Elevator steel belt damage detection and quantitative analysis system based on eddy current and magnetic flux leakage dual-mode detection

Abstract

The invention discloses an elevator steel belt damage detection and quantitative analysis system based on eddy current and magnetic flux leakage dual-mode detection, and aims to solve the problems of single-mode information loss and serious industrial field strong noise interference in the existing steel belt detection. The system synchronously integrates an eddy current sensor, a magnetic flux leakage sensor and an encoder through a multi-probe array adapter; and multi-dimensional damage physical information in the steel strip is obtained. An improved wavelet packet transform-empirical mode decomposition (WPT-EMD) collaborative noise reduction algorithm is adopted, and in combination with a sub-band energy entropy and a self-attention mechanism, non-stationary mechanical noise is effectively filtered out. A multi-rule feature extraction engine is used for extracting smooth residual errors, derivative mutation and other features, a support vector machine (DE-SVM) model introducing physical priori knowledge weights is constructed, and the small sample recognition problem is solved in combination with a virtual sample generation technology. The method can realize high-precision positioning and quantitative evaluation of steel strip damage under complex working conditions, and has the characteristics of strong anti-interference capability, high identification accuracy and good generalization performance.

Description

Technical Field

[0001] This invention belongs to the field of industrial non-destructive testing and intelligent monitoring technology, specifically relating to an elevator steel belt damage detection and quantitative analysis system based on dual-mode detection of eddy current and leakage magnetic flux. Background Technology

[0002] As the core load-bearing component of traction elevators, the traction steel belt is highly susceptible to damage from broken wires or strands, which can easily lead to major safety accidents. Traditional methods for detecting steel belt damage mainly face the following technical challenges: Limited detection methods: Traditional systems often employ a single detection principle (such as eddy current or magnetic flux leakage). Eddy current testing is sensitive to surface cracks but has limited ability to detect subcutaneous defects; magnetic flux leakage testing can detect subcutaneous defects but its application to non-magnetic materials is limited. A single method cannot comprehensively identify diverse damage characteristics.

[0003] Severe noise interference: Industrial sites are subject to noise from electromagnetic interference, mechanical vibration, and material inhomogeneity. Traditional filtering methods, while suppressing noise, are prone to losing useful damage characteristic information.

[0004] Quantitative assessment is difficult: Most existing systems are limited to qualitative identification of damage and lack the ability to accurately quantify parameters such as damage size, depth, and shape, making it difficult to meet high-quality control standards.

[0005] Poor system adaptability: Traditional detection models have fixed parameters and cannot adaptively adjust according to changes in working conditions, resulting in unstable detection performance under different production conditions.

[0006] Inefficient feature fusion: The lack of an effective multi-dimensional feature extraction and fusion mechanism makes it difficult to extract feature parameters that truly reflect damage characteristics from complex signals, affecting the accuracy of identification.

[0007] Therefore, there is an urgent need for a system that can overcome the above-mentioned defects and achieve high-precision, high-reliability, and adaptive intelligent detection. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a steel strip damage detection and quantitative identification system. This system achieves accurate detection and quantitative assessment of steel strip damage through multi-sensor information fusion, advanced signal processing technology and intelligent identification algorithm.

[0009] To achieve the above objectives, this invention provides a system for detecting and quantifying elevator steel strip damage based on dual-mode detection of eddy current and leakage magnetic flux, comprising: The dual-mode steel strip internal damage detection module synchronously connects to an eddy current sensor, a magnetic flux leakage sensor, and an encoder via a multi-probe array adapter. The eddy current sensor is used to acquire detection information at different penetration depths, the magnetic flux leakage sensor is used to detect abnormal leakage magnetic fields caused by changes in the magnetic permeability of the steel strip, and the encoder is used to provide position information. All sensor data are synchronized through a synchronous acquisition circuit to ensure time consistency. The signal collaborative noise reduction module is connected to the dual-mode steel strip internal damage detection module and is used to perform time-frequency domain analysis and background noise filtering on the collected raw data, and output the noise-reduced signal. A multi-rule feature extraction engine, connected to the signal collaborative denoising module, is configured with a variety of signal waveform analysis rules to extract multi-dimensional damage feature parameters from the denoised signal. The physical constraint recognition module is connected to the multi-rule feature extraction engine and is used to construct a machine learning model that integrates physical prior knowledge, mapping the damage feature parameters to damage quantification recognition results.

[0010] Preferably, the signal collaborative noise reduction module specifically adopts an improved wavelet packet transform-empirical mode decomposition dual-mode collaborative noise reduction algorithm; the module is equipped with an adaptive thresholding unit based on subband energy entropy, which dynamically allocates hard thresholding function and soft thresholding function by calculating the energy entropy of WPT node coefficients, and integrates a self-attention mechanism to enhance the weight of key frequency band coefficients, and then performs secondary filtering of residual noise through EMD decomposition and Hilbert envelope spectrum analysis.

[0011] Preferably, the multi-rule feature extraction engine achieves feature extraction through the TRDS multi-rule damage feature extraction mechanism, specifically including the following rules: Smoothing residual threshold rule: Calculate the residual between the signal and the smoothed signal, and extract the maximum residual value as the first feature; Derivative mutation detection rule: Extract the maximum mutation amplitude of the first derivative of the absolute value of the signal as the second feature; Local energy detection rule: The ratio of signal energy to background noise energy within the detection window is calculated as the third feature.

[0012] Preferably, the physical constraint identification module specifically constructs an improved support vector machine model based on a physical parameter feature weight allocation algorithm; the algorithm differentiates the weights of each dimension of the support vector machine model by using predefined physical weight coefficients, and calculates the optimal weight ratio through differential evolution algorithm and cross-validation algorithm to form a physical constraint feature space.

[0013] Preferably, the dual-mode steel belt internal damage detection module includes: a main frame, a leakage magnetic field probe module, an incremental encoder, a dual-mode magnetic field excitation module, guide wheels, roller bearing limit wheels, and buckles; wherein, the main frame consists of an upper shell and a lower shell, respectively placed on both sides of the traction steel belt, and two sets of guide wheels are installed in the grooves at the left and right ends of the upper shell and the lower shell to control the axial movement of the steel belt, and four sets of roller bearing limit wheels are installed at the bottom of the upper shell and the lower shell to control the circumferential movement of the traction steel belt; rectangular grooves and two sets of limit threaded holes are respectively provided between the two sets of guide wheels installed on the upper shell and the lower shell for fixing leakage magnetic field probes. The magnetic probe module has an encoder rubber wheel and an incremental encoder mounted on one end of the upper housing via a movable bracket. The lower housing is symmetrically distributed with the upper housing and does not have an encoder. The latch is located in the middle of the upper and lower housings and is used to open and close the main frame. The dual-mode excitation module is installed between the leakage magnetic probe module and the rectangular groove. The dual-mode excitation module consists of three sets of high permeability U-shaped magnetic cores arranged side by side with equal spacing and multi-winding coils surrounding the magnetic arms on both sides. The leakage magnetic probe module consists of a probe bracket and a Hall sensor array acquisition board mounted on it. The multi-winding coils are symmetrically distributed on both sides of the Hall sensor array acquisition board.

[0014] Preferably, Docker containerization is used for deployment, with each functional microservice running independently in an isolated container and communicating via a RESTful API; the device access middleware supports hot-swapping, so there is no need to restart the system when a new device is connected.

[0015] Preferably, it also includes a visual diagnostic interface, which integrates DE-SVM multi-rule feature parallel coordinate visualization, signal comparison display before and after noise reduction, damage pulse localization display, and feature dimension filtering and damage inversion analysis driven by touch gestures, forming an interactive diagnostic workbench.

[0016] Preferably, it also includes a self-optimizing learning module: configured with a dual-loop update mechanism, the short-term learning loop incrementally updates the DE-SVM model parameters through an online sequence learning algorithm; the long-term reconstruction loop performs full model retraining every 24 hours based on accumulated data; the module has a feature contribution evaluation unit based on Shapley value, which dynamically adjusts the TRDS rule weights and DE-SVM parameters to achieve continuous optimization of model performance.

[0017] Preferably, it also includes an extensible interface module, which adopts a microservice architecture and includes a device access middleware that supports both Modbus / TCP and OPC-UA protocols, an inference acceleration interface that integrates the OpenVINO™ toolkit, a vibration data exchange interface that conforms to the ISO 18436-4 standard, and a blockchain-based test report storage unit to achieve reliable traceability and auditing of the testing process.

[0018] Preferably, it also includes a data augmentation and optimization module for improving the performance of the recognition model, wherein the data augmentation and optimization module includes: A high-fidelity feature fusion unit is located in the multi-rule feature extraction engine. Based on the signal-to-noise ratio, it performs dynamic confidence weight allocation and multi-strategy optimization fusion on the features extracted from eddy current and leakage magnetic signals, and outputs high-fidelity fused feature parameters. The virtual sample generation unit, located in the physical constraint recognition module, expands the training sample set composed of the high-fidelity fused feature parameters through Gaussian process regression, noise injection, and feature space interpolation techniques to optimize the model's generalization ability under small sample conditions.

[0019] The proposed solution overcomes the shortcomings of existing technologies by using multi-sensor information fusion, advanced signal processing technology, and intelligent recognition algorithms to achieve multi-modal fusion, anti-interference, and high-precision quantification adaptive intelligent detection under all working conditions. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the dual-mode steel strip internal damage detection module of the present invention; Figure 2 This is a three-dimensional exploded view of the dual-mode steel strip internal damage detection module of the present invention; Figure 3 This is a schematic diagram of the steel strip damage detection and quantification system architecture of the present invention; Figure 4 This is a schematic diagram showing the installation location of the dual-mode steel strip internal damage detection module of the present invention in an elevator shaft.

[0021] The components include: 1. Upper housing; 2. Encoder bracket; 3. Encoder rubber roller; 4. Incremental encoder; 5. Traction steel belt; 6. Buckle; 7. Limiting threaded hole; 8. Lower housing; 9. Guide wheel; 10. Roller bearing limiting wheel; 11. Hall sensor array acquisition board; 12. Probe bracket; 13. Multi-winding coil; 14. High permeability U-shaped magnetic core.

[0022] The accompanying drawings are provided to further understand the present solution and form part of the specification. They are used together with the embodiments of the present solution to explain the present solution and do not constitute a limitation thereof. Detailed Implementation

[0023] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0024] This invention provides a system for detecting and quantifying elevator steel belt damage based on dual-mode detection of eddy current and leakage magnetic flux, such as... Figure 3 As shown, it includes: It features a dual-mode steel strip internal damage detection module, synchronously connecting an eddy current sensor, a magnetic flux leakage sensor, and an encoder via a multi-probe array adapter. The eddy current sensor employs a multi-frequency excitation method to simultaneously acquire detection information at different penetration depths; the magnetic flux leakage sensor uses a high-sensitivity magnetic element to detect abnormal leakage magnetic fields caused by changes in the steel strip's permeability; the encoder provides precise positional information for accurate damage localization. All sensor data are synchronized through a synchronous acquisition circuit to ensure time consistency.

[0025] The signal collaborative denoising module employs an improved wavelet packet transform-empirical mode decomposition (EMD) dual-mode collaborative denoising algorithm. This module includes an adaptive thresholding unit based on sub-band energy entropy. It dynamically allocates hard and soft thresholding functions by calculating the energy entropy of WPT node coefficients and integrates a self-attention mechanism to enhance the weights of key frequency band coefficients. Subsequently, EMD decomposition and Hilbert envelope spectrum analysis are used to filter out residual noise, achieving faithful reconstruction of damage features under non-stationary noise. Specifically, a WPT-EMD steel strip damage signal denoising algorithm combining adaptive threshold adjustment, self-attention, and Hilbert transform is constructed. First, the original signal is preprocessed, including baseline correction and outlier removal. Then, wavelet packet decomposition is performed. By calculating the energy entropy of each node and combining it with the adaptive threshold adjustment mechanism, a thresholding strategy is dynamically determined: a hard thresholding function is used for nodes with high energy entropy (rich features) to retain important transient features; a soft thresholding function is used for nodes with low energy entropy (noise-dominated) to smooth noise fluctuations. Finally, a self-attention mechanism is integrated to enhance the weights of node coefficients based on the correlation between frequency bands and damage features. Finally, EMD decomposition is performed to obtain several intrinsic mode functions (IMFs). The envelope spectrum of each IMF is calculated using Hilbert transform. Based on the characteristics of the envelope spectrum, residual noise components are identified and filtered out, and a high-fidelity damage signal is output.

[0026] Construct a multi-rule feature extraction engine where three feature extraction rules run in parallel: Smoothing residual threshold rule: The Savitzky-Golay filter is used to smooth the signal, the residual between the original signal and the smoothed signal is calculated, and the maximum residual value is extracted.

[0027] Derivative mutation detection rule: The first derivative of the signal is calculated using the five-point numerical differentiation method. The mutation location is identified by finding the extreme points of the derivative, and the maximum mutation amplitude is extracted.

[0028] Local energy detection rule: The signal energy is calculated using a sliding window, with the window length matching the typical size of the damage, and the ratio of signal energy to background noise energy within the window is calculated.

[0029] A physical constraint recognition module for damage features is constructed, employing an improved DE-SVM model. Its core is physically weighted SVM parameters, which introduce physical weight coefficients on top of traditional SVM parameters. These coefficients are determined based on the physical mechanism of damage and expert prior knowledge. For example, parameters such as damage signal fluctuation amplitude, envelope energy, and phase extracted by the multi-rule feature extraction engine directly reflect the degree and location information of damage. For surface cracks, higher physical weights are assigned to derivative abrupt change features; for internal defects in steel strips, higher physical weights are assigned to local energy features. The specific weight allocation ratio is generated by the dynamic confidence weight allocation unit, thus ensuring that the model learning process conforms to physical laws and enhancing the SVM model's fitting ability on small sample datasets.

[0030] like Figure 1 As shown, in one embodiment, the dual-mode steel belt internal damage detection module of the elevator steel belt damage detection and quantitative analysis system based on eddy current and leakage magnetic flux dual-mode detection of this application includes: a main frame (upper shell 1, lower shell 8), a leakage magnetic flux probe module (probe bracket 12, Hall sensor array acquisition board 11), an incremental encoder 4, a dual-mode magnetic field excitation module (high permeability U-shaped magnetic core 14, multi-winding coil 13), a guide wheel 9, a roller bearing limiting wheel 10, and a buckle 6; wherein, the main shell is composed of two parts, upper and lower, respectively placed on both sides of the traction steel belt 5. On the side, two guide wheels 9 are installed in the grooves at the left and right ends of the upper outer shell 1 and the lower outer shell 8 to control the axial movement of the steel belt. Four roller bearing limit wheels 10 are installed at the bottom to control the circumferential movement of the steel belt. Between the two guide wheels, the outer shell is provided with a rectangular groove and two limit threaded holes 7 for fixing the magnetic leakage probe module. The encoder rubber roller 3 and the incremental encoder 4 are installed on the upper right end of the upper outer shell through the encoder bracket 2. The lower outer shell 8 is symmetrically distributed with the upper outer shell and does not have an encoder. The buckle 6 is located in the middle of the upper and lower outer shells and is used to open and close the main body shell.

[0031] like Figure 2 As shown in the previous embodiment, in one example, the dual-mode excitation module of the elevator steel belt damage detection and quantitative analysis system based on eddy current and leakage magnetic flux dual-mode detection of this application consists of three sets of high permeability U-shaped magnetic cores 14 arranged side by side at equal intervals and multi-winding coils 13 surrounding the magnetic arms on both sides, symmetrically distributed on both sides of the Hall sensor array acquisition board 11. For leakage magnetic flux detection, the high permeability magnetic core can enhance the magnetic field and improve the leakage magnetic flux signal strength. For eddy current detection, the high permeability magnetic core can concentrate the alternating magnetic field, improve the eddy current signal strength, and avoid high-frequency loss. The device uses an incremental encoder 4 for axial positioning of defects and the Hall sensor array acquisition board 11 for circumferential positioning of defects, realizing high-precision positioning and tracking of traction steel belt defects in both the axial and circumferential directions.

[0032] refer to Figure 3 In one embodiment, the elevator steel strip damage detection and quantitative analysis system based on dual-mode eddy current and leakage magnetic flux detection of this application is deployed using Docker containers. Each functional microservice (such as device access, inference acceleration, and data storage) runs independently in an isolated container and communicates through a RESTful API. The device access middleware supports hot-swapping, eliminating the need to restart the system when a new device is connected. The blockchain storage unit adopts a lightweight architecture, only uploading key detection results, model parameter hash values, and timestamps to the blockchain, effectively controlling storage and computing costs while ensuring data immutability and reliable process traceability.

[0033] refer to Figure 3 In one embodiment, the elevator steel belt damage detection and quantitative analysis system based on dual-mode detection of eddy current and leakage magnetic flux of this application includes a visual diagnostic interface. The visual diagnostic interface integrates DE-SVM multi-rule feature parallel coordinate visualization, signal comparison display before and after noise reduction, damage pulse location display, and feature dimension filtering and damage inversion analysis driven by touch gestures, forming an interactive diagnostic workbench.

[0034] Specifically, the visualization interface is implemented using WebGL technology, supporting operation in mainstream browsers. The parallel coordinate visualization component supports dynamic selection and dimension rearrangement, making it easy for users to discover the correlation patterns between features. The damage location display is combined with the production line layout diagram to intuitively show the location distribution of damage on the steel strip.

[0035] refer to Figure 3 In one embodiment, the elevator steel strip damage detection and quantitative analysis system based on dual-mode eddy current and leakage magnetic flux detection of this application includes a self-optimizing learning module with a dual-loop update mechanism. The short-term learning loop incrementally updates the DE-SVM model parameters using an online sequence learning algorithm; the long-term reconstruction loop performs full model retraining every 24 hours based on accumulated data. The module includes a feature contribution evaluation unit based on Shapley values, dynamically adjusting the TRDS rule weights and DE-SVM parameters to achieve continuous optimization of model performance.

[0036] Specifically, the self-optimizing learning module employs an online sequential extreme learning machine algorithm to implement short-term learning loops, with a model update latency of less than 1 second, adapting to changes in the data stream in real time. Long-term reconstruction loops utilize full data retraining, employing early stopping strategies to prevent overfitting. The Shapley value-based feature contribution evaluation unit uses Monte Carlo sampling to approximate the marginal contribution of each TRDS feature to the model output, dynamically adjusting the rule weights and SVM kernel function parameters in the feature extraction stage to achieve closed-loop continuous optimization of model performance.

[0037] refer to Figure 3In one embodiment, the elevator steel belt damage detection and quantitative analysis system based on dual-mode eddy current and leakage magnetic flux detection of this application includes an extensible interface module. This module adopts a microservice architecture and includes a device access middleware that supports both Modbus / TCP and OPC-UA protocols, an inference acceleration interface that integrates the OpenVINO™ toolkit, a vibration data exchange interface that conforms to the ISO 18436-4 standard, and a blockchain-based test report storage unit to achieve reliable traceability and auditing of the testing process.

[0038] refer to Figure 3 In one embodiment, the multi-rule feature extraction engine of this application is provided with a dynamic confidence weight allocation unit based on signal-to-noise ratio. This unit employs multi-strategy optimization to weight and fuse the features extracted from eddy current and leakage magnetic signals, outputting high-fidelity fused feature parameters. Specifically, the results extracted by the multi-rule feature extraction engine are dynamically weighted to refine the extraction results. The dynamic confidence weight allocation unit uses multi-strategy optimization for weight allocation. Correlation analysis optimization: Based on historical data, calculate the Pearson correlation coefficient between each TRDS feature and the true degree of damage. Features with higher correlation (e.g., for surface cracks in wire ropes, the derivative mutation feature has a higher correlation) are given higher weights.

[0039] Cross-validation optimization: Use 5-fold cross-validation to evaluate the performance of different weight combinations. Divide the training set into 5 parts, and use 4 parts to train the weight combination in turn, then validate on 1 part. Select the weight combination with the smallest mean squared error and stability on multiple subsets to prevent overfitting.

[0040] Differential Evolution Algorithm Optimization: Initialization: Randomly generate 50 weight combinations as the initial population.

[0041] Mutation: For each weight vector, three different vectors are randomly selected from the population, and a mutation vector is generated using the formula V=Xr1+F(Xr2-Xr3), where F is a scaling factor.

[0042] Crossover: Cross the mutated vector with the current target vector to generate an experimental vector, thereby introducing parameter diversity.

[0043] Selection: Compare the fitness of the experimental vector with that of the current vector (e.g., the coefficient of determination R² of the SVM model on the validation set), and retain the vector with better performance.

[0044] Iteration: Repeat the mutation, crossover and selection process for 100 generations to gradually approach the optimal weight combination.

[0045] Finally, the unit outputs weighted and fused feature parameters, providing high-quality input for subsequent recognition.

[0046] refer to Figure 3 In one embodiment, the physical constraint recognition module of this application is provided with a virtual sample generation unit, which expands the training sample set through Gaussian process regression, noise injection and feature space interpolation techniques to optimize the model generalization ability under small sample conditions. The virtual sample generation unit models the distribution of existing samples through Gaussian process regression, and generates new and reasonable virtual samples in the sparse region of the feature space through noise injection and interpolation techniques, effectively solving the model training problem under small sample conditions and significantly improving the model's generalization ability and robustness.

[0047] refer to Figure 3 In one embodiment, the elevator steel belt damage detection and quantitative analysis system based on eddy current and leakage magnetic flux dual-mode detection of this application adopts a distributed architecture design, divided into a data acquisition layer, a noise reduction layer, and an identification layer. The sensor acquisition device, arranged in the elevator shaft, communicates with the server in the processing layer via industrial Ethernet. The sensor acquisition module adopts a modular design, with each acquisition unit capable of connecting multiple sensors, and synchronous trigger signals ensure the time consistency of data acquisition.

[0048] The following is a specific implementation example.

[0049] Application of damage detection for traction steel belts in a steel belt elevator in a certain residential community

[0050] 1. System Architecture Deployment

[0051] This detection system is installed inside the shaft of a steel belt traction elevator. The detection module is mounted on the crossbeam of the elevator shaft frame below the traction motor. See the attached diagram for the installation location. Figure 4 The detection module is installed on the crossbeam below the traction motor. The hardware configuration includes: a data acquisition terminal containing multiple eddy current and leakage magnetic field sensor channels and a high-precision encoder; an edge computing node deploying a signal preprocessing module; a central server equipped with a multi-core processor and a GPU accelerator card; interactive terminals including a touch screen and a mobile terminal; and a network architecture using industrial Ethernet and wireless backup to ensure real-time data transmission.

[0052] 2. Data Preprocessing Implementation

[0053] The acquired eddy current and magnetic flux leakage signals underwent joint noise reduction processing: firstly, multi-layer wavelet packet decomposition was performed, and adaptive thresholding was used to suppress noise; subsequently, empirical mode decomposition was used to further filter out residual interference. Simultaneously, precise synchronization and alignment of multi-sensor signals were achieved based on encoder pulses. Data processing results showed that the signal-to-noise ratio of the damaged signal after processing by the data noise reduction module was successfully improved from 3.8dB to 30.6dB, significantly improving signal quality and suppressing interference noise.

[0054] 3. Multidimensional damage characteristic analysis

[0055] The TRDS multi-rule feature extraction mechanism is adopted: the smoothing residual threshold rule extracts signal residual features, the derivative mutation detection rule identifies signal mutation features, and the partial energy detection rule calculates energy distribution features.

[0056] Through a dynamic confidence weight allocation unit, combined with correlation analysis, cross-validation and intelligent optimization algorithms, the system achieves optimized fusion of multi-source features. The damage detection system identified 15 different degrees of broken wires and strands, achieved complete identification of the damage waveform, and extracted multiple feature parameters reflecting the physical meaning of the damage, including: damage index, location, peak value, envelope energy, etc.

[0057] 4. Hybrid anomaly detection process

[0058] An improved support vector machine (SVM) model algorithm based on physical parameter feature weight allocation is proposed. This algorithm differentiates the parameters of each dimension of the SVM model by using predefined physical weight coefficients to form a physically constrained feature space. Combined with virtual sample generation technology, the generalization ability of the model under small sample conditions is effectively improved, enabling damage type identification and size quantification assessment. Based on the collected real damage data, the small sample dataset augmentation technique expanded the 30 sets of data to 500 sets of virtual samples, which, together with the real samples, constituted the training set. Using 100 sets of real samples as the test set, DE-SVM was trained and tested on the small sample dataset. The results showed that DE-SVM achieved an R² coefficient of 0.9854, an MSE of 0.272, and a prediction accuracy of 96%; the random forest model achieved an R² coefficient of -0.0776, an MSE of 20.4211, and a prediction accuracy of 78%; the physically weighted kernel function support vector regression model achieved an R² coefficient of 0.8945, an MSE of 2.001, and a prediction accuracy of 64%; and the traditional support vector machine model achieved an R² of -0.097, an MSE of 20.8092, and a prediction accuracy of 72%. The comparison shows that the DE-SVM model proposed in this invention significantly improves the prediction accuracy compared to traditional models, and the physically weighted mechanism also has a strong effect on improving the performance of traditional algorithms. Ultimately, after training with the dataset, the recognition algorithm achieved a damage recognition accuracy of 96%, and could identify damage as small as 5% of the metal cross-sectional area.

[0059] 5. Visualization and Decision Making

[0060] The visual interface integrates functions such as multi-rule feature parallel coordinate display, signal comparison before and after noise reduction, and damage localization display, and supports touch gesture-driven interactive operation. The system provides damage inversion analysis and treatment recommendations, forming a complete diagnostic decision support system.

[0061] 6. Effects of the Example

[0062] After noise reduction processing, the signal-to-noise ratio of the detected signal increased from 3.8dB to 30.6dB, significantly improving signal quality. The multi-rule feature extraction engine identified 15 instances of wire and strand breakage damage of varying degrees and achieved complete identification of the damage waveforms. It accurately extracted multiple feature parameters reflecting the physical significance of the damage. The DE-SVM model showed a significant improvement in prediction accuracy compared to traditional models, and the physical weighting mechanism also had a strong effect on improving the performance of traditional algorithms. Finally, after training on the dataset, the recognition algorithm achieved a damage identification accuracy of 96%, capable of identifying damage as small as 5% of the metal cross-sectional area.

[0063] This embodiment successfully identified various types of steel strip damage in practical applications, optimized the maintenance decision-making process, shortened the personnel training cycle, and verified the technical advantages and practical value of the invention in engineering practice.

[0064] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A system for detecting and quantitatively analyzing elevator steel strip damage based on dual-mode detection of eddy current and leakage magnetic flux, characterized in that, include: The dual-mode steel strip internal damage detection module synchronously connects to an eddy current sensor, a magnetic flux leakage sensor, and an encoder via a multi-probe array adapter. The eddy current sensor is used to acquire detection information at different penetration depths, the magnetic flux leakage sensor is used to detect abnormal leakage magnetic fields caused by changes in the magnetic permeability of the steel strip, and the encoder is used to provide position information. All sensor data are synchronized through a synchronous acquisition circuit to ensure time consistency. The signal collaborative noise reduction module is connected to the dual-mode steel strip internal damage detection module and is used to perform time-frequency domain analysis and background noise filtering on the collected raw data, and output the noise-reduced signal. A multi-rule feature extraction engine, connected to the signal collaborative denoising module, is configured with a variety of signal waveform analysis rules to extract multi-dimensional damage feature parameters from the denoised signal. The physical constraint recognition module is connected to the multi-rule feature extraction engine and is used to construct a machine learning model that integrates physical prior knowledge, mapping the damage feature parameters to damage quantification recognition results.

2. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, The signal collaborative noise reduction module specifically adopts an improved wavelet packet transform-empirical mode decomposition dual-mode collaborative noise reduction algorithm. This module is equipped with an adaptive thresholding unit based on subband energy entropy. It dynamically allocates hard thresholding and soft thresholding functions by calculating the energy entropy of WPT node coefficients, and integrates a self-attention mechanism to enhance the weight of key frequency band coefficients. Subsequently, residual noise is filtered out a second time through EMD decomposition and Hilbert envelope spectrum analysis.

3. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, The multi-rule feature extraction engine extracts features through the TRDS multi-rule damage feature extraction mechanism, specifically including the following rules: Smoothing residual threshold rule: Calculate the residual between the signal and the smoothed signal, and extract the maximum residual value as the first feature; Derivative mutation detection rule: Extract the maximum mutation amplitude of the first derivative of the absolute value of the signal as the second feature; Local energy detection rule: The ratio of signal energy to background noise energy within the detection window is calculated as the third feature.

4. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, The physical constraint identification module specifically constructs an improved support vector machine model based on a physical parameter feature weight allocation algorithm. This algorithm differentiates the parameters of each dimension of the support vector machine model by using predefined physical weight coefficients, and calculates the optimal weight ratio through differential evolution algorithm and cross-validation algorithm to form a physical constraint feature space.

5. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, The dual-mode steel belt internal damage detection module includes: a main frame, a magnetic flux leakage probe module, an incremental encoder (4), a dual-mode magnetic field excitation module, guide wheels (9), roller bearing limit wheels (10), and buckles (6); wherein, the main shell is composed of an upper shell (1) and a lower shell (8), which are respectively placed on both sides of the traction steel belt (5). Two sets of guide wheels (9) are installed in the grooves at the left and right ends of the upper shell (1) and the lower shell (8) to control the axial movement of the steel belt. Four sets of roller bearing limit wheels (10) are installed at the bottom of the upper shell (1) and the lower shell (8) to control the circumferential movement of the traction steel belt (5); rectangular grooves and two sets of limit threaded holes (7) are respectively provided between the two sets of guide wheels (9) installed on the upper shell (1) and the lower shell (8) for fixing. The leakage magnetic field probe module is equipped with an encoder rubber wheel (3) and an incremental encoder (4) mounted on one end of the upper housing (1) via a movable bracket (2); the lower housing (8) is symmetrically distributed with the upper housing and does not have an encoder; the buckle (6) is located in the middle of the upper housing (1) and the lower housing (8) and is used to open and close the main frame; the dual-mode excitation module is installed between the leakage magnetic field probe module and the rectangular groove. The dual-mode excitation module consists of three sets of high permeability U-shaped magnetic cores (14) arranged side by side with equal spacing and multi-winding coils (13) surrounding the magnetic arms on both sides. The leakage magnetic field probe module consists of a probe bracket (12) and a Hall sensor array acquisition board (11) mounted on it. The multi-winding coils (13) are symmetrically distributed on both sides of the Hall sensor array acquisition board (11).

6. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, It adopts Docker containerization deployment, with each functional microservice running independently in an isolated container and communicating through a RESTful API; the device access middleware supports hot-swapping, so there is no need to restart the system when a new device is connected.

7. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, It also includes a visual diagnostic interface that integrates DE-SVM multi-rule feature parallel coordinate visualization, signal comparison display before and after noise reduction, damage pulse localization display, and feature dimension filtering and damage inversion analysis driven by touch gestures, forming an interactive diagnostic workbench.

8. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, It also includes a self-optimizing learning module: configured with a dual-loop update mechanism, the short-term learning loop incrementally updates the DE-SVM model parameters through an online sequence learning algorithm; the long-term reconstruction loop performs full model retraining every 24 hours based on accumulated data; the module has a feature contribution evaluation unit based on Shapley value, which dynamically adjusts the TRDS rule weights and DE-SVM parameters to achieve continuous optimization of model performance.

9. The steel strip damage detection and quantification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, It also includes an extensible interface module, which adopts a microservice architecture and includes a device access middleware that supports both Modbus / TCP and OPC-UA protocols, an inference acceleration interface that integrates the OpenVINO™ toolkit, a vibration data exchange interface that conforms to the ISO 18436-4 standard, and a blockchain-based test report storage unit to achieve reliable traceability and auditing of the testing process.

10. The steel strip damage detection and quantification identification system based on dual-mode eddy current and leakage magnetic flux detection according to claim 1, characterized in that, It also includes a data augmentation and optimization module to improve the performance of the recognition model. The data augmentation and optimization module includes: A high-fidelity feature fusion unit is located in the multi-rule feature extraction engine. Based on the signal-to-noise ratio, the features extracted from eddy current and leakage magnetic signals are dynamically weighted and optimized by multiple strategies to output high-fidelity fused feature parameters. The virtual sample generation unit, located in the physical constraint recognition module, expands the training sample set composed of the high-fidelity fused feature parameters through Gaussian process regression, noise injection, and feature space interpolation techniques to optimize the model's generalization ability under small sample conditions.

Images