Steel wire rope nondestructive testing method and system based on sensing and physical feature fusion

Abstract

The invention discloses a steel wire rope nondestructive testing method and system based on sensing and physical feature fusion, and belongs to the technical field of steel wire rope nondestructive testing. The method comprises the following steps: firstly, carrying out saturation magnetization on a steel wire rope through an axial Gaussian difference excitation probe so as to improve the signal-to-noise ratio of a defect magnetic flux leakage signal; and meanwhile, the instantaneous acceleration is synchronously acquired through the inertial measurement unit. And establishing a speed-signal model based on the law of electromagnetic induction, and performing dynamic speed compensation on the magnetic flux leakage original voltage signal. And then, quantitatively extracting characteristic parameters with clear physical significance, such as peak voltage, maximum gradient, full width at half maximum and the like, from the standardized magnetic flux leakage signal, fusing the characteristic parameters with deep characteristics extracted by the deep learning model, and inputting a fused characteristic vector into a full-connection neural network and a classifier to realize defect type identification and degree evaluation. According to the invention, high-precision and interpretable intelligent detection and residual life prediction of steel wire rope defects can be realized, the detection reliability is obviously improved, and predictive maintenance is supported.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of nondestructive testing of steel wire ropes, and particularly relates to a nondestructive testing method and system for steel wire ropes based on fusion of sensing and physical characteristics. BACKGROUND

[0002] As a key component of the core power bearing, steel wire ropes are widely used in major equipment and infrastructure such as elevators, mine hoisting systems, port cranes, cable-stayed bridges, etc. In the long-term service process, they need to bear complex alternating loads, and the working environment is often very harsh, which is easy to gradually produce various damages such as broken wires, wear and corrosion. These damages are not only highly concealed, but also develop rapidly, and the traditional visual inspection is difficult to find internal defects, and the conventional nondestructive testing methods also have many limitations.

[0003] Although the current mainstream magnetic flux leakage detection method is widely used in steel wire rope detection, there are still several key technical bottlenecks: first, the uniformity of the magnetization field of the traditional detection probe is insufficient, and it is difficult to achieve complete saturation magnetization of the steel wire rope, resulting in weak response signal of the micro defect and deep defect, low signal-to-noise ratio, and high risk of missing detection. The actual measurement data shows that the detection rate of the traditional probe for broken wires with a diameter of less than 0.5 mm is less than 60%, and the detection rate for internal corrosion is less than 40%. Second, the fluctuation of the running speed of the steel wire rope during the field detection will directly modulate the signal amplitude, which seriously affects the accurate judgment of the damage size. The signal amplitude change caused by speed fluctuation can reach ±30%, which makes quantitative evaluation extremely difficult. Third, the existing automatic system mostly uses simple threshold alarm or pure data-driven artificial intelligence model, the former has low intelligence and high false alarm rate, and the latter is a "black box" due to the lack of physical mechanism support, and the decision-making process is difficult to explain, and the generalization ability is insufficient under the condition of sample scarcity in the industrial field, which is difficult to gain the complete trust of users.

[0004] These technical bottlenecks seriously restrict the development and application effect of the nondestructive testing technology of steel wire ropes. Therefore, it has become an urgent need of the industry to develop a new intelligent detection method and device with high reliability and high precision, which can improve the quality from the signal source, effectively overcome the speed interference, and fuse the advantages of physical mechanism and artificial intelligence. It has extremely important significance for ensuring the safe operation of major equipment and preventing safety accidents. SUMMARY

[0005] The purpose of the present application is to provide a nondestructive testing method and system for steel wire ropes based on fusion of sensing and physical characteristics, which aims to ensure the safe operation and realize predictive maintenance of steel wire ropes through high-sensitivity, anti-interference and interpretable nondestructive testing and life prediction.

[0006] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is: The steel wire rope nondestructive testing method based on sensing and physical feature fusion comprises the following steps: Step S1, saturate magnetization of the steel wire rope through an axial Gauss difference excitation probe, and synchronously collect leakage magnetic raw voltage signals on the surface of the steel wire rope by using a Hall sensor array, and simultaneously, synchronously collect instantaneous acceleration data of the probe in the axial direction at a preset sampling rate by using an inertial measurement unit fixed to the axial Gauss difference excitation probe; Step S2, perform numerical integration and filtering processing on the instantaneous acceleration data to calculate an instantaneous running speed of the steel wire rope, and based on the electromagnetic induction law, perform point-by-point dynamic compensation on the leakage magnetic raw voltage signals by using the instantaneous running speed to obtain a standardized leakage magnetic signal irrelevant to the speed; Step S3, perform automatic peak value detection on the standardized leakage magnetic signal obtained in step S2 to identify potential damage signal areas, and for each damage signal area, quantitatively extract a plurality of feature parameters with explicit physical meanings, the feature parameters including a peak voltage, a maximum gradient, a full width at half maximum and a peak-to-valley difference, and combine the extracted feature parameters to form a physical feature vector; Step S4, fuse the physical feature vector obtained in step S3 with a deep feature vector extracted from the standardized leakage magnetic signal by using a deep learning model to generate a fusion feature vector, input the fusion feature vector into a fully connected neural network and a classifier, output a diagnosis result of a damage type and a damage degree of the steel wire rope, simultaneously output physical feature parameter basis corresponding to a diagnosis decision, and predict a remaining service life of the steel wire rope based on the diagnosis result.

[0007] Further, in the step S1, the axial Gauss difference excitation probe cooperates with a main coil and an auxiliary coil to generate a steady magnetization field in a steel wire rope detection area, and the magnetization field forms a magnetic flat area with a magnetic field strength not less than 2.5T and a magnetic field uniformity not less than 95% in a center area of the air gap.

[0008] Further, the synthetic magnetic field B(z) generated by the axial Gauss difference excitation probe in the step S1 is designed by the following empirical formula: In the formula, B(z) represents a synthetic magnetic field strength of the axial Gauss difference excitation probe at an axial displacement z, B1 is a magnetic field amplitude of the main coil, B2 is a magnetic field amplitude of the auxiliary coil, and are parameters for controlling the width of the magnetic field distribution, is a constant, and z is the axial displacement.

[0009] Furthermore, in step S2, the original leakage flux voltage signal is dynamically compensated point-by-point using the instantaneous operating speed, achieved through the following compensation formula: In the formula, It is the compensated standardized leakage magnetic field signal. It is the original voltage signal of leakage flux. It refers to instantaneous velocity.

[0010] Furthermore, in step S3, when extracting the peak voltage, the signal maxima in the damaged signal region are located using a peak-finding algorithm; when extracting the maximum gradient, the maximum value of the first derivative in the damaged signal region is calculated using the five-point differential method; when extracting the full width at half peak, the time width corresponding to the half-power point of the signal is measured using an interpolation method, and the spatial width is obtained by multiplying the instantaneous running speed by the time width; when extracting the peak-valley difference, the difference between the peak and valley values ​​of the bimodal signal is calculated using an extreme point search algorithm.

[0011] Furthermore, in step S4, the deep learning model is a convolutional neural network, which sequentially extracts features from the standardized magnetic flux leakage signal through convolutional layers, activation layers, and pooling layers, and outputs a deep feature vector with a dimension of not less than 256. The classifier is a Softmax classifier, which is used to output the confidence level of each damage type, and the type with the highest confidence level is used as the final damage type diagnosis result.

[0012] Furthermore, in step S4, the deep learning model is trained by injecting the physical feature vector as prior knowledge into the training process, and the total loss function used is: In the formula, Let N be the total loss function, and N be the number of samples. For the true labels of the samples, To predict probabilities for the model, Here, M represents the weighting coefficients, and M represents the number of physical features. These are the actual physical characteristic values. Predict physical characteristic values ​​for the model.

[0013] Furthermore, in step S4, the prediction of the remaining service life is based on the diagnostic results combined with the historical load data, usage environment parameters and usage frequency data of the wire rope. The remaining service life of the wire rope is calculated and output through a multi-parameter damage evolution model.

[0014] A non-destructive testing system for steel wire ropes based on the fusion of sensor and physical features is used to execute the aforementioned non-destructive testing method for steel wire ropes based on the fusion of sensor and physical features. The system includes: The multimodal data acquisition module includes an axial Gaussian differential excitation probe, a Hall sensor array, an inertial measurement unit, and a synchronous acquisition card. The axial Gaussian differential excitation probe is used to saturate the wire rope. The Hall sensor array is used to acquire the original leakage magnetic field voltage signal on the surface of the wire rope. The inertial measurement unit is used to acquire the axial instantaneous acceleration data of the probe. The synchronous acquisition card is used to synchronously record the original leakage magnetic field voltage signal and acceleration data and add a unified timestamp. The signal processing module receives the data output from the multimodal data acquisition module, performs numerical integration and filtering on the acceleration data to calculate the instantaneous running speed, and performs point-by-point dynamic compensation on the original leakage magnetic field voltage signal based on the instantaneous running speed, outputting a standardized leakage magnetic field signal that is independent of speed. The feature extraction module is used to automatically detect peak values ​​of standardized magnetic flux leakage signals, identify damaged signal regions, extract physical feature parameters, and output physical feature vectors. The intelligent diagnostic module includes a convolutional neural network unit, a feature fusion unit, and a classification decision unit. The convolutional neural network unit is used to extract the depth feature vector of the standardized magnetic flux leakage signal. The feature fusion unit is used to splice the physical feature vector and the depth feature vector to form a fused feature vector. The classification decision unit is used to output the diagnostic results of damage type and damage degree, as well as the corresponding physical feature parameters. The life prediction module is used to calculate and output the remaining service life of the wire rope by combining damage diagnosis results, load history data, environmental parameters and usage frequency data through a multi-parameter damage evolution model.

[0015] Furthermore, the Hall sensor array is uniformly arranged along the circumference of the steel wire rope, and its sampling rate is not less than 100kHz; the sampling rate of the inertial measurement unit is not less than 1000Hz.

[0016] The beneficial effects of the above scheme are as follows: 1. This invention achieves ultra-high sensitivity detection of minute and hidden defects, significantly reducing the risk of missing minute defects. Traditional magnetic flux leakage probes suffer from insufficient uniformity of the magnetization field, resulting in weak response signals to minute defects. This solution, through an innovative axial Gaussian difference excitation probe design, generates a saturated magnetization field with a magnetic field strength of 2.5T and a uniformity exceeding 95% in the detection area, effectively exciting the magnetic flux leakage signal of minute defects. The unique magnetic circuit design improves the signal-to-noise ratio of the probe to 28dB for detecting 0.3mm broken wires, providing a guarantee for the timely detection of safety hazards from the physical source.

[0017] 2. This invention significantly suppresses the impact of speed fluctuations, achieving true quantitative detection and ensuring the comparability and reliability of results. Addressing the industry pain point of speed fluctuations reaching ±30% during on-site detection, this solution employs a 1000Hz sampling rate IMU to monitor motion in real time, establishing a speed-signal mathematical model based on the law of electromagnetic induction, combined with an adaptive filtering algorithm. This technology can reduce the signal amplitude error caused by speed fluctuations from the traditional ±25% to ±1.5%, meaning that the detected signal amplitude is essentially consistent for the same defect within a speed range of 0.5-5 m / s. Laboratory and field verifications have shown that the system maintains stable detection accuracy within this speed range, providing the possibility for precise damage quantification.

[0018] 3. This invention constructs a highly accurate and interpretable intelligent diagnostic system, overcoming the limitations of black-box models. This invention quantifies and extracts physical features directly related to the geometry of damage from standardized signals, such as peak voltage, maximum gradient, full width at half maximum (FWHM), and peak-to-valley difference, and then fuses these features with deep features automatically extracted by CNN. The system provides physical evidence simultaneously when outputting diagnostic results. For example, when the diagnosis is "broken wire," it displays quantified physical evidence such as "signal gradient greater than 18V / s, peak-to-valley difference exceeding 0.85V"; when the diagnosis is "corrosion," it displays characteristic parameters such as "signal width greater than 8mm, gradient less than 5V / s." This dual-drive mode provides complete decision-making evidence, greatly enhancing the reliability and credibility of engineering applications.

[0019] 4. This invention leverages few-shot learning capabilities, significantly reducing application barriers and costs. Addressing the challenge of scarce defect samples in industrial settings, this solution injects prior knowledge into physical features and combines it with a transfer learning framework, achieving diagnostic accuracy comparable to traditional pure CNN models with only 200-500 labeled samples. The system supports online incremental learning, reducing the required labeled samples for new defect types to less than 50. This allows for rapid deployment and optimization, greatly reducing model training costs, shortening implementation cycles, and enhancing the practicality and widespread applicability of the solution.

[0020] 5. This invention features predictive maintenance capabilities, facilitating the full lifecycle management of wire ropes. The system establishes a multi-parameter fusion damage evolution model, comprehensively considering data such as load history, environmental factors, and usage frequency, to accurately predict remaining service life. This function can avoid downtime losses caused by sudden fractures, truly realizing the transformation from passive inspection to predictive maintenance, and possesses broad application value and socio-economic benefits. Attached Figure Description

[0021] Figure 1 This is an overall flowchart of the non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features of the present invention; Figure 2This is a framework diagram of the wire rope non-destructive testing system based on the fusion of sensing and physical features of the present invention. Detailed Implementation

[0022] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0023] It should be noted that, unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0024] Example 1 like Figure 1 As shown, this embodiment provides a non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features, including the following steps: Step S1: The wire rope is saturated magnetized using an Axial Gaussian-Difference Excitation Probe (AGDE), and the original voltage signal of leakage magnetic flux on the surface of the wire rope is synchronously acquired using a Hall sensor array. At the same time, the instantaneous acceleration data of the probe axis is synchronously acquired using an Inertial Measurement Unit (IMU) fixed to the AGDE probe at a preset sampling rate. Step S2: Perform numerical integration and filtering on the instantaneous acceleration data to calculate the instantaneous running speed of the wire rope. Based on the law of electromagnetic induction, use the instantaneous running speed to perform point-by-point dynamic compensation on the original leakage magnetic field voltage signal to obtain a standardized leakage magnetic field signal that is independent of speed. Step S3: Perform automatic peak detection on the standardized leakage magnetic field signal obtained in step S2 to identify potential damage signal regions. For each damage signal region, quantify and extract multiple feature parameters with clear physical meaning. The feature parameters include peak voltage, maximum gradient, full width at half maximum (FWHM), and peak-to-valley difference. Combine the extracted feature parameters to form a physical feature vector. Step S4: The physical feature vector obtained in step S3 is fused with the deep feature vector extracted from the standardized leakage magnetic signal through the deep learning model to generate a fused feature vector. The fused feature vector is then input into a fully connected neural network and a classifier to output the diagnostic results of the wire rope damage type and damage degree, as well as the physical feature parameters corresponding to the diagnostic decision. Based on the diagnostic results, the remaining service life of the wire rope is predicted.

[0025] The following is a detailed explanation of the implementation process of a non-destructive testing method for wire ropes based on the fusion of sensor and physical characteristics, using the non-destructive testing of the hoisting wire rope of a 40-ton gantry crane at an international container terminal as an example. The crane operates for 18 hours daily, and the wire rope has been in use for 11 months. The testing environment was 28℃, 65% humidity, and 3.4 m / s wind speed. The specific steps are as follows: Step S1: Multimodal data acquisition and preprocessing This step achieves synchronous acquisition of multimodal data through an axial Gaussian difference excitation probe, a Hall sensor array, an inertial measurement unit, and a synchronous acquisition card. The core is to construct a high-intensity, high-uniformity magnetization field to ensure the acquisition of high-quality raw data.

[0026] Design and optimization of axial Gaussian difference excitation probe: Based on the empirical formula for the synthesis magnetic field: In the formula, This represents the resultant magnetic field strength of the axial Gaussian difference excitation probe at an axial displacement z. It is the amplitude of the magnetic field of the main coil. It is the amplitude of the magnetic field of the auxiliary coil. and It is a parameter that controls the width of the magnetic field distribution, where C is a constant and z is the axial displacement.

[0027] The probe parameters were optimized using finite element simulation software. Through multiple simulations and adjustments, an optimal set of parameters was found to determine the amplitude of the main coil magnetic field. =3.2T, Auxiliary coil magnetic field amplitude =0.7T =5mm, and =3mm, so that the probe forms a magnetically flat region with a magnetic field strength of not less than 2.5T and a magnetic field uniformity of not less than 95% in the central region of the air gap, which meets the requirements for saturation magnetization of the wire rope.

[0028] The optimized axial Gaussian difference excitation probe was installed at the detection position of the hoisting wire rope on the gantry crane, ensuring that the wire rope could pass through the central air gap of the probe at a uniform speed. Thirty-two high-sensitivity Hall sensors were evenly arranged around the probe to form a Hall sensor array, with a sampling rate set to 100kHz, to acquire the original leakage magnetic field voltage signal on the wire rope surface. An inertial measurement unit was securely mounted on the probe, with a sampling rate set to 1000Hz, to acquire the instantaneous acceleration data along the probe's axis. A synchronous acquisition card was also connected to ensure that the original leakage magnetic field voltage signal and acceleration data were recorded synchronously and stamped with a unified timestamp, avoiding data time deviations from affecting subsequent analysis.

[0029] A preset constant current is applied to the main and auxiliary excitation coils, with the main coil current at 15A and the auxiliary coil current at 5A, to establish a stable magnetization field. The data acquisition system is then activated. As the gantry crane hoists the wire rope, the Hall sensor array continuously collects the original voltage signal of the leakage magnetic flux on the wire rope surface. The inertial measurement unit synchronously acquires the probe's axial instantaneous acceleration data. The synchronous acquisition card stores two data streams in real time and adds a unified timestamp. This acquisition process lasted two hours. Table 1 details the key parameters obtained through direct measurement and calculation during the data acquisition phase. The data was primarily obtained through direct measurement: the three-dimensional leakage magnetic field strength on the surface of the wire rope was directly measured using a 32-Hall sensor array on the AGDE probe and converted into a voltage signal, thus obtaining the original signal peak value and background noise level; the probe's triaxial acceleration data was acquired in real time using an IMU, and the instantaneous running speed was obtained after integration; the signal-to-noise ratio was calculated by dividing the original signal peak value by the background noise level. These measured data provide a reliable data foundation for all subsequent advanced analyses.

[0030] Table 1 Raw test data The original signal peak reached 3.27V, far exceeding the normal range upper limit of 0.5V by 654%, strongly indicating severe metal loss or damage. The signal-to-noise ratio of 40.2dB is well above the minimum requirement of 30dB, indicating excellent signal quality. This is mainly due to the highly uniform magnetization field design of the AGDE probe, which effectively suppresses environmental electromagnetic interference. The velocity value of 2.10m / s is within the normal operating range of 1.8-2.5m / s, ruling out the possibility of signal abnormalities caused by velocity anomalies. The background noise level of 0.08V is below the 0.1V threshold, indicating a good electromagnetic environment and excellent probe shielding, providing a reliable foundation for subsequent accurate signal analysis.

[0031] Step S2: Dynamic speed compensation and signal standardization The core objective of this step is to eliminate the linear modulation effect of velocity fluctuations on the amplitude of the original signal, converting the unstable original signal into a standardized signal that is only related to the geometry of the damage and independent of the motion conditions. This is crucial for achieving quantitative damage identification and comparability of results.

[0032] First, the acceleration data collected in step S1 Preprocessing is performed. A 50Hz low-pass filter is used to filter out high-frequency noise (mainly noise generated by environmental vibration), and then the instantaneous running speed is calculated based on numerical integration. The calculation formula is: In the formula, This represents the instantaneous speed of the wire rope at time t. It is the initial velocity at t=0, which is set according to the actual running speed of the wire rope at the start of the test.

[0033] To further improve the accuracy of velocity calculation, a Kalman filter algorithm is used to smooth the integration results, eliminating the DC drift that may be introduced by the integration, and finally obtaining a smooth and accurate instantaneous velocity sequence. Verification showed that the speed calculation error was controlled within ±0.02m / s.

[0034] Then, dynamic speed compensation is performed. Based on the dynamic speed compensation formula: In the formula, It is a standardized leakage magnetic field signal after speed compensation. It is the original voltage signal of leakage flux. It refers to instantaneous velocity.

[0035] Point-by-point compensation is performed on the original leakage flux voltage signal. This is done for each discrete sampling time point. The original leakage magnetic field voltage signal value at that moment Divide by the corresponding instantaneous running speed value The standardized leakage magnetic signal value at that point is obtained. This ultimately generates a complete standardized leakage magnetic signal sequence.

[0036] Table 2 shows the processing effect of the velocity compensation algorithm. All data were calculated: signal amplitude fluctuation was calculated by statistically analyzing the standard deviation of the signal before and after compensation; velocity sensitivity was obtained by fitting the linear relationship between signal amplitude and velocity to obtain the slope value; and the normalized signal amplitude was calculated point by point using the compensation algorithm. All calculations were based on the physical model of the law of electromagnetic induction.

[0037] Table 2. Processing effect of speed compensation algorithm Velocity compensation significantly reduced signal amplitude fluctuations from ±27% to ±1.2%, an improvement of 22.5 times. This remarkable improvement demonstrates that the compensation algorithm effectively eliminates the impact of velocity fluctuations on the signal. Velocity sensitivity decreased from 0.35V / (m / s) to 0.02V / (m / s), a reduction of 94%, indicating that the compensated signal is almost unaffected by velocity changes. The normalized signal amplitude is 1.89V, a value that eliminates the velocity factor and accurately reflects the physical characteristics of the defect. The Kalman filtering algorithm used in the compensation process effectively suppressed noise from acceleration measurements, and the precision of the numerical integration algorithm ensured the accuracy of velocity calculations, providing stable and reliable input data for subsequent damage identification.

[0038] This step strictly follows the basic physical laws of electromagnetic induction, mathematically and precisely removes interference terms, fundamentally solving the core problem that the detection results depend on speed, and ensuring that the signal amplitude obtained by detecting the same defect at different speeds remains consistent.

[0039] Step S3: Physical Feature Extraction and Enhancement This step automatically detects peak values ​​in the standardized magnetic flux leakage signal obtained in step S2, identifies potential damage signal regions, and quantifies and extracts feature parameters with clear physical meaning to form a physical feature vector. The specific implementation process is as follows: (1) Perform automatic peak detection: Use an adaptive threshold peak finding algorithm to detect the standardized leakage magnetic field signal. The process involves several steps. First, the average amplitude and standard deviation of the signal are calculated. The average amplitude plus three times the standard deviation is set as the initial detection threshold. Then, the signal is scanned point by point along the time axis. When the signal value exceeds the threshold and the signal values ​​of adjacent points show an initial increase followed by a decrease, it is identified as a potential damage signal peak point, thus determining the start and end positions of the damage signal region. A total of 12 potential damage signal regions were identified in this detection, among which one region exhibited significant signal characteristics and requires focused analysis.

[0040] (2) Feature parameter extraction: For the identified significant damage signal areas, four feature parameters are extracted respectively: peak voltage, maximum gradient, full width at half maximum and peak-to-valley difference.

[0041] Peak voltage extraction: ,in Represents the standardized leakage magnetic field signal after velocity compensation. The maximum amplitude. By traversing all data points within the damaged signal region, the maximum signal value is found, which is the peak voltage. .

[0042] Maximum gradient extraction: ,in Represents standardized magnetic leakage signal The maximum rate of change. The first derivative of the damaged signal region is calculated using the five-point differential method. The first derivative sequence is obtained through calculation, and the maximum value of the sequence is found, which is the maximum gradient.

[0043] Full width at half maximum (FWHM) extraction: ,in The width of the space representing the defect. It is the instantaneous operating speed. This refers to the time width corresponding to the half-power point of the signal. First, determine the voltage value corresponding to 50% of the peak voltage of the damaged signal (i.e., the half-power point). Then, use linear interpolation to find the time points corresponding to the half-power point voltage value at the rising and falling edges of the signal. The difference between the two time points is the time width corresponding to the half-power point.

[0044] Peak-valley difference extraction: ,in Peak value representing a bimodal signal Valley value The difference between them. An extreme point search algorithm is used to traverse the damaged signal region and find the signal peak. Valley value The difference between the two is the peak-valley difference.

[0045] (3) Formation of physical feature vector: The extracted peak voltage, maximum gradient, full width at half maximum (FWHM), and peak-to-valley difference are combined to form a physical feature vector. .

[0046] This step aims to extract feature parameters with clear physical meaning from the standardized signal. Table 3 lists the quantized physical feature parameters obtained by the feature extraction algorithm. Each parameter has been measured multiple times and averaged to ensure stability.

[0047] Table 3 Quantitative Physical Characteristic Parameters The peak amplitude of 1.89V exceeds the upper limit of the normal range by 0.6V, reaching 315%. Such a high peak voltage indicates significant metal loss, typically corresponding to severe damage. The maximum gradient of 22.4V / s exceeds the standard of 8V / s by 280%. This large gradient value indicates a very sharp damage edge, a typical characteristic of wire breakage, as the fracture surface generates strong magnetic field distortion. The full width at half maximum (FWHM) of 6.8mm is within the normal range of 2-10mm, indicating that the axial dimension of the damage is normal and not a characteristic of large-area wear. The peak-to-valley difference of 1.25V exceeds the standard of 0.3V by 416%. Such a large peak-to-valley difference is a hallmark of wire breakage signals in magnetic flux leakage detection, as wire breakage produces a typical double-peak signal. The combination of these four characteristic parameters clearly points to wire breakage damage, and the degree of damage is relatively severe.

[0048] The features extracted in this step have clear physical meaning, are directly related to the geometric shape of the defect, and are mathematical expressions of domain expert knowledge. This greatly enhances the interpretability of the system, injects prior physical knowledge into the intelligent diagnostic process, provides strong guidance for the diagnostic model, and helps improve the model's generalization ability and decision credibility when there are insufficient samples.

[0049] Step S4: Intelligent Diagnosis and Remaining Useful Life Prediction This step will use physical feature vectors The deep feature vectors extracted by the deep learning model are fused together and input into a fully connected neural network and classifier for intelligent diagnosis. Simultaneously, the remaining lifespan is calculated based on the diagnostic results and multiple parameters. The specific implementation process is as follows: (1) Deep learning model construction and training: A one-dimensional convolutional neural network (CNN) is used as the deep learning model. The network structure includes an input layer, a feature extraction layer, and an output layer. The feature extraction layer contains three groups of convolutional layers, activation layers, and pooling layers. The parameters for each group are uniform: kernel size 3, stride 1, and padding method "same". The activation layers all use the ReLU function, and the pooling layers all use max pooling with a kernel size of 2 and a stride of 2. The only difference between the groups is the number of convolutional kernels, which are 32, 64, and 128 respectively. The output layer consists of one fully connected layer, outputting a depth feature vector with a dimension of 256. .

[0050] The training dataset uses a wire rope leakage magnetic signal dataset containing various damage types such as broken wires, wear, and corrosion, with a total of 500 samples (180 broken wires, 170 wear, and 150 corrosion). 80% of the samples are used as the training set and 20% as the validation set.

[0051] The physical feature vectors extracted in step S3 are injected into the training process as prior knowledge, specifically by adding a physical feature constraint term to the loss function. The total loss function is a weighted sum of the cross-entropy loss function and the sum of squared errors of the physical features, as shown in the formula: In the formula, Let N be the total loss function, and N be the number of samples. For the true labels of the samples, To predict probabilities for the model, Here, M represents the weighting coefficients, and M represents the number of physical features. These are the actual physical characteristic values. Predict physical characteristic values ​​for the model.

[0052] In this embodiment, The value was set to 0.3. The Adam optimizer was used, with the initial learning rate set to 0.001. Every 50 epochs, the learning rate was decayed to 0.5. After 100 training iterations, the final model achieved an accuracy of 99% on the validation set, meeting the diagnostic requirements.

[0053] (2) Feature fusion and intelligent diagnosis: The physical feature vector obtained in step S3 The feature vector Vfused is formed by concatenating and fusing the deep feature vectors extracted by the convolutional neural network described above, using a method of linking the vectors end to end.

[0054] Feature fusion formula: In the formula, It is the fused feature vector. It is a physical feature vector. It is a feature vector extracted by deep learning. This represents a vector concatenation operation.

[0055] fuse feature vectors The input is a fully connected neural network, which contains two hidden layers. The first hidden layer has 128 neurons with the ReLU activation function, and the second hidden layer has 64 neurons with the ReLU activation function. The output layer is connected to a Softmax classifier, which outputs the confidence score of each damage type. The type with the highest confidence score is used as the final damage type diagnosis result. At the same time, the physical feature parameters corresponding to the diagnosis decision are also output.

[0056] Remaining service life prediction: Based on the above diagnostic results, combined with the wire rope's historical load data (average load of 35 tons and maximum load of 40 tons over the past 11 months) and environmental parameters (temperature 28℃, humidity 65%, salt spray concentration 0.05mg / m), 3 Based on usage frequency data (18 hours of work per day, approximately 200 lifting operations), a multi-parameter damage evolution model was used to calculate the remaining service life. The multi-parameter damage evolution model is based on linear cumulative damage theory, combined with the damage growth law of wire ropes. The model formula is as follows: In the formula, Indicates the remaining service life of the wire rope. D represents the rated service life of the wire rope, and D represents the current degree of damage. , , , For model parameters, For average load, For ambient temperature, For usage frequency.

[0057] In this embodiment, the rated service life of this type of wire rope is 24 months. Based on the two broken wires, D=0.3 was determined, and this was obtained through fitting historical data. , , =0.8 .

[0058] Table 4 shows the output results of the intelligent diagnosis. All data were derived through model inference: the damage type was determined by the probability distribution of the multi-class softmax output; the damage degree was quantified by the regression network output; the damage location was calculated based on high-precision encoder data and decoding algorithms; and the remaining lifetime prediction was derived through time series prediction model inference. All outputs are accompanied by confidence assessments.

[0059] Table 4 Output results of the intelligent diagnostic system The diagnostic results showed that the damage type was wire breakage, with a confidence level of 98.7%. The corresponding physical characteristic parameters were a maximum gradient of 22.4V / s and a peak-to-valley difference of 1.25V, which are consistent with the typical characteristics of wire breakage signals. A regression network was used to quantitatively assess the damage level, outputting a damage level of 2 broken wires, with a confidence level of 95.3%. This confidence level indicates that the model is very confident in this quantitative judgment, mainly based on the matching degree between the peak amplitude of 1.89V and two wire breakage cases in historical data. The damage location was accurate to 125.6m, and the positioning error of ±0.2m fully meets the maintenance requirements, thanks to the high-precision absolute encoder and strict speed synchronization algorithm used in the system. The remaining life prediction was 48 days, with an 85% confidence level indicating high prediction reliability. This prediction integrates multiple factors such as the current damage level, historical usage data, and load conditions, and is calculated using a degradation model based on the Wiener process.

[0060] Example 2 like Figure 2 As shown, this embodiment provides a non-destructive testing system for steel wire ropes based on the fusion of sensor and physical features, used to perform the aforementioned non-destructive testing method for steel wire ropes based on the fusion of sensor and physical features. The system includes a multi-modal data acquisition module, a signal processing module, a feature extraction module, an intelligent diagnostic module, and a life prediction module. The specific structure and functions are as follows: (1) Multimodal data acquisition module The multimodal data acquisition module mainly consists of an axial Gaussian differential excitation probe, a Hall sensor array, an inertial measurement unit, and a synchronous acquisition card. Its core function is to realize the synchronous acquisition of the original voltage signal of the wire rope leakage magnetic field and the instantaneous axial acceleration data of the probe.

[0061] Axial Gaussian difference excitation probe: Utilizing the optimized structure described in step S1 above, the main coil and auxiliary coil work together to generate a steady-state magnetization field in the wire rope detection area. The magnetic field strength in the air gap center region is no less than 2.5T, and the magnetic field uniformity is no less than 95%, ensuring the wire rope is saturated with magnetization and providing a magnetic field foundation for obtaining high-quality leakage magnetic signals. The probe housing is made of high-strength aluminum alloy, possessing excellent impact and corrosion resistance, making it suitable for harsh working environments such as ports.

[0062] Hall sensor array: Composed of 32 high-sensitivity Hall sensors, uniformly arranged along the circumference of the wire rope with a sensor spacing of 11.25° and a sampling rate of 100kHz, it can quickly and accurately acquire the original voltage signal of leakage magnetic flux on the surface of the wire rope. The sensor adopts a differential output mode, effectively suppressing common-mode interference and improving the signal anti-interference capability. Its operating temperature range is -40℃ to 85℃, meeting the detection needs in different environments.

[0063] Inertial Measurement Unit (IMU): A six-axis IMU is employed, comprising a three-axis accelerometer and a three-axis gyroscope. The sampling rate is set to 1000Hz. It is fixed to the axial Gaussian difference excitation probe via a dedicated bracket to ensure consistency with the probe's motion state, enabling precise acquisition of instantaneous acceleration data along the probe's axis. The IMU exhibits excellent stability and accuracy, with an acceleration measurement range of ±16g and a measurement error of less than 0.1%, providing reliable data for subsequent instantaneous velocity calculations.

[0064] Synchronous Acquisition Card: Employing a high-speed 16-bit resolution synchronous acquisition card, it features two analog input channels, connecting to the Hall sensor array and the inertial measurement unit respectively. With a sampling clock synchronization accuracy of less than 1μs, it enables synchronized recording of the original leakage magnetic field voltage signal and acceleration data, and assigns a unified timestamp to both data streams to avoid errors caused by data time asynchrony. The acquisition card connects to a host computer for real-time data transmission and storage, meeting the data storage requirements for long-term testing.

[0065] (2) Signal processing module The signal processing module uses an industrial-grade embedded processor. Its core function is to receive the data output from the multimodal data acquisition module, perform numerical integration and filtering on the acceleration data, calculate the instantaneous running speed, and perform point-by-point dynamic compensation on the original leakage magnetic field voltage signal based on the instantaneous running speed, and output a standardized leakage magnetic field signal.

[0066] Data reception and preprocessing: The signal processing module receives data transmitted by the synchronous acquisition card through the Ethernet interface. First, it performs format parsing and verification on the data and removes invalid data. Then, it uses a 50Hz low-pass filter to filter out high-frequency noise generated by environmental vibrations, preparing for subsequent velocity calculations.

[0067] Instantaneous velocity calculation: Based on the preprocessed acceleration data, the instantaneous operating velocity is calculated using the numerical integration algorithm and Kalman filter algorithm from step S2 above. The calculation period is 1ms to ensure real-time performance. Simultaneously, a velocity anomaly monitoring function is set up. When the calculated instantaneous velocity exceeds the normal range, an alarm signal is issued to remind operators to check the equipment's operating status.

[0068] Dynamic speed compensation: Based on the dynamic speed compensation formula, the original leakage magnetic field voltage signal is compensated point by point. The compensation algorithm is implemented with hardware acceleration. A dedicated computing unit is built through FPGA to improve the compensation processing speed and ensure that leakage magnetic field signals with a sampling rate of 100kHz can be processed in real time. The standardized leakage magnetic field signal after compensation is transmitted to the feature extraction module through the Ethernet interface.

[0069] (3) Feature extraction module The core function of the feature extraction module is to automatically detect the peak value of the standardized magnetic flux leakage signal output by the signal processing module, identify the damaged signal region, extract physical feature parameters, and output a physical feature vector.

[0070] Automatic peak detection program: An adaptive threshold peak finding program is written in MATLAB. The program first reads the standardized leakage magnetic field signal data, calculates the average amplitude and standard deviation of the signal, and automatically sets the detection threshold. Then, it traverses the signal through a sliding window, identifies potential damage signal regions, and marks the start and end indices of the regions. The program runs in less than 10 seconds, meeting the requirements for real-time detection.

[0071] Feature Parameter Extraction Program: For the identified damage signal regions, a feature parameter extraction program was developed to calculate peak voltage, maximum gradient, full width at half maximum (FWHM), and peak-to-valley difference. The maximum gradient is calculated using the five-point differential method, the FWHM using linear interpolation, and the peak-to-valley difference using an extreme point search algorithm. The program can automatically handle damage signal regions of different lengths, with an extraction accuracy error of less than 5%. The extracted feature parameters are stored in a text file format and simultaneously transmitted to the intelligent diagnostic module via the network.

[0072] (4) Intelligent diagnostic module The intelligent diagnosis module and the feature extraction module share a high-performance computer. The core function is to extract the deep feature vector of the standardized magnetic flux leakage signal through the convolutional neural network unit, use the feature fusion unit to splice the physical feature vector and the deep feature vector to form a fused feature vector, and then output the diagnosis results of damage type and damage degree and the corresponding physical feature parameters through the classification decision unit.

[0073] Convolutional Neural Network Unit: Construct the above one-dimensional convolutional neural network model, perform data preprocessing according to the model input requirements, and then input the processed data into the convolutional neural network to extract the depth feature vector.

[0074] Feature fusion unit: The feature fusion program reads the physical feature vector transmitted by the feature extraction module and the deep feature vector output by the convolutional neural network unit, and forms a fused feature vector through vector concatenation operation. The concatenation process automatically verifies the vector dimension to ensure the correctness of the fusion.

[0075] Classification Decision Unit: This unit constructs a fully connected neural network and a Softmax classifier. The Softmax classifier outputs the confidence score for each damage type. This unit inputs the fused feature vectors into the fully connected neural network, calculates the damage type and severity, and simultaneously calls upon physical feature parameter data to output the physical feature parameter basis for the diagnostic decision.

[0076] (5) Lifetime prediction module The life prediction module is also deployed on a high-performance computer. Its core function is to combine the damage diagnosis results output by the intelligent diagnosis module, the historical load data input from the outside, environmental parameters and usage frequency data, and calculate and output the remaining service life of the wire rope through a multi-parameter damage evolution model.

[0077] The data input interface supports manual input or import of historical load data (such as average load, maximum load), environmental parameters (such as temperature, humidity, salt spray concentration), and usage frequency data (such as daily working hours, number of lifts). After data input, the format is automatically checked to ensure data validity.

[0078] Multi-parameter damage evolution model program: Based on linear cumulative damage theory, a multi-parameter damage evolution model calculation program was developed. The program reads damage diagnosis results and input multi-parameter data, substitutes them into the model formula, and calculates the remaining service life. The prediction results are displayed in text and chart form, and a detection report is generated simultaneously. The report includes information such as damage type, damage degree, damage location, and remaining service life, which facilitates the equipment management department in formulating maintenance plans.

[0079] On-site verification and effect evaluation Based on the diagnostic results obtained from the above testing methods and system, on-site verification was conducted on the hoisting wire rope of a 40-ton gantry crane at an international container terminal. Maintenance personnel verified the damage location (125.6m from the end) using three methods: visual inspection, magnetic particle testing, and ultrasonic testing, based on the system-output damage location. Table 5 shows the verification results; the data were obtained through on-site measurements. Visual inspection: Using a high-powered magnifying glass to observe the damage location, two obvious broken wires were found. The ends of the broken wires showed obvious fatigue fracture characteristics, which was consistent with the damage type and degree diagnosed by the system.

[0080] Magnetic particle testing: Wet magnetic particle testing is used. Magnetic suspension is sprayed onto the damaged area. After applying a magnetic field, a clear magnetic mark appears at the broken wire. The length of the broken wire is measured to be about 15 mm, further confirming the existence of the broken wire damage.

[0081] Ultrasonic testing: A 2.5MHz ultrasonic probe was used to perform transverse wave detection on the damaged area. No other hidden damage was found, indicating that the system's diagnostic results were accurate and no damage was missed.

[0082] Table 5. On-site verification results In summary, the on-site verification results were completely consistent with the system diagnosis, confirming the presence of two broken wires with a positional error of only 0.1m. This level of accuracy significantly exceeds that of traditional detection methods, demonstrating the system's high accuracy. The damage was in its early stages, and the system promptly issued an early warning, providing ample time for preventative maintenance and preventing a potential sudden breakage accident. The recommended treatment was replacement within 30 days, a timeframe that considered both safety margins and the impact on maintenance resources and production plans. This inspection successfully prevented a potentially major safety accident caused by wire rope breakage, estimating a direct reduction in downtime losses of approximately 150,000 yuan. This case fully demonstrates the reliability and practical value of this technical solution in engineering practice, showcasing the crucial role of intelligent detection systems in equipment safety management.

[0083] Effect evaluation Detection sensitivity: Under the condition of detection speed ≤5m / s, the system has a detection rate of 99.8% for 0.5mm broken wires and a detection rate of 97.5% for internal micro-corrosion with a depth ≥0.1mm. Compared with traditional detection methods, the detection sensitivity is greatly improved.

[0084] Anti-interference performance: Through dynamic speed compensation technology, the signal amplitude error caused by speed fluctuations is reduced from the traditional ±25% to ±1.5%, and stable detection accuracy can be maintained in the speed range of 0.5-5m / s, effectively overcoming speed fluctuation interference.

[0085] Diagnostic interpretability: When outputting diagnostic results, the intelligent diagnostic system simultaneously provides physical characteristic parameters (such as maximum gradient and peak-valley difference), which enhances the credibility of the diagnostic results and achieves a diagnostic accuracy rate of 99.2%.

[0086] Small sample learning capability: This system only needs 200-500 labeled samples to achieve the training effect of tens of thousands of samples in traditional methods. The requirement for labeled samples for new defect types is reduced to less than 50, which greatly reduces the application threshold and shortens the implementation cycle.

[0087] This invention effectively solves the problems of low sensitivity, susceptibility to speed interference, poor interpretability, and reliance on a large number of labeled samples in traditional non-destructive testing methods for wire ropes by innovative axial Gaussian difference excitation probe design, dynamic velocity compensation technology, intelligent diagnostic model integrating physical and depth features, and multi-parameter damage evolution model. Field verification and effect evaluation show that this method and system have high detection accuracy, good stability, and strong practicality, and can provide reliable technical support for the safety inspection of wire ropes in major equipment and infrastructure, and has broad application value.

[0088] Finally, it should be noted that any parts of this invention not described in detail are prior art. Those skilled in the art will understand that the above descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features, characterized in that, Includes the following steps: Step S1: The wire rope is saturated magnetized by an axial Gaussian difference excitation probe, and the original voltage signal of leakage magnetic field on the surface of the wire rope is synchronously collected by a Hall sensor array. At the same time, the instantaneous acceleration data of the probe axis is synchronously collected by an inertial measurement unit fixed to the axial Gaussian difference excitation probe at a preset sampling rate. Step S2: Perform numerical integration and filtering on the instantaneous acceleration data to calculate the instantaneous running speed of the wire rope. Based on the law of electromagnetic induction, use the instantaneous running speed to perform point-by-point dynamic compensation on the original leakage magnetic field voltage signal to obtain a standardized leakage magnetic field signal that is independent of speed. Step S3: Perform automatic peak detection on the standardized leakage magnetic field signal obtained in step S2 to identify potential damage signal regions. For each damage signal region, quantify and extract multiple feature parameters with clear physical meaning. The feature parameters include peak voltage, maximum gradient, full width at half maximum (FWHM), and peak-to-valley difference. Combine the extracted feature parameters to form a physical feature vector. Step S4: The physical feature vector obtained in step S3 is fused with the deep feature vector extracted from the standardized leakage magnetic signal through the deep learning model to generate a fused feature vector. The fused feature vector is then input into a fully connected neural network and a classifier to output the diagnostic results of the wire rope damage type and damage degree, as well as the physical feature parameters corresponding to the diagnostic decision. Based on the diagnostic results, the remaining service life of the wire rope is predicted.

2. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 1, characterized in that, In step S1, the axial Gaussian difference excitation probe generates a steady-state magnetization field in the wire rope detection area through the cooperation of the main coil and the auxiliary coil. This magnetization field forms a magnetically flat region with a magnetic field strength of not less than 2.5T and a magnetic field uniformity of not less than 95% in the air gap center region.

3. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 2, characterized in that, The synthetic magnetic field B(z) generated by the axial Gaussian difference excitation probe in step S1 is designed by the following empirical formula: In the formula, This represents the resultant magnetic field strength of the axial Gaussian difference excitation probe at an axial displacement z. It is the amplitude of the magnetic field of the main coil. It is the amplitude of the magnetic field of the auxiliary coil. and It is a parameter that controls the width of the magnetic field distribution. is a constant, and z is the axial displacement.

4. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 1, characterized in that, In step S2, the original leakage flux voltage signal is dynamically compensated point by point using the instantaneous operating speed, which is achieved through the following compensation formula: In the formula, It is the compensated standardized leakage magnetic field signal. It is the original voltage signal of leakage flux. It refers to instantaneous velocity.

5. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 1, characterized in that, In step S3, when extracting the peak voltage, the signal maxima in the damaged signal region are located using a peak-finding algorithm; when extracting the maximum gradient, the maximum value of the first derivative in the damaged signal region is calculated using the five-point differential method; when extracting the full width at half peak, the time width corresponding to the half-power point of the signal is measured using an interpolation method, and the spatial width is obtained by multiplying the instantaneous running speed by the time width; when extracting the peak-valley difference, the difference between the peak and valley values ​​of the bimodal signal is calculated using an extreme point search algorithm.

6. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 1, characterized in that, In step S4, the deep learning model is a convolutional neural network, which sequentially extracts features from the standardized magnetic flux leakage signal through convolutional layers, activation layers, and pooling layers, and outputs a deep feature vector with a dimension of not less than 256. The classifier is a Softmax classifier, which is used to output the confidence level of each damage type, and the type with the highest confidence level is used as the final damage type diagnosis result.

7. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 6, characterized in that, In step S4, the deep learning model is trained by injecting the physical feature vector as prior knowledge into the training process, and the total loss function used is: In the formula, Let N be the total loss function, and N be the number of samples. For the true labels of the samples, To predict probabilities for the model, Here, M represents the weighting coefficients, and M represents the number of physical features. These are the actual physical characteristic values. Predict physical characteristic values ​​for the model.

8. The non-destructive testing method for steel wire ropes based on the fusion of sensing and physical features according to claim 1, characterized in that, In step S4, the remaining service life is predicted based on the diagnostic results combined with the wire rope's load history data, usage environment parameters, and usage frequency data. The remaining service life of the wire rope is calculated and output through a multi-parameter damage evolution model.

9. A non-destructive testing system for steel wire ropes based on the fusion of sensing and physical features, characterized in that, A method for performing the non-destructive testing of wire ropes based on the fusion of sensing and physical features as described in any one of claims 1 to 8, comprising: The multimodal data acquisition module includes an axial Gaussian differential excitation probe, a Hall sensor array, an inertial measurement unit, and a synchronous acquisition card. The axial Gaussian differential excitation probe is used to saturate the wire rope. The Hall sensor array is used to acquire the original leakage magnetic field voltage signal on the surface of the wire rope. The inertial measurement unit is used to acquire the axial instantaneous acceleration data of the probe. The synchronous acquisition card is used to synchronously record the original leakage magnetic field voltage signal and acceleration data and add a unified timestamp. The signal processing module receives the data output from the multimodal data acquisition module, performs numerical integration and filtering on the acceleration data to calculate the instantaneous running speed, and performs point-by-point dynamic compensation on the original leakage magnetic field voltage signal based on the instantaneous running speed, outputting a standardized leakage magnetic field signal that is independent of speed. The feature extraction module is used to automatically detect peak values ​​of standardized magnetic flux leakage signals, identify damaged signal regions, extract physical feature parameters, and output physical feature vectors. The intelligent diagnostic module includes a convolutional neural network unit, a feature fusion unit, and a classification decision unit. The convolutional neural network unit is used to extract the depth feature vector of the standardized magnetic flux leakage signal. The feature fusion unit is used to splice the physical feature vector and the depth feature vector to form a fused feature vector. The classification decision unit is used to output the diagnostic results of damage type and damage degree, as well as the corresponding physical feature parameters. The life prediction module is used to calculate and output the remaining service life of the wire rope by combining damage diagnosis results, load history data, environmental parameters and usage frequency data through a multi-parameter damage evolution model.

10. The wire rope non-destructive testing system based on sensor and physical feature fusion according to claim 9, characterized in that, The Hall sensor array is uniformly arranged along the circumference of the steel wire rope, and its sampling rate is not less than 100kHz; the sampling rate of the inertial measurement unit is not less than 1000Hz.

Images