zero frequency; This represents the L2 norm square operator. This represents the inner product operation, used to measure the matching relationship between the Lagrange multiplier and the mode reconstruction error over the entire time domain; These are the modal narrowband constraint weighting parameters, used to control the spectral concentration of each modal signal near its corresponding center frequency; modal narrowband constraint weighting parameters The preferred setting is 1000 to 3000, as this range can generally balance modal spectrum concentration, avoid modal aliasing, and suppress over-decomposition.

[0146] S433: Within the augmented Lagrange framework, modal signal updates, center frequency updates, and Lagrange multiplier updates are performed sequentially using an alternating iterative approach.

[0147] Under the condition of fixed center frequency and Lagrange multipliers, each modal signal is updated. Based on the updated modal signal, the corresponding center frequency is updated according to the spectral energy distribution. And based on the current modal signal and the enhanced damped response signal... The reconstruction error between them updates the Lagrange multipliers;

[0148] S434: When the change in modal signal between two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, the variational mode decomposition is completed and all modal signals are output.

[0149] Specifically, in step S44, the modal signals obtained from the decomposition are... Perform a Hilbert transform to construct its analytic signal:

[0150] ,

[0151] in, Represents the imaginary unit; This represents the Hilbert transform operator.

[0152] The modal signal is obtained by analyzing the modulus of the signal. The envelope amplitude signal is expressed as follows:

[0153] ,

[0154] In the digital signal processing implementation, to adapt to the discrete sampling characteristics of millimeter-wave radar and facilitate subsequent numerical calculations, the envelope amplitude signal is... Discretize the data according to the sampling time to obtain the envelope amplitude sequence. ,and ;in, For discrete-time indexing, The number of sampling points. It should be noted that, because stochastic resonance enhancement and variational mode decomposition are essentially just different levels of signal transformation and analysis of the same radar sampled signal, without introducing resampling, time reconstruction, or asynchronous update mechanisms, the discrete signals in each processing stage naturally inherit the original sampling time axis of the millimeter-wave radar, and their discrete time index... This corresponds to a discrete time series with the radar sampling period as the time step.

[0155] To convert the envelope amplitude sequence The probability distribution is transformed into a probability distribution that can be used for entropy calculation. After normalizing the envelope amplitude, the probability distribution expression is as follows:

[0156] ,

[0157] From the above normalization definition, it can be seen that the... And satisfy Its probability distribution sequence constitutes a discrete probability distribution of the proportion of the envelope amplitude on the time axis.

[0158] Therefore, modal signals can be calculated using the information entropy calculation method. The corresponding envelope entropy is:

[0159] ,

[0160] Among them, envelope entropy The discrete entropy value characterizes the envelope distribution of a modal signal. A smaller envelope entropy indicates a more concentrated discrete probability distribution, meaning the envelope energy tends to accumulate at a few sampling points, which is more conducive to highlighting anomalous impacts or characteristic transient components. Conversely, a larger envelope entropy indicates a more uniform energy distribution, typically corresponding to modes dominated by background vibration or noise. Extracting the envelope entropy... Minimum modal signal As the target modal signal.

[0161] In step S5, this invention utilizes Hilbert transform to extract envelope amplitude and phase information from the target modal signal, and constructs damping characteristic indices and stiffness characteristic indices respectively, thereby achieving a quantitative characterization of the attenuation and elastic properties of the polyurethane buffer. Furthermore, these two indices are constructed into a health feature vector, providing a feature data foundation for subsequent anomaly identification of the polyurethane buffer. In one embodiment of this invention, step S5 includes S51 to S54.

[0162] Step S51: Perform a Hilbert transform on the target modal signal to obtain the target modal analytic signal.

[0163] ,

[0164] in, Represents the imaginary unit; Represents the Hilbert transform operator;

[0165] The target modal analytical signal is moduloed point by point during continuous sampling time to obtain the target envelope amplitude sequence;

[0166] Target modal analytical signal The expression for obtaining the target envelope amplitude signal by modulus is:

[0167] ,

[0168] The target envelope amplitude signal directly reflects the energy decay characteristics of the polyurethane buffer after the compression rebound ends, providing a basis for subsequent damping characteristic analysis.

[0169] The argument of the target modal analytical signal is calculated point by point for each value in the continuous sampling time to obtain the target phase sequence;

[0170] Target modal analytical signal The expression for solving for the argument is:

[0171] ,

[0172] Target phase signal The rate of change over time corresponds to the target angular frequency information of the vibration system, which is used to characterize the main frequency characteristics of the target modal signal vibration and to provide a basis for subsequent stiffness characteristic analysis.

[0173] Step S52: Perform a natural logarithmic operation on the target envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence;

[0174] For the target envelope amplitude signal Performing the natural logarithm operation yields the logarithmic envelope function. The expression is:

[0175] ;

[0176] Based on the decay trend of the logarithmic envelope amplitude sequence, a preset linear fitting interval corresponding to the amplitude decaying from the initial peak to a preset proportion is selected. Within this interval, a least-squares method is used for linear regression to obtain the regression slope. The absolute value of the regression slope is then taken to obtain the damping characteristic index. .

[0177] In practical applications, based on the decay trend of the target envelope amplitude, a linear interval with clear physical meaning and minimal noise influence during the vibration decay process is selected. Preferably, this is the linear fitting interval corresponding to the amplitude decaying from the initial peak to a preset proportion (preferably set to 90% to 10%). This is to avoid interference with the fitting results during the initial transient impact phase and the low signal-to-noise ratio range at the tail end. Among these, Indicates the start time point. Indicates the end time.

[0178] In the selected linear fitting interval Within, the least squares method is used to analyze the logarithmic envelope function. Perform linear regression and establish a regression model:

[0179] ,

[0180] in, For the regression intercept, The regression slope, This is the residual term.

[0181] It should be noted that the aforementioned and For target mode analytical signal The established continuous-time representation is used to characterize the theoretical variation law of vibration attenuation; in actual engineering calculations, millimeter-wave radar signals are acquired through discrete sampling, therefore the target envelope amplitude signal is... The target envelope amplitude sequence is obtained by discretizing the data at the sampling time. The corresponding logarithmic envelope sequence is Subsequent regression slope estimates are all based on this discrete sequence; among which, It is a discrete-time index, and , This represents the number of sampling points.

[0182] By using least squares regression within the linear fitting interval, the log-envelope sequence is analyzed. A linear relationship between time and regression slope is fitted to obtain the corresponding regression slope, thereby estimating the equivalent damping coefficient. This method does not depend on the number of complete vibration cycles, nor does it require precise acquisition of individual peak positions, making it particularly suitable for engineering applications where the duration of the free damping damping vibration process of a polyurethane buffer is short and the effective vibration cycle is limited. The regression slope is then used to... The absolute value of is defined as the damping characteristic index, and its expression is:

[0183] .

[0184] Step S53: The target angular frequency is defined as the target phase. The first derivative with respect to time, i.e. ;

[0185] The discrete target angular frequency sequence is calculated by taking the phase difference between adjacent sampling times from the target phase sequence. The expression is as follows:

[0186] ;

[0187] in, Indicates the first The instantaneous phase values ​​extracted from the target modal analytical signal at each discrete sampling time; This represents the instantaneous phase value corresponding to the next sampling moment. The sampling period; This indicates that the instantaneous phase difference between adjacent sampling points is used to determine the phase difference at the th sampling point. The discrete target angular frequency value, obtained by approximating the vibration process between one sampling time and the next sampling time, is used to characterize the instantaneous vibration frequency characteristics of the target modal signal within the corresponding sampling interval.

[0188] The discrete target angular frequency sequence is statistically averaged over the time interval corresponding to the linear fitting interval to obtain an estimated principal oscillation frequency; the estimated principal oscillation frequency is then squared. stiffness characteristic index .

[0189] Stiffness characteristic indicators can effectively reflect the trend of changes in the state of the buffer material, such as hardening, changes in elastic modulus, or deterioration of structural integrity. At the same time, they avoid dependence on mass parameters and complex mechanical models, making them suitable for on-site non-contact testing scenarios.

[0190] It should be noted that when discretizing the target angular frequency sequence in step S53, the aforementioned target envelope amplitude sequence... With discrete target angular frequency sequence All are constructed based on the same discrete sampling time axis of millimeter-wave radar, and their discrete time indexes are... With the number of sampling points While maintaining consistency in the temporal sense, the target envelope amplitude sequence is directly calculated from a single sampling point, and its index range covers all sampling points. The discrete target angular frequency sequence, on the other hand, is obtained by the instantaneous phase difference between adjacent sampling points, used to characterize the vibration frequency characteristics between adjacent sampling times. Therefore, its effective index range typically has an offset of one sampling point relative to the envelope amplitude sequence. These differences stem from different signal processing methods and are normal engineering implementation variations, not affecting the unified analysis and comprehensive evaluation of all features on the same time axis.

[0191] During the free-damping decay vibration stage of the polyurethane buffer, although the vibration amplitude gradually decreases over time, the instantaneous frequency of the system typically fluctuates slightly around its natural frequency. In step S53, to reduce the impact of local fluctuations in instantaneous frequency and noise interference on the results, within the selected linear fitting interval... Within, for discrete target angular frequency sequences The average value was statistically averaged and used as the estimated principal frequency of the polyurethane buffer during the free damping damping vibration process.

[0192] ,

[0193] Where W is the estimated value of the dominant oscillation frequency. and These represent the selected linear fitting intervals. The start and end index points on the discrete sampling time axis; This represents the index range corresponding to the selected effective fitting interval on the discrete sampling time axis; This represents the number of discrete target angular frequency sample points that participate in the statistical averaging within the effective fitting interval, and its value satisfies... ;coefficient This method converts statistical results in angular frequency form into a physical frequency representation in Hertz. This processing method improves the stability and noise immunity of the principal oscillator frequency estimation results while preserving the physical meaning of the frequency.

[0194] Step S54: Using the damping characteristic index and stiffness characteristic index Constructing a health feature vector:

[0195] ,

[0196] in, This indicates the transpose operation.

[0197] This invention establishes a health baseline space through step S6 and uses Mahalanobis distance to measure the deviation of health feature vectors and determine thresholds. This comprehensively reflects the degree of change of the current state of the polyurethane buffer relative to its healthy state, effectively improving the accuracy and robustness of anomaly identification results. In one embodiment of this invention, step S6 includes:

[0198] S61: Collect multiple sets of health benchmark samples, and obtain health feature vectors of multiple sets of health benchmark samples by calculating damping characteristic index and stiffness characteristic index; calculate the average of each health feature vector according to the feature dimension to obtain the health sample benchmark mean; subtract each health feature vector from the health sample benchmark mean vector to obtain the corresponding deviation vector, and calculate the average of the outer product of each deviation vector and its transpose vector to obtain the health benchmark covariance matrix; perform regularization processing on the health benchmark covariance matrix to obtain the invertible covariance matrix; construct the health benchmark space using the binary tuple composed of the health benchmark mean vector and the invertible covariance matrix.

[0199] S62: Calculate the squared Mahalanobis distance of the health feature vector obtained in step S5 relative to the health reference space, and use the squared Mahalanobis distance as the anomaly score;

[0200] S63: Compare the abnormal score with a preset abnormal judgment threshold. When the abnormal score is not greater than the abnormal judgment threshold, determine that the current health feature vector is within the health reference space and output a normal result; when the abnormal score is greater than the abnormal judgment threshold, determine that the current health feature vector is outside the health reference space and output an abnormal result.

[0201] Specifically, in step S61:

[0202] After the polyurethane buffer is installed and confirmed to be in good condition, a compression rebound test is performed, and the obtained multiple sets of health benchmark samples are recorded as follows. , Indicates the first The health feature vector obtained from this test is used as the health baseline feature vector, and ; Index for health benchmark samples; This represents the number of healthy baseline samples, which is preferred in engineering. To ensure the stability of covariance estimation;

[0203] The health baseline center is defined as the sample mean, and the health baseline sample mean is:

[0204] ,

[0205] Among them, the mean of the health baseline sample represents the central position of the health status in the feature space;

[0206] To characterize the fluctuation scale of various health indicators and their correlations, the health benchmark covariance matrix is ​​defined as follows:

[0207] ,

[0208] Among them, the health baseline covariance matrix It is a symmetric positive semi-definite matrix; Indicates the transpose operation; The diagonal elements represent the variance of each index, while the off-diagonal elements represent the correlation between the indices. For example, the decrease in stiffness due to temperature rise is linked to the change in damping.

[0209] To avoid the health baseline covariance matrix becoming irreversible or ill-conditioned due to small samples and strong correlations, this embodiment of the invention uses regularization to generate an invertible covariance matrix:

[0210] ,

[0211] in, This is the invertible covariance matrix after regularization; The inverse matrix can be represented as This is used to scale and normalize the indicators in each dimension of the health benchmark feature vector and eliminate the correlation between indicators. For example, the regularization coefficient, (e.g.) (It can be adaptively set by the baseline data level). for An identity matrix, where 2 represents the number of health benchmark characteristic indicators, and satisfies... ;

[0212] Will be composed of binary pairs The defined statistical space is denoted as the health baseline space:

[0213] .

[0214] The health baseline space is used to characterize the statistical distribution characteristics of various health indicators of the polyurethane buffer under healthy conditions. The health baseline space parameters can be stored in a non-volatile storage unit inside the detection device to ensure that the health baseline parameters are not lost after the device is powered off or restarted, and to be updated when the polyurethane buffer is maintained, replaced, or recalibrated.

[0215] It should be noted that the health benchmark samples in this embodiment of the invention are preferably samples obtained from polyurethane buffers of the same specification under confirmed healthy conditions. This is because polyurethane buffers of different specifications typically have inherent differences in structural dimensions, material parameters, rated stroke, stiffness level, and damping response. Even if they are all in a healthy state, the distribution range of their corresponding health feature vectors may differ. To avoid introducing additional statistical dispersion due to specification differences and affecting the characterization accuracy of the health benchmark space, it is preferable to establish the health benchmark space based on healthy samples of the same specification. When applicable to buffers of multiple specifications, corresponding health benchmark spaces can also be established for different specifications, and the health benchmark space matching the specification of the buffer under test can be called for anomaly identification during testing.

[0216] When the system performs a test on the current polyurethane buffer, it outputs a set of health feature vectors. At that time, steps S62 and S63 are performed to output one detection result. Specifically, the Mahalanobis distance... and abnormal scores The expression is:

[0217] ,

[0218] ,

[0219] in, This represents the deviation vector of the health feature vector obtained from this test result relative to the mean of the health benchmark sample; The offset vector is in transpose form and is used in distance calculation. Indicates the transpose operation; The inverse of the invertible covariance matrix of the healthy baseline sample is used to reflect the fluctuation scale and correlation of various health indicators under healthy conditions, thereby performing a statistically weighted evaluation of the deviation vectors. It should be noted that in Mahalanobis distance calculation, the inverse of the invertible covariance matrix is ​​located in the middle of the deviation vectors, serving as a statistical metric bridge connecting the left and right deviation vectors. It defines the scale relationship and correlation structure between various health indicators under healthy conditions, thus enabling the distance calculation to reflect the rationality of joint deviations rather than the absolute change of a single indicator.

[0220] In Mahalanobis distance anomaly measurement, the inverse of the covariance matrix is ​​used to reflect the statistical fluctuation characteristics of each health indicator under healthy conditions and their interrelationships, thereby enabling differentiated evaluation of different types of deviations. Specifically, when a health indicator is significantly affected by environmental factors and has a wide natural fluctuation range under normal conditions, its variance in healthy samples is large. After weighting with the inverse of the covariance matrix, the contribution of this indicator to the anomaly score is relatively reduced, allowing the system to tolerate normal fluctuations caused by environmental disturbances such as temperature changes and load differences, thus avoiding misjudgments. Conversely, when a health indicator has high stability and a small fluctuation range under healthy conditions, even small-amplitude anomaly deviations will significantly affect the anomaly score after weighting, thereby improving the system's sensitivity to early anomalies or performance degradation.

[0221] Calculate the health baseline eigenvector The outlier score is used as the baseline sample score. Its expression is:

[0222] ;

[0223] Calculate the baseline sample score mean health benchmark score and standard deviation of health benchmark score Its expression is:

[0224] ,

[0225] ,

[0226] in, Indicates the first The health baseline feature vector obtained from this test, and ; Index for health benchmark samples; This represents the number of healthy baseline samples, which is preferred in engineering. This is to ensure the stability of the mean and standard deviation of the health baseline score.

[0227] Anomaly detection threshold in this invention embodiment Set as:

[0228] ,

[0229] in, The threshold coefficient is preferably... This ensures that score changes caused by environmental noise and operating condition fluctuations in a healthy state will not lead to misjudgment, while also providing an effective response when the damping and stiffness characteristics of the polyurethane buffer deviate significantly.

[0230] When the system performs a test on the current polyurethane buffer, if the anomaly score is... satisfy When the polyurethane buffer is deemed normal, a normal output result is given, and its indicators are within the statistical fluctuation range allowed by the health baseline; if the abnormal score is... Appear When an abnormality is detected in the polyurethane buffer, an abnormal result is output, and the corresponding test sample is an abnormal score sample.

[0231] To reduce the impact of occasional interference on anomaly detection results and improve the stability and reliability of anomaly identification, this embodiment of the invention introduces a continuous triggering mechanism as a means of preventing false alarms during the anomaly detection stage. The occasional interference includes, but is not limited to, short-term fluctuations in health characteristics caused by factors such as foreign object impacts in the pit, instantaneous strong noise, communication link jitter, or data packet loss. Specifically, this invention does not directly output anomaly alarms based on a single detection result, but instead sets a threshold for the number of consecutive triggers. Only when in consecutive During the testing process, the system only determines that the polyurethane buffer is in an abnormal state and outputs a final abnormal alarm signal when all calculated abnormal scores exceed the corresponding abnormal judgment threshold. If the abnormal score only exceeds the threshold in a few testing cycles and does not meet the continuous triggering condition, the situation is recorded as a suspected abnormality or transient disturbance, and a retest or continuous monitoring process can be triggered without immediately outputting an abnormal alarm.

[0232] Among them, the threshold for the number of consecutive triggers The configuration can be adjusted according to the specific application scenario, elevator operating environment, and false alarm tolerance, for example, set to Through the aforementioned continuous triggering mechanism, this embodiment can effectively suppress false alarms caused by occasional interference, while ensuring reliable identification of real anomalies or persistent deterioration states.

[0233] As an example, the executable instructions in the detection method of the present invention can be deployed to be executed on a computing device, or on multiple computing devices located at one location, or on multiple computing devices distributed in multiple locations and interconnected by a communication network.

[0234] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0235] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0236] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0237] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for detecting anomalies in elevator polyurethane buffers based on millimeter-wave radar, characterized in that, Includes the following steps: S1: Aim the main beam of the millimeter-wave radar at the side wall of the elevator polyurethane buffer and collect the echo signal within a preset time window after the polyurethane buffer has finished compressing and rebounding. S2: Determine the target range cell through the echo signal and extract the corresponding phase change information, and generate the vibration displacement signal in the radar line-of-sight direction according to the phase-displacement mapping relationship; S3: Identify the intervals in the vibration displacement signal where the overall vibration envelope shows a decaying trend, and extract the vibration displacement signal corresponding to the interval as the damping response signal; S4: After preprocessing the damping response signal, bistable stochastic resonance enhancement is performed. The enhanced damping response signal is then subjected to variational mode decomposition to obtain multiple mode signals. The envelope entropy of each mode signal is calculated, and the mode signal with the smallest envelope entropy is selected as the target mode signal. S5: After analyzing the target modal signal, extract the target envelope amplitude information and the target phase information. Calculate the damping characteristic index with the target envelope amplitude and the stiffness characteristic index with the target phase. Construct a health feature vector through the damping characteristic index and the stiffness characteristic index. S6: Construct a health benchmark space based on the health benchmark sample, and determine whether the health feature vector obtained in step S5 is located in the health benchmark space. If it is, it is judged to be normal; otherwise, it is judged to be abnormal. Step S4 includes: S41: Perform frequency shift scaling on the damping response signal, and then perform mean removal and amplitude normalization on the frequency shift scaling signal to obtain a preprocessed signal; S42: The preprocessed signal is used as an external driving input to the bistable stochastic resonance system. In the bistable stochastic resonance system, random noise is superimposed on the preprocessed signal to drive the state of the bistable stochastic resonance system to transition between the two potential wells, and the enhanced damped response signal is output. S43: Perform variational mode decomposition on the enhanced damping response signal: Set the constraint that the decomposed modal signals can reconstruct the enhanced damping response signal, and take the center frequency corresponding to each modal signal as the parameter to be estimated. Under the premise of satisfying the constraint, construct a variational optimization problem that minimizes the bandwidth of each modal signal near the center frequency; use the augmented Lagrange method to process the variational optimization problem, and perform modal signal update, center frequency update and Lagrange multiplier update in an alternating iterative manner until the change of the modal signal in two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, and output all decomposed modal signals; S44: Perform Hilbert transform on each modal signal to obtain modal analytical signals; calculate the modulus of each modal analytical signal at each time step in continuous sampling to obtain the envelope amplitude sequence of each modal signal; normalize each envelope amplitude sequence to obtain the probability distribution of the envelope energy of each modal signal along the time axis; calculate the envelope entropy of each modal signal using the information entropy calculation method according to the probability distribution; select the modal signal with the smallest envelope entropy as the target modal signal; Step S5 includes: S51: Perform Hilbert transform on the target modal signal to obtain the target modal analytical signal; calculate the modulus of the target modal analytical signal at each time step in the continuous sampling time to obtain the target envelope amplitude sequence; calculate the argument of the target modal analytical signal at each time step in the continuous sampling time to obtain the target phase sequence. S52: Perform a natural logarithmic operation on the target envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence; based on the decay trend of the logarithmic envelope amplitude sequence, select a preset linear fitting interval corresponding to the amplitude decaying from the initial peak to a preset proportion range, and perform linear regression using the least squares method within the linear fitting interval to obtain the regression slope; take the absolute value of the regression slope to obtain the damping characteristic index. ; S53: The target angular frequency is defined as the first derivative of the target phase with respect to time. A discrete target angular frequency sequence is obtained by calculating the phase difference between adjacent sampling times of the target phase sequence. The discrete target angular frequency sequence is then statistically averaged over the time interval corresponding to the linear fitting interval to obtain an estimated principal frequency. The estimated principal frequency is then squared to obtain the stiffness characteristic index. ; S54: Through the aforementioned damping characteristic index and stiffness characteristic index Constructing a health feature vector: ， in, This indicates the transpose operation.

2. The method according to claim 1, characterized in that, Step S2 includes: S21: Perform a range-direction fast Fourier transform on the discrete sampled data of the echo signal to obtain complex echo data distributed along the range direction, and then divide the complex echo data into multiple range cells according to the range resolution of the millimeter-wave radar. S22: Within a preset distance range, calculate the modulus square of the complex echo data of the distance unit at continuous sampling time, accumulate or average the modulus square data to obtain the echo energy of each distance unit, and select the distance unit with the largest echo energy as the target distance unit corresponding to the polyurethane buffer. S23: Extract the complex echo data of the target distance unit at continuous sampling time, calculate the instantaneous phase at the sampling time of the complex echo data, and obtain a continuous phase sequence through phase unwinding processing; The continuous phase of the target distance unit at the first sampling time within the preset time window As a reference phase, for any sampling time continuous phase By performing differential processing, the phase change is obtained: ; S24: Based on the phase-displacement mapping relationship, the phase change is converted into a vibration displacement signal. The formula for the phase-displacement mapping relationship is as follows: ， in, Represents the sampling time The output vibration displacement signal reflects the real-time minute vibration displacement of the polyurethane buffer along the line of sight of the millimeter-wave radar. For millimeter wave wavelengths; It is a time variable.

3. The method according to claim 1, characterized in that, Step S3 includes: S31: Perform Hilbert transform on the vibration displacement signal to obtain an analytical signal, and calculate the vibration envelope amplitude sequence by taking the modulus of the values ​​of the analytical signal at each time point in the continuous sampling time. S32: Perform natural logarithmic operation on the vibration envelope amplitude sequence to obtain a logarithmic envelope amplitude sequence; perform least squares linear fitting on the logarithmic envelope amplitude sequence within a sliding time window to obtain the fitting slope and goodness of fit; determine the sliding time window where the fitting slope, goodness of fit, and vibration envelope amplitude meet the preset conditions as a candidate interval with a local decay trend; merge temporally adjacent candidate intervals to obtain an interval where the overall vibration envelope has a decay trend; S33: Extract the vibration displacement signal corresponding to the interval where the overall vibration envelope shows a decaying trend from the vibration displacement signal, and use it as the damping response signal.

4. The method according to claim 3, characterized in that, In step S32, the preset conditions for the fitting slope, goodness of fit, and vibration envelope amplitude are as follows: Within the sliding window, the fitting slope is less than zero, the goodness of fit is not less than the preset threshold, and the average value of the vibration envelope amplitude is higher than the preset noise threshold.

5. The method according to claim 1, characterized in that, The method for outputting the enhanced damped response signal of the bistable stochastic resonance system in S42 includes: The potential function of the bistable stochastic resonance system is: ， in, For the state variables of a bistable stochastic resonance system; and For the bistable state trap parameters, and and ; The dynamic equations of the bistable stochastic resonance system are as follows: ， in, The state output signal of the bistable stochastic resonance system serves as the enhanced damped response signal. The noise intensity coefficient, Zero-mean, unit-intensity Gaussian white noise For preprocessed signals; The enhanced damping response signal is obtained by iteratively solving the dynamic equations point by point using the Euler-Maruyama numerical discretization method. .

6. The method according to claim 1, characterized in that, S43 includes: S431: The constraint condition is set as the ability of the decomposed modal signals to reconstruct the enhanced damped response signal. The functional expression of the constraint condition is as follows: ， in, Indicates the first One modal signal; For modal index, ; To decompose the total number of modes; S432: The center frequency corresponding to each modal signal As parameters to be estimated, under the premise of satisfying the aforementioned constraints, a mechanism is constructed that ensures each modal signal is at its corresponding center frequency. The variational optimization problem with the minimum nearby bandwidth is addressed by using the augmented Lagrangian method. The augmented Lagrange function is expressed as: ， in, Represents the set of all modal signals; Represents the set of center frequencies of all modal signals. For Lagrange multipliers, This represents the time partial derivative operator. Represents the Dirac function, Represents the imaginary unit. This represents the convolution operation. It is a complex exponential modulation factor. This represents the L2 norm square operator. This represents the inner product operation. These are the modal narrowband constraint weight parameters; S433: Within the augmented Lagrange framework, modal signal updates, center frequency updates, and Lagrange multiplier updates are performed sequentially using an alternating iterative approach. Under the condition of fixed center frequency and Lagrange multipliers, each modal signal is updated. Based on the updated modal signal, the corresponding center frequency is updated according to the spectral energy distribution. And based on the current modal signal and the enhanced damped response signal... The reconstruction error between them updates the Lagrange multipliers; S434: When the change in modal signal between two adjacent iterations is lower than the preset threshold or the number of iterations reaches the maximum allowable value, the variational mode decomposition is completed and all modal signals are output.

7. The method according to claim 1, characterized in that, Step S6 includes: S61: Collect multiple sets of health benchmark samples, and obtain health feature vectors of multiple sets of health benchmark samples by calculating damping characteristic index and stiffness characteristic index; calculate the average of each health feature vector according to the feature dimension to obtain the health sample benchmark mean vector; subtract each health feature vector from the health sample benchmark mean vector to obtain the corresponding deviation vector, and calculate the average of the outer product of each deviation vector and its transpose vector to obtain the health benchmark covariance matrix; perform regularization processing on the health benchmark covariance matrix to obtain the invertible covariance matrix; construct the health benchmark space using the binary tuple composed of the health sample benchmark mean vector and the invertible covariance matrix. S62: Calculate the squared Mahalanobis distance of the health feature vector obtained in step S5 relative to the health reference space, and use the squared Mahalanobis distance as the anomaly score; S63: Compare the abnormal score with a preset abnormal judgment threshold. When the abnormal score is not greater than the abnormal judgment threshold, determine that the current health feature vector is within the health reference space and output a normal result; when the abnormal score is greater than the abnormal judgment threshold, determine that the current health feature vector is outside the health reference space and output an abnormal result.

8. The method according to claim 7, characterized in that, In step S61: Multiple sets of health baseline samples are denoted as , Indicates the first The health feature vector obtained from this test is used as the health baseline feature vector, and ; Index for health benchmark samples; Indicates the number of healthy baseline samples; The health baseline center is defined as the sample mean, and the health baseline sample mean is: ， Define the health baseline covariance matrix as follows: ， Among them, the health baseline covariance matrix It is a symmetric positive semi-definite matrix; Indicates the transpose operation; Invertible covariance matrix: ， in, This is the invertible covariance matrix after regularization; The inverse matrix can be represented as , The regularization coefficient is . for An identity matrix, where 2 represents the number of health benchmark characteristic indicators, and satisfies... ; The health baseline space is: 。

Images