Elevator steel wire rope broken wire damage identification method and system

Abstract

The application discloses a kind of elevator steel wire rope broken wire damage identification method and system, it is related to elevator damage identification technical field.The application is with rotary encoder as synchronous trigger source, synchronous acquisition magnetic flux leakage signal and visual image and uniform conversion into spatial equal-interval sequence;Magnetic flux leakage signal is carried out truncated gaussian weighted adaptive baseline correction, and visual image is carried out helical texture development, eliminates the exclusive interference of two types of modal;Using sliding window type wavelet packet decomposition energy entropy, based on multi-working condition health benchmark constructs single-mode DIH damage index;Through real-time confidence sharpening adaptive weighting mechanism fusion double-mode DIH, combined with spatial continuity and time repeatability cascade verification output final identification result.The application realizes the high-precision online detection of elevator steel wire rope broken wire, has internal and surface broken wire detection capability, strong anti-interference ability, low false detection rate, adapts to elevator complex operating condition, is suitable for elevator steel wire rope routine safety monitoring and operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of elevator damage identification technology, and in particular relates to a method and system for identifying broken wire damage in elevator steel wire ropes. Background Technology

[0002] Elevator wire ropes are the core load-bearing components of the traction system. During long-term service, they are prone to wire breakage due to fatigue, wear, and corrosion. If not detected and replaced in time, this can lead to serious safety accidents such as elevator falls. Current mainstream wire breakage detection technologies fall into two categories: magnetic flux leakage detection and visual inspection. Magnetic flux leakage detection alone is insufficiently sensitive to surface micro-broken wires and is easily affected by electromagnetic interference and baseline drift. Visual inspection alone is greatly affected by oil and rust obstructions, and the inherent spiral texture of the wire rope can easily be confused with broken wire characteristics, resulting in false positive and false negative rates that are difficult to meet engineering requirements.

[0003] Therefore, developing a broken wire damage identification method that is suitable for continuous operation of elevator wire ropes, has strong anti-interference ability, and high identification accuracy is a technical problem that this solution urgently needs to solve. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for identifying broken wire damage in elevator steel wire ropes. Through the coordinated efforts of dual-branch dedicated preprocessing, sliding window wavelet packet energy entropy analysis, dual-modal adaptive weighted fusion, and spatiotemporal cascade verification, high-precision online detection of broken wire damage in elevator steel wire ropes is achieved. This solves the problems of insufficient sensitivity of existing magnetic leakage detection to surface micro-broken wires and the significant impact of oil stains and rust on visual detection.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: This invention relates to a method for identifying broken wire damage in elevator wire ropes, comprising the following steps: Simultaneously acquire the magnetic leakage signal and visual image of the wire rope, and convert the magnetic leakage signal and visual image into a spatially equally spaced sampling sequence corresponding to the axial position of the wire rope; Adaptive baseline correction is performed on the magnetic flux leakage signal, and spiral texture unfolding is performed on the visual image to obtain the magnetic flux leakage feature sequence and the visual feature sequence. The spatially equally spaced sampling sequence is divided into sliding windows with a preset window length and step size. Wavelet packet decomposition is performed on the leakage magnetic feature sequence and visual feature sequence in each window to obtain multiple equal bandwidth frequency band components of the corresponding layer. The signal energy and relative energy ratio of each frequency band in each window are calculated in turn to form the frequency band energy distribution vector corresponding to each window. At the same time, the wavelet packet energy entropy of each window is calculated as an auxiliary index based on the relative energy ratio of each frequency band. Based on healthy steel wire rope samples, a benchmark value for the relative energy ratio of each frequency band is established. The offset between the relative energy ratio of each frequency band in the current window and the benchmark value, as well as the standard deviation of all offsets, are calculated. The single-mode DIH damage index values ​​of the leakage magnetic mode and visual mode corresponding to the window are generated. Calculate the real-time confidence scores of the magnetic flux leakage mode and the visual mode within the current window respectively; calculate the corresponding dynamic weights based on the two confidence scores, and perform weighted fusion of the two DIH indices to obtain the fused DIH index; A damage assessment threshold is set. When the fusion DIH index corresponding to the sliding window exceeds the damage assessment threshold, it is determined that there is a broken wire damage at the corresponding position of the window. Continuous windows that exceed the damage assessment threshold are identified as candidate broken wire segments. Within the candidate broken wire segments, the fusion DIH index values ​​corresponding to each window are compared. It is determined that there is a broken wire position within the sliding window with the largest fusion DIH index. Combining the fusion DIH index corresponding to the window where the broken wire position is located with the physical characteristics of the two modes, the number of broken wires and the broken wire depth are calculated through a pre-calibrated regression model to complete the damage degree classification and output the final identification result.

[0006] Optionally, the spatially equidistant sampling sequence uses the pulses output by the traction wheel rotary encoder as the global synchronization trigger source and the reference zero point of the traction wheel rotary encoder as the starting position. Based on the pulse count at each sampling moment, the diameter of the traction wheel, and the number of pulses per revolution of the encoder, the axial position of the wire rope corresponding to the sampling point is calculated. By interpolation resampling, the two signals are unified into a spatially equidistant sequence with a 1mm interval, so as to achieve precise alignment of the spatial position.

[0007] Optionally, the adaptive baseline correction specifically involves: fitting the baseline of the magnetic flux leakage signal using a fifth-order polynomial, introducing a truncated Gaussian weight function during the fitting process, assigning zero weight to sampling points whose signal amplitude deviates from the window mean by more than two standard deviations; subtracting the fitted baseline from the original signal to obtain the baseline-corrected stable magnetic flux leakage signal.

[0008] Optionally, the specific process for obtaining the wavelet packet energy entropy corresponding to each window includes: Set the window length and sliding step size, and divide the preprocessed magnetic flux leakage sequence and grayscale sequence into continuous windows along the axial spatial sequence to generate multiple sets of overlapping sub-signal segments. The db8 wavelet is selected as the wavelet basis function. Wavelet packet decomposition is performed on the leakage sub-signal and gray sub-signal in each sliding window to decompose the single signal into multiple frequency band components with equal bandwidth, and the reconstructed sub-signal corresponding to each frequency band is output. Energy statistics are performed on the reconstructed sub-signals of each frequency band, and the sum of the squares of the signal amplitudes is taken as the total energy of the frequency band. The total energy of the window is obtained by summing the energy of all frequency bands within a single window, and then the energy of each frequency band is divided by the total energy to obtain the relative energy ratio of each frequency band. The wavelet packet energy entropy value corresponding to the window is calculated based on the relative energy proportion of each frequency band within the window. After calculating all sliding windows sequentially along the axial direction, the leakage magnetic energy entropy sequence and the visual energy entropy sequence, which are continuously distributed along the rope length, are output.

[0009] Optionally, the specific steps for calculating the wavelet packet energy entropy include: Energy statistics are performed on each frequency band reconstructed sub-signal within each sliding window, and the sum of squares of the amplitudes of each reconstructed sub-signal is used to characterize the signal energy of the corresponding frequency band. The total energy of the signals in each frequency band within the same window is normalized to obtain the relative energy of each frequency band. Based on the relative energy, the wavelet packet energy entropy corresponding to the window is calculated in the form of information entropy. When the relative energy percentage of a certain frequency band is zero, the entropy contribution of that frequency band is ignored.

[0010] The specific formula is as follows: For the first The first sliding window Each frequency band is used to calculate the signal energy of that frequency band. The specific formula is as follows: ; In the formula, Let the amplitude of the reconstructed signal in the j-th frequency band be the value at the k-th sampling point. This represents the number of sampling points within the window. Calculate the relative energy percentage of each frequency band. The specific formula is as follows: ; Calculate the wavelet packet energy entropy for this window. The specific formula is as follows: ; In the formula, For the wavelet packet decomposition level, when hour, .

[0011] Optionally, the DIH curve generation process is as follows: Standard samples of healthy steel wire ropes were collected at various operating speeds and ambient temperatures to obtain energy entropy data for each frequency band under various operating conditions. The statistical mean of the energy entropy of the same frequency band under all operating conditions was taken as the health benchmark mean of the corresponding frequency band, forming a standardized health benchmark library. For a single sliding window to be detected, extract the energy entropy value of each frequency band within the window, and subtract it from the mean of the health benchmark for the corresponding frequency band in the health benchmark library one by one to obtain the energy entropy difference sequence of each frequency band. Calculate the global average value of the energy entropy difference of all frequency bands in the current window, and then use the global average value as a benchmark to calculate the unbiased standard deviation of the energy entropy difference of all frequency bands. Use the unbiased standard deviation as the single-mode DIH damage index value corresponding to the current window. The DIH index of all sliding windows is calculated sequentially along the axial direction of the wire rope. The DIH values ​​of each window are then sequentially spliced ​​together according to the axial position corresponding to their center to form a single-mode DIH curve that is continuously distributed along the rope length. The leakage magnetic mode DIH curve and the visual mode DIH curve are output respectively.

[0012] Optionally, the formula for calculating the DIH curve, which is continuously distributed along the rope length, is as follows: ; In the formula, For the first Single-mode DIH damage index values ​​for each window. For the first The first window The difference between the energy entropy of each frequency band and the mean of the health baseline For the first The global mean of the energy entropy differences across all frequency bands within a window. This represents the number of wavelet packet decomposition layers.

[0013] Optionally, the step of obtaining the fused DIH index includes: For the magnetic flux leakage mode, two operating condition features are extracted: signal-to-noise ratio and baseline fluctuation coefficient within the current window. For the visual mode, two operating condition features are extracted: the proportion of oil smudge occlusion area and image clarity within the current window. Based on two operating condition characteristics of the leakage magnetic field mode, the real-time comprehensive confidence level of the leakage magnetic field mode is calculated; based on two operating condition characteristics of the visual mode, the real-time comprehensive confidence level of the visual mode is calculated. The combined confidence scores of the two modes are sharpened, and then the sharpened confidence scores are normalized to obtain the fusion weights of the magnetic leakage mode and the visual mode, respectively. The sum of the two weights is 1. The leakage magnetic mode DIH index and the visual mode DIH index are weighted and summed according to their respective fusion weights to obtain the final fused DIH index.

[0014] Optionally, all continuous segments on the DIH curve that exceed the judgment threshold are marked as candidate wire breakage segments, and the start and end axial positions of each segment are recorded. Within each candidate wire breakage segment, the maximum value of the fused DIH index is searched as the peak point, and the axial position corresponding to the peak point is determined as the precise axial position of the wire breakage. The fused DIH value corresponding to the peak value is recorded, and four physical features corresponding to the current wire breakage position are extracted: peak amplitude of leakage magnetic signal, half-width of leakage magnetic peak, area of ​​visual wire breakage region, and contrast of visual wire breakage region. Combined with the fused DIH peak value and the real-time fusion weight of the two modes, the quantitative values ​​of the number of wire breakages and the depth of wire breakage are calculated respectively through a pre-calibrated multivariate regression model.

[0015] This invention is a broken wire damage identification system for elevator wire ropes, comprising: The multimodal synchronous acquisition and spatial domain conversion unit includes a leakage magnetic field sensor array, a linear industrial camera, a rotary encoder, and a synchronous trigger controller. It is used to synchronously acquire the leakage magnetic field detection signal and surface visual image of the wire rope by using the output pulse of the rotary encoder at the traction wheel as a global trigger source; and converts the two signals acquired in the time domain into an axial spatial equally spaced sampling sequence based on the encoder pulse count. The dual-branch signal preprocessing unit is used to perform adaptive baseline correction on the leakage magnetic signal, perform spiral texture unfolding processing on the visual image, fit the spiral trajectory of the steel wire and perform vertical offset compensation on the pixels, and extract the gray-scale sequence distributed along the axis. The sliding window wavelet packet decomposition and energy entropy calculation unit is used to perform multi-level wavelet packet decomposition on the leakage magnetic sequence and grayscale sequence in each window to obtain multiple equal bandwidth frequency band components. The signal energy and relative energy ratio of each frequency band in each window are calculated in turn to obtain the wavelet packet energy entropy corresponding to each window. The single-mode DIH damage index generation unit is used to calculate the difference between the relative energy ratio of each frequency band and the mean of the health baseline for each current sliding window, and then calculate the unbiased standard deviation of the differences of all frequency bands to obtain the single-mode DIH damage index value of the corresponding window, and output the DIH curve that is continuously distributed along the rope length. The dual-modal DIH adaptive fusion unit is used to calculate the real-time confidence of the magnetic leakage mode and the visual mode within the current window, respectively. Based on the two confidences, the corresponding dynamic weights are calculated, and the two DIH indices are weighted and fused to obtain the fused DIH index. The damage identification and quantitative assessment unit is used to determine the presence of broken wires at the corresponding location when the fused DIH index exceeds the threshold. The axial position corresponding to the DIH peak point is used as the precise location of the broken wire. Combining the fused DIH peak value and the physical characteristics of the two modes, the number of broken wires and the depth of broken wires are calculated through a pre-calibrated multivariate regression model to complete the damage level classification and early warning output.

[0016] The present invention has the following beneficial effects: This invention achieves high-precision online detection of broken wire damage in elevator steel wire ropes through multimodal synchronous acquisition and unified spatial domain conversion, combined with dual-branch dedicated preprocessing, sliding window wavelet packet energy entropy analysis, dual-modal adaptive weighted fusion, and spatiotemporal cascade verification. On the one hand, by complementing the information from both magnetic flux leakage and visual dual-modal approaches, it covers both internal and surface broken wires, expanding the detectable damage range and improving the broken wire detection rate. On the other hand, by ensuring accurate spatial matching of the two data streams through unified spatial domain alignment, combined with adaptive dynamic weighting and spatiotemporal dual verification mechanisms, it effectively suppresses false detections and missed detections caused by fluctuations in operating speed, changes in ambient temperature, on-site oil contamination, and electromagnetic interference. Without increasing hardware costs, it significantly improves the accuracy and operational robustness of elevator steel wire rope broken wire detection.

[0017] This invention addresses two types of specific interference in wire rope detection scenarios by designing a dual-branch independent preprocessing mechanism. For the leakage magnetic signal, it employs truncated Gaussian weighted adaptive baseline correction to automatically shield the interference of broken wire peak points, accurately fitting and eliminating the slowly changing baseline caused by operational jitter and sensor drift. For the visual signal, it uses spiral trajectory fitting and pixel offset compensation to flatten the inherent spiral texture of the wire rope, eliminating the contribution of periodic grayscale fluctuations to energy distribution. This dual preprocessing purifies the detection signal from the source, ensuring that subsequent wavelet packet energy entropy changes are dominated only by damage, significantly improving the significance of micro-broken wire features. This effectively solves the technical problems of large inherent structural interference and easy masking of micro-damage when existing DIH methods are directly applied to wire rope scenarios.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the steps of a method for identifying broken wire damage in elevator wire ropes according to the present invention. Figure 2 This is a schematic diagram of the structure of an elevator wire rope broken wire damage identification system according to the present invention; Figure 3 This is a schematic diagram illustrating the principle of sliding window wavelet packet decomposition and DIH index calculation in this invention. Figure 4 This is a visual comparison image of the steel wire rope spiral texture before and after unfolding according to the present invention; Figure 5 This is a logic block diagram of the dual-modal DIH adaptive weighted fusion of the present invention; Figure 6 This is a schematic diagram comparing the fused DIH curve and the single-mode DIH curve under the single micro-broken wire condition of the present invention. Figure 7 This is a schematic diagram illustrating the integration of DIH damage assessment and candidate broken wire labeling. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0023] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-7 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0024] Please see Figure 1 As shown, the present invention is a method for identifying broken wire damage in elevator wire ropes, comprising the following steps: Step S1, Multimodal Signal Synchronous Acquisition and Spatial Domain Unified Conversion: The pulses output by the traction wheel rotary encoder are used as the global synchronous trigger source to synchronously acquire the leakage magnetic signal and visual image of the wire rope; the axial position of the wire rope corresponding to each sampling point is calculated based on the encoder pulse count, and the two signals acquired in the time domain are uniformly converted into an axial spatial equally spaced sampling sequence, so that the spatial resolution of the two signals is completely consistent and the spatial position is accurately aligned.

[0025] Step S1 is implemented through the following sub-steps: Step S11, Hardware Synchronous Trigger Acquisition: Connect the pulse signal output by the rotary encoder at the traction wheel to the synchronous trigger controller, use the encoder pulse as the global synchronization clock, and simultaneously trigger the leakage magnetic field sensor array and the linear scan industrial camera to perform data acquisition, ensuring that the hardware time synchronization error of the two raw data is within the set range. Step S12, Axial position mapping calculation: Using the encoder's reference zero pulse as the coordinate origin of the wire rope's axial position, based on the encoder's cumulative pulse count at each sampling moment, combined with the traction sheave diameter and encoder pulse count per revolution parameters, the axial displacement value of the wire rope corresponding to that sampling point is calculated, and a corresponding position label is generated for each sampling data. Step S13, Spatial Equal Interval Resampling Alignment: Based on a fixed spatial interval, linear interpolation resampling is performed on the leakage magnetic signal with position label and the visual grayscale sequence respectively, unifying the two signals with non-uniform sampling in the time domain into an axial spatial equal interval sampling sequence, eliminating the problem of uneven spatial sampling caused by elevator acceleration and deceleration, and finally realizing a one-to-one correspondence between the spatial positions of the two signals.

[0026] In this invention, a 1024-line incremental rotary encoder is installed on the side of the elevator traction sheave, and the nominal diameter of the traction sheave is... The encoder pulse signal is input to the synchronous trigger controller, which simultaneously triggers the leakage magnetic field sensor array and the 2K linear scan industrial camera to acquire data, with a hardware synchronization time error of less than 1ms. The leakage magnetic field sensor adopts a linear array of Hall elements with a sampling frequency of 2kHz; the linear scan camera has a horizontal frequency of 10kHz and a horizontal resolution of 2048 pixels.

[0027] The acquired raw signal is a time-domain sequence. Due to fluctuations in the wire rope's running speed, time-domain sampling cannot correspond to uniform spatial intervals. In step S1, the encoder zero point is used as the spatial reference, and the corresponding axial position is calculated based on the pulse count at each sampling moment: the number of pulses per revolution is 1024, and the circumference of the traction sheave is approximately 1256.6 mm. Therefore, a single pulse corresponds to a wire rope displacement of approximately 1.23 mm. Through linear interpolation resampling, the leakage magnetic signal and the visual grayscale sequence are uniformly converted into spatially equally spaced sequences with 1 mm intervals, achieving precise spatial alignment of the two signals.

[0028] It should be noted that the original acquired signal is a time-domain sequence. Because the speed of the steel cable fluctuates during elevator acceleration and deceleration, time-domain equal-interval sampling cannot correspond to uniform spatial intervals, and direct analysis would introduce positional bias. This step converts the time-domain signal into an axially spatially equal-interval sequence. The conversion formula is as follows: ; In the formula, For the first The axial position of the wire rope corresponding to each sampling point, in mm; This is the encoder pulse count value at that sampling moment; The pulse count value at the encoder's reference zero point; The nominal diameter of the traction sheave is in mm. This represents the number of pulses per encoder revolution, which is 1024 in this embodiment.

[0029] pass After calculating the axial position of all sampling points, linear interpolation is used for resampling to unify the magnetic flux leakage signal and the visual grayscale sequence into a spatially equidistant sequence with a 1mm interval, so as to achieve a one-to-one spatial correspondence between the two signals. Step S1 transforms the time-domain sampling points into absolute spatial positions by mapping encoder pulses, breaking the limitation of traditional time-domain sampling being affected by elevator speed fluctuations. This ensures that the spatial reference for subsequent wavelet packet decomposition and DIH calculation is consistent. At the same time, it transforms non-uniform time-domain sampling into uniform spatial-domain sampling, completely eliminating the interference of uneven sampling density on energy entropy calculation during elevator acceleration and deceleration, and ensuring the stability and comparability of DIH indicators under different operating speeds.

[0030] Step S2, Dual-branch signal-specific preprocessing: Adaptive baseline correction is performed on the leakage magnetic signal to eliminate the slow baseline changes caused by running jitter and zero-point drift; the visual image is processed by spiral texture unfolding to eliminate the periodic grayscale fluctuations caused by the inherent spiral texture of the wire rope and extract the axial grayscale sequence. Specifically, step S2 includes the following sub-steps: Step S21: Sliding window division of magnetic flux leakage signal: The magnetic flux leakage signal after spatial domain transformation is divided into sliding windows according to the window length and step size consistent with the subsequent wavelet packet analysis. Each window contains axial sampling points of a continuous fixed length. Step S22, Adaptive Baseline Fitting and Correction: Within each window, a fifth-order polynomial with truncated Gaussian weighting is used to perform baseline fitting on the magnetic flux leakage signal. During the fitting process, broken wire peak points with amplitudes deviating from the window mean by more than two standard deviations are automatically masked to eliminate the interference of broken wire characteristics on the baseline fitting. The original magnetic flux leakage signal is subtracted from the fitted baseline to output the baseline-corrected stable magnetic flux leakage signal. Input the original leakage magnetic field signal sequence within the current window and calculate the mean of the signal within the window. with standard deviation The fitting weight for each sampling point is calculated using a truncated Gaussian weighting function. Points with amplitudes within twice the standard deviation of the mean are weighted according to the Gaussian function; points with amplitudes exceeding twice the standard deviation (i.e., the peak points of broken wires) have their weights reset to zero and do not participate in baseline fitting. A fifth-order polynomial baseline is obtained by fitting using the weighted least squares method. Subtract the fitted baseline from the original signal point by point to obtain the baseline-corrected stationary signal, and output it to step S3 for wavelet packet decomposition.

[0031] Step S23, Wire Helix Trajectory Detection and Parameter Fitting: Edge detection is performed on the acquired visual image of the wire rope surface. The helix edge trajectory of a single wire is identified by Hough transform, and three sets of characteristic parameters, namely helix amplitude, twist pitch and initial phase, are fitted. The original grayscale image of the steel wire rope surface is input, and the edge pixels of the steel wire are first extracted using Canny edge detection. Based on the Hough transform of the sine curve, parameter space voting is performed on the edge points to obtain the helical amplitude A and the lay length corresponding to the peak values. with initial phase The three sets of spiral parameters obtained from the output fitting are then fed into S24 for texture unwrapping.

[0032] Step S24, Spiral Texture Unfolding and Gray-Level Sequence Extraction: Based on the fitted spiral parameters, the corresponding offset compensation is performed on each column of pixels in the image along the longitudinal direction to flatten the spiral wire surface into parallel straight line texture, eliminating the periodic gray-level fluctuations caused by the inherent spiral texture; the average gray-level value of each column of the flattened image is extracted to form a gray-level sequence distributed along the axial direction. Input the original visual image and the spiral parameters obtained by fitting S23; for each column of the image The longitudinal offset of the steel wire column is calculated based on the spiral equation, and the entire column of pixels is then positioned along... Directional offset compensation corrects the spiral steel wire ridges to a horizontal straight line. After flattening, the average gray value of all pixels in each column of the image is calculated to form a one-dimensional axial gray value sequence, which is then output to step S3 for wavelet packet decomposition.

[0033] In this embodiment of the invention, due to elevator start-stop jitter and sensor zero-point drift, the baseline of the leakage magnetic signal changes slowly. Direct wavelet packet decomposition would severely interfere with the calculation of low-frequency band energy entropy, leading to misjudgment. Therefore, when performing adaptive baseline correction on the leakage magnetic signal, a truncated Gaussian-weighted fifth-order polynomial is used to fit the leakage magnetic signal baseline. During the fitting process, a truncated Gaussian weight function is introduced, and sampling points whose signal amplitude deviates from the window mean by more than two standard deviations are assigned zero weight, automatically eliminating the interference of broken wire peak points on the baseline fitting. The original signal is subtracted from the fitted baseline to obtain the baseline-corrected stable leakage magnetic signal.

[0034] It should be noted that the leakage magnetic signal within the window is... The fitted baseline is The fitting weight formula for each sampling point is: ; In the formula, Axial position Baseline fitting weights at sampling points; This represents the amplitude of the leakage magnetic signal at the corresponding position within the window. This represents the average value of the leakage magnetic field signal within the current sliding window. The standard deviation of the leakage magnetic signal within the current sliding window is used. For points where the signal amplitude deviates from the mean of the window by less than twice the standard deviation, the fitting weights are assigned according to the Gaussian function. For points that deviate from the mean by more than twice the standard deviation (usually broken wire peaks or strong interference points), the weights are reset to 0, thereby eliminating the interference of extreme points on the baseline fitting and realizing adaptive baseline correction.

[0035] The corrected signal is: Using the above formula, the corrected signal baseline is stable, eliminating the influence of slow drift on subsequent energy entropy calculations.

[0036] In this embodiment of the invention, the inherent helical texture of the wire rope generates periodic grayscale fluctuations, and its energy distribution can mask the subtle characteristics of micro-broken wires, leading to false detections or missed detections. This embodiment uses helical trajectory fitting and pixel offset compensation to flatten the helical surface into parallel straight-line textures, eliminating the periodic grayscale fluctuations caused by the inherent helical texture of the wire rope. The specific process for extracting the axial grayscale sequence is as follows: First, the helical edge of a single steel wire in the image is detected by Hough transform, and the equation of the helix is ​​obtained by fitting: ; In the formula, The amplitude is the helical amplitude, in mm. Twist pitch, in mm; The initial phase is expressed in rad.

[0037] For each column of the image , along the pixel Directional offset compensation is applied to achieve spiral flattening: ; In the formula, The original image grayscale values, This represents the grayscale value of the image after it has been flattened.

[0038] like Figure 4As shown, the left (a) image is a schematic diagram of the original surface of the wire rope before unfolding: multiple wires are arranged in a spiral pattern, and the inherent spiral structure of the wire rope forms a periodic oblique edge texture; the broken wire defect is mixed in the spiral texture and is covered by the background oblique edge, resulting in low feature recognition. When directly extracting features, it is easily confused with texture fluctuations, which is the core interference source causing false detection and missed detection. The right (b) image is a schematic diagram of the surface of the wire rope after unfolding: after fitting the spiral trajectory and compensating for the vertical offset of pixels, all the wire edges are corrected to be horizontal straight lines that are parallel to each other, and the periodic gray-level fluctuations caused by the inherent spiral texture are completely eliminated; the broken wire defect appears as a horizontal fracture mark independent of the background texture, which is significantly distinguished from the flat background. The significance of the broken wire feature is greatly improved, which eliminates the inherent structural interference for subsequent wavelet packet energy entropy calculation and DIH damage index construction, and ensures the detection accuracy of micro-broken wires. As can be seen from the figure, the surface of the wire rope becomes a parallel straight line texture after unfolding, and the periodic gray-level fluctuations are eliminated. Then, the average gray value of each column of images is extracted to form an axial gray-level sequence. This is used for subsequent wavelet packet analysis.

[0039] Please see Figure 3 As shown, step S3, sliding window wavelet packet decomposition and energy entropy calculation: Set a sliding window with a length covering twice the wire rope lay length, and slide it continuously along the axial spatial sequence with a fixed step size; perform wavelet packet decomposition on the leakage magnetic field sequence and grayscale sequence in each window to obtain multiple equal bandwidth frequency band components of the corresponding layer; calculate the signal energy and relative energy of each frequency band in each window in turn, and finally obtain the wavelet packet energy entropy corresponding to each window; Step S3 specifically includes the following sub-steps: Step S31, Sliding window parameter configuration and sequence division: The window length and sliding step size are the same as those in step S2 preprocessing. The window length is set to twice the nominal twist pitch of the wire rope, and the sliding step size is consistent with the spatial sampling interval. The preprocessed leakage magnetic sequence and grayscale sequence are divided into continuous windows along the axial spatial sequence to generate multiple sets of overlapping sub-signal segments. It should be noted that setting the window length to twice the twist pitch ensures that each window contains at least two complete spiral texture cycles, avoiding frequency band energy calculation deviations caused by boundary effects; the sliding step size is consistent with the spatial sampling interval, ensuring that the final output energy entropy sequence matches the original sampling resolution, thus guaranteeing positioning accuracy; Step S32, Wavelet Packet Parameter Selection and Multi-Level Decomposition: Select the db8 wavelet as the wavelet basis function and set the decomposition level to six levels; perform wavelet packet decomposition on the leakage sub-signal and grayscale sub-signal in each sliding window, decompose the single signal into sixty-four frequency band components with equal bandwidth, and output the reconstructed sub-signal corresponding to each frequency band; optimize the cost function of the wire rope broken wire sample, and finally determine that the db8 wavelet basis and six-level decomposition are the optimal combination: the compact support and regularity of the db8 wavelet are suitable for the abrupt change characteristics of the broken wire signal, and the six-level decomposition can completely cover the characteristic frequency band corresponding to the broken wire of the wire rope, while taking into account the computational efficiency; Specifically, the frequency band division table for each layer of wavelet packet decomposition is as follows: Step S33: Calculation of frequency band energy and relative energy ratio: Perform energy statistics on the reconstructed sub-signals of each frequency band, and use the sum of the squares of the signal amplitudes as the total energy of the frequency band; sum the energy of all frequency bands within a single window to obtain the total energy of the window, and then divide the energy of each frequency band by the total energy to obtain the relative energy ratio of each frequency band. In practice, 64 frequency band reconstructed sub-signals of a single window are input; for each sub-signal, the square of the amplitude is calculated point by point and then summed to obtain the total energy of that frequency band; the energy of the 64 frequency bands is added together to obtain the total signal energy of the current window; the energy of each frequency band is divided by the total energy to obtain the relative energy ratio of each frequency band, and the sum of the relative energies of all frequency bands is 1; a 64-dimensional relative energy ratio vector is output and sent to the subsequent energy entropy calculation stage.

[0040] Step S34, Wavelet Packet Energy Entropy Calculation and Sequence Output: Based on the principle of information entropy, calculate the wavelet packet energy entropy value corresponding to the window according to the relative energy ratio of each frequency band within the window; after completing the calculation of all sliding windows along the axial direction, output the leakage magnetic energy entropy sequence and the visual energy entropy sequence that are continuously distributed along the rope length.

[0041] In practice, the 64-dimensional relative energy ratio vector of a single window is input; according to the information entropy formula, the logarithmic weighted sum of the relative energy is calculated band by band to obtain the energy entropy value of the window; after the energy entropy calculation of all sliding windows is completed sequentially along the axial direction, they are spliced ​​together to form a leakage magnetic energy entropy sequence and a visual energy entropy sequence that are continuously distributed along the rope length; the two energy entropy sequences are output to step S4 to construct the DIH damage index.

[0042] In this embodiment of the invention, the sliding window length is set to 240mm (twice the twist pitch), containing 240 spatial sampling points; the window slides continuously along the axial direction in 1mm steps to ensure that the spatial resolution of the output DIH curve is consistent with the sampling resolution. The db8 wavelet basis function is selected to perform 6-level wavelet packet decomposition on the signal within each window, resulting in 64 equal-bandwidth frequency components.

[0043] For the first For signals within a sliding window, calculate the energy entropy using the following steps: Bandwidth energy calculation: for the first The first sliding window Each frequency band is used to calculate the signal energy of that frequency band. The specific formula is as follows: ; In the formula, Let the amplitude of the reconstructed signal in the j-th frequency band be the value at the k-th sampling point. The number of sampling points within the window is represented by the sum of squares of the discrete signal amplitudes, which characterizes the energy carried by the frequency band component and is a fundamental statistic for wavelet packet energy analysis.

[0044] Relative energy calculation: Divide the energy of each frequency band by the total energy within the window to obtain the relative energy percentage; calculate the relative energy percentage of each frequency band. The specific formula is as follows: ; In the formula, In this embodiment, the wavelet packet decomposition level is [number]. .

[0045] Wavelet packet energy entropy calculation: Based on information entropy theory, calculate the wavelet packet energy entropy of this window. : ; Agreement hour, .

[0046] It should be noted that, based on information entropy theory, entropy characterizes the degree of dispersion / disorder in the energy distribution of a signal within a window: the more concentrated the energy is in a few frequency bands, the lower the entropy value, and the more ordered the signal; the more dispersed the energy is across multiple frequency bands, the higher the entropy value, and the more complex and disordered the signal. When a wire breaks in the steel rope, the time-frequency structure of the signal changes abruptly, the energy distribution dispersion increases, and the entropy value changes accordingly. (Convention) hour, This is a general mathematical convention for calculating information entropy: since the logarithm of 0 has no mathematical definition, but according to the property of limits, hour Therefore, this convention is made to ensure that the calculation is legal and does not affect the physical meaning of the entropy value.

[0047] Step S4: Construction of Single-Mode DIH Damage Index: Standard samples of healthy steel wire ropes under multiple working conditions are collected in advance, and the average health benchmark of the relative energy ratio of each frequency band is statistically obtained; for each current sliding window, the difference between the relative energy ratio of each frequency band and the average health benchmark is calculated, and then the standard deviation of the differences of all frequency bands is calculated to obtain the single-mode DIH damage index value corresponding to the window, and the DIH curve continuously distributed along the rope length is output. Step S4 specifically includes the following sub-steps: Step S41: Construction of a multi-condition health benchmark library: Collect standard samples of healthy steel wire ropes at various operating speeds and ambient temperatures, and obtain energy entropy data for each frequency band under each condition through a processing procedure that is completely consistent with steps S1 to S3; take the statistical mean of the energy entropy of the same frequency band under all conditions, and use it as the health benchmark mean of the corresponding frequency band to form a standardized health benchmark library. In practice, multiple sets of raw signals from healthy steel wire ropes covering the commonly used elevator operating speed range and the ambient temperature range are collected. For each set of healthy samples, steps S1 (spatial domain transformation), S2 (preprocessing), and S3 (wavelet packet decomposition and energy entropy calculation) are performed to obtain a 64-dimensional frequency band energy entropy vector for each sample. The arithmetic mean of the energy entropy under all operating conditions is calculated for each frequency band and used as the standardized health benchmark for that frequency band. The average values ​​of the health benchmarks for all frequency bands are combined into a benchmark vector and stored in the health benchmark library for subsequent detection.

[0048] Step S42, Calculation of energy entropy difference in single window frequency band: For the single sliding window to be detected, extract the energy entropy value of each frequency band in the window, and subtract it from the mean of the health benchmark of the corresponding frequency band in the health benchmark library one by one to obtain the energy entropy difference sequence of each frequency band; In practice, the input is the 64-dimensional frequency band energy entropy vector of the current sliding window and the benchmark mean vector in the health benchmark library; the frequency band numbers are matched one by one, and subtraction is performed bit by bit to obtain the deviation value of each frequency band relative to the health state; the output is a 64-dimensional energy entropy difference sequence, where positive values ​​represent an increase in the energy entropy of the frequency band and negative values ​​represent a decrease, together reflecting the time-frequency distribution changes caused by damage.

[0049] Step S43: Calculate the single-window DIH damage index: Calculate the global average value of the energy entropy difference of all frequency bands in the current window, and then use the global average value as the benchmark to calculate the unbiased standard deviation of the energy entropy difference of all frequency bands. Use the unbiased standard deviation as the single-mode DIH damage index value corresponding to the current window. In practice, the energy entropy difference sequence of the current window is input. First, the arithmetic mean of the differences of all frequency bands is calculated to obtain the global mean. Then, the square of the deviation between the difference and the global mean is calculated for each frequency band, and all the squared deviation values ​​are accumulated. The accumulated result is divided by the number of frequency bands minus one (degree of freedom correction) and the square root is taken to obtain the unbiased standard deviation, which is the DIH index value of the current window. The single-window DIH scalar value is output. The larger the value, the higher the dispersion of the frequency band energy entropy distribution, and the greater the probability of damage at the corresponding location.

[0050] Step S44: Continuous DIH curve generation: Calculate the DIH index of all sliding windows sequentially along the axial direction of the wire rope, and splice the DIH values ​​of each window in an orderly manner according to the axial position corresponding to its center to form a single-mode DIH curve continuously distributed along the rope length. Output the leakage magnetic mode DIH curve and the visual mode DIH curve respectively. In practice, the DIH value of all sliding windows is calculated sequentially along the axial direction of the wire rope, and the axial position corresponding to the center of each window is recorded. All DIH values ​​are sorted and spliced ​​according to the axial position from smallest to largest to form a one-dimensional continuous curve. The leakage magnetic mode DIH curve and the visual mode DIH curve are output respectively. The horizontal axis of the curve is the axial position of the wire rope, and the vertical axis is the DIH index value.

[0051] In this embodiment of the invention, a health benchmark library is first constructed: standard samples of healthy steel wire ropes are collected at three operating speeds of 0.5 m / s, 1.0 m / s, and 1.6 m / s, and at three ambient temperatures of 10℃, 25℃, and 40℃, and the statistical mean of the energy entropy of 64 frequency bands is calculated respectively. This serves as a baseline average for health.

[0052] For the current m-th sliding window, calculate the difference between the relative energy percentage of each frequency band and the healthy baseline: ; Calculate the global mean of the differences across all frequency bands: ; The final DIH damage index is defined as the unbiased standard deviation of the difference: ; After continuous calculation along the rope length, continuously distributed leakage magnetic flux density (DIH) curves and visual DIH curves are obtained. When a wire breaks at a certain position in the wire rope, the time-frequency distribution of the signal at that position changes abruptly, the dispersion of the relative energy proportion of each frequency band increases significantly, and the DIH curve forms a distinct peak. The peak position corresponds to the axial position of the broken wire, and the peak amplitude corresponds to the severity of the damage.

[0053] Step S5, Adaptive Weighted Fusion of Dual-Modal DIH: Calculate the real-time confidence scores of the magnetic flux leakage mode and the visual mode within the current window respectively; calculate the corresponding dynamic weights based on the two confidence scores, and perform weighted fusion of the two DIH indices to obtain the fused DIH index; Step S5 specifically includes the following sub-steps: Step S51: Extraction of single-mode operating condition feature factors: For the magnetic flux leakage mode, extract two operating condition features: signal-to-noise ratio and baseline fluctuation coefficient within the current window; for the visual mode, extract two operating condition features: oil stain occlusion area ratio and image clarity within the current window. In practice, the preprocessed magnetic flux leakage signal and visual image within the current window, along with the corresponding single-mode DIH calculation results, are input. For the magnetic flux leakage mode, the ratio of the signal peak value to the root mean square of the background noise within the window is calculated to obtain the signal-to-noise ratio (SNR). The ratio of the maximum fluctuation of the fitted baseline within the window to the signal amplitude is calculated to obtain the baseline fluctuation coefficient. For the visual mode, oil-stained areas are identified through color space segmentation, and the proportion of oil-stained pixels to the total pixels in the window is statistically analyzed to obtain the proportion of oil-stained occlusion area. The gradient variance of the image grayscale is calculated to obtain the image sharpness index. Four operating condition feature factors are output and sent to the subsequent comprehensive confidence calculation stage.

[0054] Step S52, Single-mode comprehensive confidence calculation: Based on the two working condition features of the magnetic flux leakage mode, the real-time comprehensive confidence of the magnetic flux leakage mode is calculated; based on the two working condition features of the visual mode, the real-time comprehensive confidence of the visual mode is calculated; the confidence values ​​of both channels are in the range of 0 to 1, and the higher the value, the stronger the reliability of the detection data of the corresponding mode. In practical implementation, the four operating condition characteristic factors output by S51 are input; the signal-to-noise ratio (SNR) term uses a normalized mapping to compress the SNR values ​​from 0 to infinity to a minimum. The interval is then multiplied by the baseline stability coefficient; the greater the baseline fluctuation, the lower the confidence level. Multiplying these two values ​​together yields the overall confidence level for magnetic flux leakage. The oil contamination percentage is mapped using the complement; the more oil contamination, the lower the confidence level. Sharpness is assessed using a relative ratio; the blurrier the image, the lower the confidence level. Multiplying these two values ​​together yields the overall visual confidence level. Two output channels are provided. The overall confidence level of the interval is as follows: the higher the confidence level, the more reliable the current data for that modality and the higher the reference value of the detection results.

[0055] Step S53, Adaptive Dynamic Fusion Weight Generation: Sharpen the overall confidence of the two modes to improve the distinction between high and low confidence; then normalize the sharpened confidence to obtain the fusion weights of the magnetic leakage mode and the visual mode respectively, and the sum of the two weights is 1. In practice, the combined confidence level of the two modes is input. Take the confidence scores for the two paths respectively. The process of sharpening is performed by amplifying the difference between high and low confidence levels, causing the weights of modalities with severe interference and low confidence levels to decrease rapidly, thus preventing them from degrading the fusion result. The sharpened confidence levels are then normalized and divided by the sum of the two values ​​to obtain the final fusion weights. The two normalized weights are then output and fed into the subsequent weighted fusion stage.

[0056] Step S54, weighted fusion of dual-modal DIH indices: according to their respective fusion weights, the magnetic leakage modal DIH index and the visual modal DIH index are weighted and summed to obtain the final fused DIH index, which is then output to the subsequent damage identification stage. In practice, the two single-mode DIH values ​​and their corresponding fusion weights are input; the two DIH values ​​are weighted and summed according to their weights to obtain the fusion DIH index; the weighted calculation of all windows is completed sequentially along the wire rope axis to generate a continuously distributed fusion DIH curve; the fusion DIH curve is output to step S6 for damage determination, location and quantitative assessment.

[0057] In the embodiments of the present invention, such as Figure 5 As shown, the real-time confidence scores of the two modes are first calculated, with values ​​ranging from [value range missing]. .

[0058] Confidence of leakage magnetic mode It is determined by both the signal-to-noise ratio within the window and the baseline fluctuation coefficient: ; In the formula, The signal-to-noise ratio of the leakage magnetic signal. Baseline fluctuation coefficient ( The higher the signal-to-noise ratio and the more stable the baseline, the higher the confidence level.

[0059] Visual modal confidence The image sharpness is determined by both the percentage of the area obscured by oil stains within the window and the image clarity. ; In the formula, The percentage of the area covered by oil stains ( ), Image gradient variance (a sharpness index). This represents the standard sharpness threshold. The less occlusion, the sharper the image, and the higher the confidence level.

[0060] The two confidence scores are then squared-sharpened and normalized to obtain the fusion weights: ; In the formula, The weighted sharpening coefficient is 2; and When a certain mode is severely disturbed and its confidence level decreases, its fusion weight is automatically reduced.

[0061] The final DIH metric is: .

[0062] Please see Figure 7 As shown, step S6, damage identification and quantitative assessment: Based on the statistical results of healthy samples, a damage judgment threshold is set. When the fused DIH index exceeds the threshold, it is determined that there is a broken wire at the corresponding position. The axial position corresponding to the DIH peak point is taken as the precise location of the broken wire. Combining the fused DIH peak and the physical characteristics of the two modes, the number of broken wires and the depth of broken wires are calculated through a pre-calibrated regression model to complete the damage degree classification. Step S6 specifically includes the following sub-steps: Step S61, Damage Judgment Threshold Setting and Initial Screening of Candidate Broken Wires: Based on the fusion DIH statistical data of healthy samples, the damage judgment threshold is determined using the three-standard-deviation principle; all continuous segments on the fusion DIH curve that exceed the judgment threshold are marked as candidate broken wire segments, and the start and end axial positions of each segment are recorded. It should be noted that the formula for calculating the damage assessment threshold is as follows: ; In the formula, The threshold for determining DIH damage is set at the fusion level. The statistical mean of the DIH index for the healthy sample; The statistical standard deviation of the DIH index for the healthy sample; The input is a fused DIH sequence obtained from a complete process of inputting healthy samples under multiple operating conditions. The DIH value distribution of all healthy samples is statistically analyzed. The arithmetic mean and standard deviation of the DIH values ​​of healthy samples are calculated, and a judgment threshold is calculated based on the three-times-standard-deviation principle. This threshold corresponds to the upper limit of the 99.7% value distribution under healthy conditions. Exceeding the threshold can be judged as abnormal damage. The fused DIH curve to be detected is compared with the judgment threshold point by point. Signal segments that continuously exceed the threshold are marked as candidate broken wire segments. The start position, end position and segment length of each segment are recorded. The set of all candidate broken wire segments is output and sent to the subsequent precise positioning stage.

[0063] Step S62, Precise axial positioning of broken wire: Within each candidate broken wire segment, search for the maximum value of the fused DIH index as the peak point, determine the axial position corresponding to the peak point as the precise axial position of the broken wire, and record the fused DIH value corresponding to the peak. In practice, the fused DIH sequence and corresponding axial position coordinates of a single candidate broken wire segment are input; the maximum value of the DIH value is searched within the segment range, and the window number and axial position corresponding to the point are recorded; the axial position of the peak point is used as the precise location of the broken wire, and the peak DIH value is recorded as the core input parameter for damage quantification; the precise location of the broken wire and the peak DIH value are output and sent to the damage quantification process.

[0064] Step S63, Multi-feature Quantitative Calculation of Damage Degree: Extract four physical features corresponding to the current broken wire location: peak amplitude of leakage magnetic signal, half-width of peak leakage magnetic signal, area of ​​visual broken wire region, and contrast of visual broken wire region. Combine the real-time fusion weights of DIH peak value and two-channel mode, and calculate the quantitative values ​​of the number of broken wires and the depth of broken wires respectively through a pre-calibrated multivariate regression model. In practice, the input consists of the fused DIH value corresponding to the peak position of the broken wire, the fusion weights of the two modes, and the physical and visual physical features of the leakage magnetic field at the corresponding position. The physical features of the two modes are multiplied by their respective fusion weights: the modal features with higher confidence have larger weights and contribute more to the quantization results; the modal features with lower confidence have smaller weights to avoid interfering with the quantization accuracy. The weighted physical features and the fused DIH peak are used as inputs and substituted into a pre-calibrated multivariate regression model to calculate the quantization results of the number of broken wires and the depth of broken wires. The quantized values ​​of the number of broken wires and the depth of broken wires are output and sent to the damage grading stage.

[0065] Step S64, Damage Level Classification and Graded Early Warning Output: Based on the calculated number of broken wires and the depth of broken wires, the broken wire damage is classified into three levels: minor, moderate, and severe according to the preset grading rules; corresponding to different damage levels, graded early warning signals are output for recording and archiving, prompting maintenance, and immediate shutdown.

[0066] In this embodiment of the invention, based on the statistical results of healthy samples, three times the maximum value of the fused DIH in the healthy state is taken as the damage determination threshold. .when When the corresponding position is determined to be damaged by wire breakage, the axial position corresponding to the DIH peak point is taken as the precise location of the wire breakage.

[0067] By combining features such as the peak value of the DIH signal, the peak amplitude of the leakage magnetic field signal, and the area of ​​the visually broken wire region, the number of broken wires n and the broken wire depth d are quantitatively calculated using a pre-calibrated multiple regression model. ; ; In the formula, This represents the peak amplitude of the leakage magnetic signal. The half-width at half-peak of leakage flux is 0.5%. The pixel area of ​​the visual broken wire region. Contrast of the broken wire area; These are the regression coefficients obtained through calibration experiments; like Figure 6 As shown, the horizontal dashed line represents the damage assessment threshold (0.30, based on the three-standard-deviation principle). If a segment continuously exceeds the threshold, it is marked as a candidate wire breakage segment; the peak point (at 200mm) within the segment is marked as the precise location of the wire breakage.

[0068] It should be noted that damage is typically classified into three levels based on the number and depth of broken filaments: Minor damage: a single broken wire with a depth less than 1 / 3 of the wire diameter; record and archive. General injuries: Root severance or depth at The diameter of the steel wire indicates an upcoming maintenance check. Severe damage: Four or more broken wires or a depth greater than 1 / 2 the diameter of the steel wire will immediately trigger a shutdown warning.

[0069] Specifically, the classification standard for broken wire damage is as follows: Step S7, Spatiotemporal Continuity Verification and False Detection Removal: Based on the spatiotemporal continuity of the wire rope movement, the spatial continuity and temporal repeatability of the candidate broken wire positions are verified, isolated single-point peaks and non-repeating interference signals are removed, and the final identification result is output. Step S7 specifically includes the following sub-steps: Step S71, Spatial continuity verification of candidate broken wires: For each candidate broken wire segment obtained from the initial screening, count the number of windows in the segment whose fused DIH value continuously exceeds the judgment threshold, and at the same time verify whether the DIH curve in the segment presents a single-peak smooth convex shape; judge the candidate segments with fewer than the set threshold or whose shape does not conform to the smooth convex feature as random noise and remove them, and retain the candidate broken wires that conform to the spatial continuity feature. In specific implementation, input all candidate broken wire segment data output in step S6. Each segment contains an axial position sequence and a corresponding fused DIH value sequence. Perform the first filtering: count the total number of windows that continuously exceed the judgment threshold for each segment, compare it with the minimum continuous window threshold, and directly judge isolated single-point segments with less than the threshold as random noise and remove them. Perform the second filtering: for the retained segments, calculate the second difference of the DIH curve point by point, verify the consistency of the difference sign on both sides of the peak point, eliminate the interference of multi-peak spikes, and retain only candidate broken wires with a smooth single-peak shape. Output the set of candidate broken wires that have passed the spatial double verification and send it to the subsequent time repeatability verification stage.

[0070] It should be noted that for a single candidate broken wire segment, the statistical satisfaction is... Total number of consecutive sliding windows Set minimum continuous window threshold (In this scheme, we take 3), if This is then identified as isolated noise and removed. The smoothness of the bulge shape is determined by the second-order difference of the DIH curve; the formula for calculating the second-order difference is: ; In the formula, For the first The second difference of the DIH curve at each window; if the second difference of all points to the left of the peak point... Second difference of all points on the right If a peak is found to be a smooth single peak that conforms to physical characteristics, it is considered a smooth single peak. If multiple sharp peaks with alternating positive and negative values ​​appear, they are considered to be burr interference and are removed.

[0071] Step S72, Multi-round time repeatability matching: Collect the full rope length detection data of the elevator running back and forth multiple times, and match the remaining candidate broken wires according to the axial position in the detection results of different rounds. Calculate the number of times the candidate broken wires at the same position are repeated and the peak feature similarity. In practice, the system inputs the full rope length detection results from at least three round trips of the elevator, as well as the candidate broken wire set that passed spatial verification. Using the candidate broken wire positions from the first round of detection as a benchmark, it searches for candidate peak values ​​with a positional deviation of less than 10mm in the detection results of each subsequent round. For peak value pairs that successfully match positions, it calculates the peak feature similarity; if the similarity is greater than 0.7, it is determined to be a duplicate occurrence of the same broken wire. After traversing all rounds, it counts the total number of times each candidate broken wire appears repeatedly in all detection rounds, while simultaneously recording the corresponding position and peak data for each round. It outputs the number of times each candidate broken wire appears repeatedly, the multi-round position set, and the multi-round peak value set, and sends them to the final judgment stage.

[0072] Step S73, Real Damage Judgment and Result Output: Candidate broken wires that have repeated occurrences to the set requirement and whose position deviation is less than the set threshold are judged as real broken wires; candidate broken wires that only occur once are judged as transient interference and eliminated; finally, the axial position of the broken wire, damage quantification parameters and damage level results after double verification are output. In practice, the system inputs the number of repetitions, multiple rounds of position data, and multiple rounds of peak data for each candidate broken wire. A final judgment is made based on the set repetition threshold and position deviation threshold. Wires meeting the criteria are marked as genuine broken wires, while those not meeting the criteria are marked as interference and removed. For samples determined to be genuine broken wires, the average position and average peak value from multiple rounds of detection are calculated to reduce the impact of random fluctuations in a single detection and improve result stability. Combining the damage quantification model from step S6, the final number of broken wires and the depth of broken wires are calculated based on the average peak value to complete the damage level determination. The final identification result is output, including the axial position of the broken wire, damage quantification parameters, and damage level, and is simultaneously pushed to the operation and maintenance system and the early warning module to trigger the corresponding level of early warning action.

[0073] In this embodiment of the invention, the actual broken wire is a fixed defect on the steel wire rope, and the corresponding DIH peak value will show a smooth convex shape in multiple consecutive windows; if only a single window has a peak value and the adjacent windows before and after are all below the threshold, it is determined to be random noise and is removed.

[0074] During the elevator's round trip operation, the same steel wire rope will pass through the detection point multiple times. If the DIH peak value at the same axial position appears repeatedly in at least two independent runs, and the positional deviation is less than 10mm, it is determined to be real damage. A single peak value is determined to be transient interference and is eliminated.

[0075] like Figure 6 As shown, in this embodiment, for a single micro-broken wire with a depth of 1 / 4 of the diameter, the single magnetic flux leakage DIH peak is not obvious and close to the threshold, while the single visual DIH fluctuates due to local oil contamination. After fusion by the present invention, the DIH peak is significantly improved, clearly exceeding the judgment threshold, and there are no additional false detection peaks, verifying the effectiveness of the method.

[0076] Tests showed that the solution in this embodiment achieved a detection rate of 98.5% for single micro-broken wires, an identification accuracy of 96.2% under conditions where the oil stain coverage area was 30%, a false detection rate of 1.2% throughout the process, and a broken wire positioning error of ≤2mm, meeting the engineering requirements for elevator on-site inspection.

[0077] See Figure 2 As shown, this invention is an elevator wire rope broken wire damage identification system, which can be used to execute the method described in this invention, including: The multimodal synchronous acquisition and spatial domain conversion unit is used to synchronously acquire the leakage magnetic field detection signal and surface visual image of the wire rope by using the output pulse of the rotary encoder at the traction wheel as a global trigger source; and based on the encoder pulse count, it converts the two signals acquired in the time domain into an axial spatial equally spaced sampling sequence, so that the spatial resolution of the two signals is completely consistent and the spatial position is accurately aligned. The dual-branch signal preprocessing unit is used to perform adaptive baseline correction on the leakage magnetic signal. It uses truncated Gaussian weighted polynomial fitting to extract and remove the baseline, eliminating the slow baseline changes caused by operating jitter and sensor zero-point drift. At the same time, it performs spiral texture unfolding processing on the visual image, fits the spiral trajectory of the steel wire and performs vertical offset compensation on the pixels to eliminate the periodic gray-level fluctuations caused by the inherent spiral texture and extracts the gray-level sequence distributed along the axis. The sliding window wavelet packet decomposition and energy entropy calculation unit is used to set a sliding window whose length covers twice the wire rope lay length, and slides continuously along the axial spatial sequence with a fixed step size; multi-level wavelet packet decomposition is performed on the leakage magnetic sequence and grayscale sequence in each window to obtain multiple equal bandwidth frequency band components; the signal energy and relative energy ratio of each frequency band in each window are calculated in turn, and finally the wavelet packet energy entropy corresponding to each window is obtained. The single-mode DIH damage index generation unit is used to statistically obtain the health benchmark mean of the relative energy ratio of each frequency band based on multiple sets of standard samples of healthy steel wire ropes under different operating speeds and ambient temperatures. For each current sliding window, the difference between the relative energy ratio of each frequency band and the health benchmark mean is calculated, and then the unbiased standard deviation of the differences of all frequency bands is calculated to obtain the single-mode DIH damage index value of the corresponding window, and output the DIH curve that is continuously distributed along the rope length. The dual-modal DIH adaptive fusion unit is used to calculate the real-time confidence scores of the magnetic flux leakage mode and the visual mode within the current window, respectively. The confidence score of the magnetic flux leakage mode is determined by the signal-to-noise ratio and baseline fluctuation within the window, while the confidence score of the visual mode is determined by the proportion of oil smudge occlusion area and image clarity within the window. Based on the two confidence scores, corresponding dynamic weights are calculated, and the two DIH indices are weighted and fused to obtain the fused DIH index. The damage identification and quantitative assessment unit is used to set damage judgment thresholds based on the statistical results of healthy samples. When the fused DIH index exceeds the threshold, it is determined that there is filament breakage damage at the corresponding location. The axial position corresponding to the DIH peak point is used as the precise location of the filament breakage. Combining the fused DIH peak value and the physical characteristics of the two modes, the number of filament breakages and the filament breakage depth are calculated through a pre-calibrated multivariate regression model to complete the damage degree classification and early warning output. The spatiotemporal continuity verification and result output unit is used to perform spatial continuity verification and temporal repeatability verification on candidate broken wire positions based on the spatiotemporal continuity of the wire rope motion, eliminate isolated single-point peaks and non-repeating interference signals, and output the final broken wire identification result.

[0078] It is worth noting that in the above system embodiments, the various units are divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0079] Furthermore, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0080] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for identifying broken wire damage in elevator wire ropes, characterized in that, Includes the following steps: Simultaneously acquire the magnetic leakage signal and visual image of the wire rope, and convert the magnetic leakage signal and visual image into a spatially equally spaced sampling sequence corresponding to the axial position of the wire rope; Adaptive baseline correction is performed on the magnetic flux leakage signal, and spiral texture unfolding is performed on the visual image to obtain the magnetic flux leakage feature sequence and the visual feature sequence. The spatially equally spaced sampling sequence is divided into sliding windows with a preset window length and step size. Wavelet packet decomposition is performed on the magnetic flux leakage feature sequence and visual feature sequence in each window to obtain multiple equal bandwidth frequency band components of the corresponding layer. The signal energy and relative energy ratio of each frequency band within each window are calculated sequentially to form the frequency band energy distribution vector corresponding to each window. At the same time, the wavelet packet energy entropy of each window is calculated based on the relative energy ratio of each frequency band as an auxiliary indicator. Based on healthy steel wire rope samples, a benchmark value for the relative energy ratio of each frequency band is established. The offset between the relative energy ratio of each frequency band in the current window and the benchmark value, as well as the standard deviation of all offsets, are calculated. The single-mode DIH damage index values ​​of the leakage magnetic mode and visual mode corresponding to the window are generated. Calculate the real-time confidence scores of the magnetic flux leakage mode and the visual mode within the current window respectively; calculate the corresponding dynamic weights based on the two confidence scores, and perform weighted fusion of the two DIH indices to obtain the fused DIH index; A damage assessment threshold is set. When the fusion DIH index corresponding to the sliding window exceeds the damage assessment threshold, it is determined that there is a broken wire damage at the corresponding position of the window. Continuous windows that exceed the damage assessment threshold are identified as candidate broken wire segments. Within the candidate broken wire segments, the fusion DIH index values ​​corresponding to each window are compared. It is determined that there is a broken wire position within the sliding window with the largest fusion DIH index. Combining the fusion DIH index corresponding to the window where the broken wire position is located with the physical characteristics of the two modes, the number of broken wires and the broken wire depth are calculated through a pre-calibrated regression model to complete the damage degree classification and output the final identification result.

2. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, The spatially equidistant sampling sequence uses the pulses output by the traction wheel rotary encoder as the global synchronization trigger source and the reference zero point of the traction wheel rotary encoder as the starting position. Based on the pulse count at each sampling moment, the diameter of the traction wheel, and the number of pulses per revolution of the encoder, the axial position of the wire rope corresponding to the sampling point is calculated. Through interpolation resampling, the two signals are unified into a spatially equidistant sequence with the same interval, so as to achieve precise alignment of spatial positions.

3. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, The adaptive baseline correction specifically involves: fitting the baseline of the magnetic flux leakage signal using a fifth-order polynomial; introducing a truncated Gaussian weighting function during the fitting process; assigning zero weight to sampling points whose signal amplitude deviates from the window mean by more than two standard deviations; and subtracting the fitted baseline from the original signal to obtain the baseline-corrected stable magnetic flux leakage signal.

4. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, The specific process for obtaining the wavelet packet energy entropy corresponding to each window includes: Set the window length and sliding step size, and divide the preprocessed magnetic flux leakage sequence and grayscale sequence into continuous windows along the axial spatial sequence to generate multiple sets of overlapping sub-signal segments. The db8 wavelet is selected as the wavelet basis function. Wavelet packet decomposition is performed on the leakage sub-signal and gray sub-signal in each sliding window to decompose the single signal into multiple frequency band components with equal bandwidth, and the reconstructed sub-signal corresponding to each frequency band is output. Energy statistics are performed on the reconstructed sub-signals of each frequency band, and the sum of the squares of the signal amplitudes is taken as the total energy of the frequency band. The total energy of the window is obtained by summing the energy of all frequency bands within a single window, and then the energy of each frequency band is divided by the total energy to obtain the relative energy ratio of each frequency band. The wavelet packet energy entropy value corresponding to the window is calculated based on the relative energy proportion of each frequency band within the window. After calculating all sliding windows sequentially along the axial direction, the leakage magnetic energy entropy sequence and the visual energy entropy sequence, which are continuously distributed along the rope length, are output.

5. The method for identifying broken wire damage in elevator wire ropes according to claim 4, characterized in that, The specific calculation steps for the wavelet packet energy entropy include: Energy statistics are performed on each frequency band reconstructed sub-signal within each sliding window, and the sum of squares of the amplitudes of each reconstructed sub-signal is used to characterize the signal energy of the corresponding frequency band. The total energy of the signals in each frequency band within the same window is normalized to obtain the relative energy of each frequency band. Based on the relative energy, the wavelet packet energy entropy corresponding to the window is calculated in the form of information entropy. When the relative energy percentage of a certain frequency band is zero, the entropy contribution of that frequency band is ignored.

6. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, The DIH curve generation process is as follows: Standard samples of healthy steel wire ropes were collected at various operating speeds and ambient temperatures to obtain energy entropy data for each frequency band under various operating conditions. The statistical mean of the energy entropy of the same frequency band under all operating conditions was taken as the health benchmark mean of the corresponding frequency band, forming a standardized health benchmark library. For a single sliding window to be detected, extract the energy entropy value of each frequency band within the window, and subtract it from the mean of the health benchmark for the corresponding frequency band in the health benchmark library one by one to obtain the energy entropy difference sequence of each frequency band. Calculate the global average value of the energy entropy difference of all frequency bands in the current window, and then use the global average value as a benchmark to calculate the unbiased standard deviation of the energy entropy difference of all frequency bands. Use the unbiased standard deviation as the single-mode DIH damage index value corresponding to the current window. The DIH index of all sliding windows is calculated sequentially along the axial direction of the wire rope. The DIH values ​​of each window are then sequentially spliced ​​together according to the axial position corresponding to their center to form a single-mode DIH curve that is continuously distributed along the rope length. The leakage magnetic mode DIH curve and the visual mode DIH curve are output respectively.

7. The method for identifying broken wire damage in elevator wire ropes according to claim 5, characterized in that, The calculation steps for the single-modal DIH damage index include: Calculate the deviation of the energy entropy of each frequency band within each window from the mean of the health baseline; The average deviation value is calculated for each frequency band deviation; Calculate the degree of dispersion of each frequency band deviation relative to the average deviation value, and use the square root of the degree of dispersion as the single-mode DIH damage index value for that window. The DIH curve is obtained by mapping the single-mode DIH damage index corresponding to each window according to the axial position.

8. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, The steps for obtaining the fused DIH index include: For the magnetic flux leakage mode, two operating condition features are extracted: signal-to-noise ratio and baseline fluctuation coefficient within the current window. For the visual mode, two operating condition features are extracted: the proportion of oil smudge occlusion area and image clarity within the current window. Based on two operating condition characteristics of the leakage magnetic field mode, the real-time comprehensive confidence level of the leakage magnetic field mode is calculated; based on two operating condition characteristics of the visual mode, the real-time comprehensive confidence level of the visual mode is calculated. The combined confidence scores of the two modes are sharpened, and then the sharpened confidence scores are normalized to obtain the fusion weights of the magnetic leakage mode and the visual mode, respectively. The sum of the two weights is 1. The leakage magnetic mode DIH index and the visual mode DIH index are weighted and summed according to their respective fusion weights to obtain the final fused DIH index.

9. The method for identifying broken wire damage in elevator wire ropes according to claim 1, characterized in that, All continuous segments on the DIH curve that exceed the judgment threshold are marked as candidate wire breakage segments, and the start and end axial positions of each segment are recorded. Within each candidate wire breakage segment, the maximum value of the fused DIH index is searched as the peak point, and the axial position corresponding to the peak point is determined as the precise axial position of the wire breakage. The fused DIH value corresponding to the peak value is recorded, and four physical features corresponding to the current wire breakage position are extracted: peak amplitude of leakage magnetic signal, half-width of leakage magnetic peak, area of ​​visual wire breakage region, and contrast of visual wire breakage region. Combined with the fused DIH peak value and the real-time fusion weight of the two modes, the quantitative values ​​of the number of wire breakages and the depth of wire breakage are calculated respectively through a pre-calibrated multivariate regression model.

10. A system for identifying broken wire damage in elevator wire ropes, characterized in that, include: The multimodal synchronous acquisition and spatial domain conversion unit includes a leakage magnetic field sensor array, a linear industrial camera, a rotary encoder, and a synchronous trigger controller. It is used to synchronously acquire the leakage magnetic field detection signal and surface visual image of the wire rope by using the output pulse of the rotary encoder at the traction wheel as a global trigger source; and converts the two signals acquired in the time domain into an axial spatial equally spaced sampling sequence based on the encoder pulse count. The dual-branch signal preprocessing unit is used to perform adaptive baseline correction on the leakage magnetic signal, perform spiral texture unfolding processing on the visual image, fit the spiral trajectory of the steel wire and perform vertical offset compensation on the pixels, and extract the gray-scale sequence distributed along the axis. The sliding window wavelet packet decomposition and energy entropy calculation unit is used to perform multi-level wavelet packet decomposition on the leakage magnetic sequence and grayscale sequence in each window to obtain multiple equal bandwidth frequency band components. The signal energy and relative energy ratio of each frequency band in each window are calculated in turn to obtain the wavelet packet energy entropy corresponding to each window. The single-mode DIH damage index generation unit is used to calculate the difference between the relative energy ratio of each frequency band and the mean of the health baseline for each current sliding window, and then calculate the unbiased standard deviation of the differences of all frequency bands to obtain the single-mode DIH damage index value of the corresponding window, and output the DIH curve that is continuously distributed along the rope length. The dual-modal DIH adaptive fusion unit is used to calculate the real-time confidence of the magnetic leakage mode and the visual mode within the current window, respectively. Based on the two confidences, the corresponding dynamic weights are calculated, and the two DIH indices are weighted and fused to obtain the fused DIH index. The damage identification and quantitative assessment unit is used to determine the presence of broken wires at the corresponding location when the fused DIH index exceeds the threshold. The axial position corresponding to the DIH peak point is used as the precise location of the broken wire. Combining the fused DIH peak value and the physical characteristics of the two modes, the number of broken wires and the depth of broken wires are calculated through a pre-calibrated multivariate regression model to complete the damage level classification and early warning output.

Images