Metal magnetic memory signal defect identification method and system based on feature enhancement

Abstract

The embodiment of the application discloses a metal magnetic memory signal defect identification method and system based on feature enhancement, the system comprises a signal acquisition module, a feature extraction module and a defect identification module; the method comprises: acquiring the multi-channel metal magnetic memory signal of the target metal component by using the signal acquisition module; based on the multi-channel metal magnetic memory signal, the multi-channel time domain feature and the multi-channel frequency domain feature are extracted respectively by using the feature extraction module, and the multi-channel enhanced feature is obtained; based on the pre-determined optimal feature subset, the corresponding target feature is selected from the multi-channel enhanced feature, and the optimal feature vector is obtained; the optimal feature vector is input into the pre-trained random forest model by using the defect identification module, and the defect identification result for the target metal component is obtained; wherein, the pre-trained random forest model is generated by an unsupervised clustering algorithm based on the label-free historical multi-channel enhanced feature, and is obtained by training based on the training label.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing of industrial components, and relates to, but is not limited to, a method and system for identifying defects in metal magnetic memory signals based on feature enhancement. Background Technology

[0002] Metal components, with their superior properties such as high strength, high toughness, and impact resistance, are widely used in key industrial fields such as hoisting machinery, mining equipment, bridge engineering, and port machinery. However, during long-term service, metal components are affected by multiple factors such as load cycles, friction and wear, and corrosion aging, making them prone to defects such as wire breakage, wear, stress concentration, and cracks. If these defects are not detected and identified in a timely manner, they will continue to deteriorate with the extension of service time, eventually leading to component fracture and failure, causing serious consequences such as equipment failure, production interruption, and even personal injury or death, resulting in huge economic losses and adverse social impacts. Therefore, early defect detection and identification of metal components is a key link in ensuring industrial production safety and reducing operating costs.

[0003] Currently, commonly used non-destructive testing methods for metal components in the industrial field mainly include electromagnetic testing, visual inspection, acoustic emission testing, and ultrasonic testing. Among them, visual inspection is greatly affected by lighting conditions, component surface condition, and obstruction, resulting in limited detection accuracy and difficulty in identifying internal and minute surface defects. Acoustic emission testing equipment is expensive, has weak anti-interference capabilities, and performs poorly in complex industrial environments with background noise. Ultrasonic testing requires highly skilled operators and a certain level of component surface cleanliness, resulting in low detection efficiency. Although traditional electromagnetic testing has high detection sensitivity, it requires a complex excitation device, making the equipment bulky and cumbersome to operate. It also has stringent requirements for component surface cleanliness; impurities such as sludge and rust on the surface can severely attenuate the detection signal, leading to a significant decrease in detection accuracy. Furthermore, it requires tedious surface cleaning before testing, severely impacting detection efficiency.

[0004] Metal magnetic memory testing technology, as a novel non-destructive testing technology, has the advantage of not requiring an external magnetic field. It utilizes the irreversible change in the magnetic state of metal components under the combined influence of the Earth's magnetic field and operating loads, and detects changes in weak magnetic signals on the component surface using sensors to achieve early defect detection. This technology also features ease of operation, high detection efficiency, and the ability to predict potential failure risks, making it promising for industrial testing. However, existing metal magnetic memory testing technologies suffer from several drawbacks. First, they rely on a single feature extraction dimension, depending on only a few signal features for defect identification, making it difficult to comprehensively characterize the complex characteristics of defects. Second, their anti-interference capabilities are insufficient, easily affected by environmental noise and surface impurities, resulting in low defect identification accuracy. Third, model training relies on a large amount of manually labeled data, while defect samples are scarce in industrial settings, leading to high costs and low efficiency in manual labeling. Fourth, the optimization of detection model parameters is insufficient, often relying on empirical values ​​or local optimization methods, making it difficult to achieve globally optimal performance. Summary of the Invention

[0005] In view of this, the present application provides a method and system for identifying defects in metal magnetic memory signals based on feature enhancement, which at least solves the problem of low defect identification accuracy caused by the single dimension of feature extraction and the reliance on manual annotation of a large number of defect samples in the prior art.

[0006] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a method for identifying defects in metal magnetic memory signals based on feature enhancement, applied to a system for identifying defects in metal magnetic memory signals based on feature enhancement; the system includes: a signal acquisition module, a feature extraction module, and a defect identification module; the method includes: The signal acquisition module is used to acquire multi-channel metal magnetic memory signals of the target metal component; Using the feature extraction module, multi-channel time-domain features and multi-channel frequency-domain features are extracted based on the multi-channel metallic magnetic memory signal to obtain multi-channel enhanced features; based on a pre-determined optimal feature subset, corresponding target features are selected from the multi-channel enhanced features to obtain the optimal feature vector; Using the defect identification module, the optimal feature vector is input into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

[0007] Secondly, embodiments of this application provide a metal magnetic memory signal defect identification system based on feature enhancement, the system comprising: a signal acquisition module, a feature extraction module, and a defect identification module; The signal acquisition module is used to acquire the multi-channel metal magnetic memory signal of the target metal component; The feature extraction module is used to extract multi-channel time-domain features and multi-channel frequency-domain features based on the multi-channel metal magnetic memory signal to obtain multi-channel enhanced features; and to select corresponding target features from the multi-channel enhanced features based on a pre-determined optimal feature subset to obtain the optimal feature vector. The defect identification module is used to input the optimal feature vector into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

[0008] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement a method for identifying defects in metal magnetic memory signals based on feature enhancement.

[0009] The beneficial effects of the technical solutions provided in this application include at least the following: This application acquires multi-channel metallic magnetic memory signals of target metal components, providing a crucial raw data foundation for subsequent multi-channel feature extraction. Compared to single-channel signals, multi-channel metallic magnetic memory signals contain more comprehensive spatial magnetic field information of the target metal component, enabling the method in this application to fully explore the changes in metallic magnetic memory signals caused by defects from multiple directions, laying a data foundation for improving the accuracy of defect identification. Based on the multi-channel metallic magnetic memory signals, multi-channel time-domain features and multi-channel frequency-domain features are extracted separately to obtain multi-channel enhanced features. The multi-channel time-domain features and multi-channel frequency-domain features characterize the defects of the target metal component from two dimensions: macroscopic mechanical response and microscopic physical mechanism, respectively. The fusion of multi-channel time-domain features and multi-channel frequency-domain features overcomes the limitation of the limited dimensionality of single features, improves the comprehensive characterization ability of complex defect characteristics, provides comprehensive feature input for subsequent random forest models, and improves the defect identification accuracy of target metal components from multiple dimensions. Based on a pre-determined optimal feature subset, corresponding target features are selected from the multi-channel enhanced features to obtain the optimal feature vector. By employing feature dimensionality reduction and feature importance filtering, redundant and low-contribution features are eliminated, reducing the computational complexity of the model and improving its generalization ability and operational efficiency. The optimal feature vector is input into a pre-trained random forest model to obtain defect identification results for the target metal component. The pre-trained random forest model is based on unlabeled historical multi-channel enhanced features, generating training labels through an unsupervised clustering algorithm, and then training based on these labels. This addresses the problems of scarce defect samples in industrial settings and high costs of manual annotation, reduces the random forest model's dependence on labeled data, and enhances the practicality and generalizability of the proposed method. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating a method for identifying defects in metal magnetic memory signals based on feature enhancement, provided in an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of the method provided in the embodiments of this application; Figure 3 This is a schematic diagram of a feature-enhanced metal magnetic memory signal defect identification system provided in an embodiment of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0013] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0014] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0015] This application provides a method for identifying defects in metal magnetic memory signals based on feature enhancement. Figure 1 A flowchart illustrating a method for identifying defects in metal magnetic memory signals based on feature enhancement, as provided in this application embodiment, is shown below. Figure 1 As shown, the method includes at least the following steps: Step S110: Use the signal acquisition module to acquire the multi-channel metal magnetic memory signal of the target metal component.

[0016] The target metal component can be any type of ferromagnetic metal component, such as load-bearing components like wire ropes, cables, and steel structures.

[0017] Multi-channel metal magnetic memory signals are multi-channel data reflecting the spatial leakage magnetic field distribution on the surface of a target metal component, synchronously acquired by an array of magnetic sensors deployed on the component's surface. Specifically, "multi-channel" refers to independent signal acquisition performed simultaneously in multiple spatial directions, thus obtaining more comprehensive magnetic field vector information than signals from a single direction.

[0018] By acquiring multi-channel metal magnetic memory signals, a crucial raw data foundation is provided for subsequent multi-channel feature extraction. Compared to single-channel signals, multi-channel metal magnetic memory signals contain more comprehensive spatial magnetic field information of the target metal component, enabling the method in this application to fully explore the changes in magnetic memory signal characteristics caused by defects from multiple directions, thus laying a data foundation for improving the accuracy of defect identification.

[0019] Step S120: Using the feature extraction module, based on the multi-channel metal magnetic memory signal, extract multi-channel time-domain features and multi-channel frequency-domain features respectively to obtain multi-channel enhanced features; based on a pre-determined optimal feature subset, select the corresponding target features from the multi-channel enhanced features to obtain the optimal feature vector.

[0020] Multi-channel time-domain characteristics can intuitively reflect the macroscopic statistical characteristics of signals as they change with spatial location, directly corresponding to the mechanical behavior such as stress concentration and distribution range at the defect location.

[0021] Multi-channel frequency domain characteristics can reveal the distribution and energy characteristics of frequency components within multi-channel metallic magnetic memory signals, and capture high-frequency responses caused by microstructural damage in materials.

[0022] Multi-channel time-domain features and multi-channel frequency-domain features characterize the defects of the target metal component from two dimensions: macroscopic mechanical response and microscopic physical mechanism, respectively. The fusion of multi-channel time-domain features and multi-channel frequency-domain features constitutes multi-channel enhanced features. Multi-channel enhanced features overcome the limitations of single feature dimensions, improve the comprehensive characterization ability of complex defect characteristics, provide richer feature inputs for subsequent random forest models, and improve the defect identification accuracy of the target metal component from multiple dimensions.

[0023] The pre-determined optimal feature subset is determined during the training phase of the random forest model by evaluating and filtering historical features based on their importance.

[0024] In the practical application phase of the model, based on the optimal feature subset, corresponding features are selected from the extracted multi-channel enhancement features to form the optimal feature vector. Through feature dimensionality reduction and importance screening, redundant and low-contribution features are effectively eliminated, which not only significantly reduces the computational complexity of the model but also helps to improve the model's generalization ability and operating efficiency.

[0025] Step S130: Using the defect identification module, the optimal feature vector is input into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

[0026] Defect identification results can include the spatial coordinates of the defect and model performance metrics. The optimal feature vector is input into a pre-trained random forest model (hereinafter referred to as the random forest model). With the help of the random forest model's ensemble learning mechanism, the optimal feature vector is fused and the decision is corrected to improve the identification accuracy.

[0027] Experiments show that the method achieves an accuracy of 98.16% on the test set, effectively identifying defects.

[0028] The random forest model is based on unlabeled historical multi-channel augmented features. Training labels are generated through an unsupervised clustering algorithm to train the random forest model. This solves the problems of scarce defect samples in industrial fields and high costs of manual annotation, reduces the random forest model's dependence on labeled data, and improves the practicality and scalability of the proposed method.

[0029] This application acquires multi-channel metallic magnetic memory signals of target metal components, providing a crucial raw data foundation for subsequent multi-channel feature extraction. Compared to single-channel signals, multi-channel metallic magnetic memory signals contain more comprehensive spatial magnetic field information of the target metal component, enabling the method in this application to fully explore the changes in metallic magnetic memory signals caused by defects from multiple directions, laying a data foundation for improving the accuracy of defect identification. Based on the multi-channel metallic magnetic memory signals, multi-channel time-domain features and multi-channel frequency-domain features are extracted separately to obtain multi-channel enhanced features. The multi-channel time-domain features and multi-channel frequency-domain features characterize the defects of the target metal component from two dimensions: macroscopic mechanical response and microscopic physical mechanism, respectively. The fusion of multi-channel time-domain features and multi-channel frequency-domain features overcomes the limitation of the limited dimensionality of single features, improves the comprehensive characterization ability of complex defect characteristics, provides comprehensive feature input for subsequent random forest models, and improves the defect identification accuracy of target metal components from multiple dimensions. Based on a pre-determined optimal feature subset, corresponding target features are selected from the multi-channel enhanced features to obtain the optimal feature vector. By employing feature dimensionality reduction and feature importance filtering, redundant and low-contribution features are eliminated, reducing the computational complexity of the model and improving its generalization ability and operational efficiency. The optimal feature vector is input into a pre-trained random forest model to obtain defect identification results for the target metal component. The pre-trained random forest model is based on unlabeled historical multi-channel enhanced features, generating training labels through an unsupervised clustering algorithm, and then training based on these labels. This addresses the problems of scarce defect samples in industrial settings and high costs of manual annotation, reduces the random forest model's dependence on labeled data, and enhances the practicality and generalizability of the proposed method.

[0030] Optionally, the multi-channel metal magnetic memory signal includes an X-channel metal magnetic memory signal, a Y-channel metal magnetic memory signal, and a Z-channel metal magnetic memory signal; acquiring the multi-channel metal magnetic memory signal of the target metal component includes: deploying an array-type magnetic sensor on the surface of the target metal component, moving the array-type magnetic sensor at a constant speed to detect the X-channel metal magnetic memory signal, the Y-channel metal magnetic memory signal, and the Z-channel metal magnetic memory signal; preprocessing the X-channel metal magnetic memory signal, the Y-channel metal magnetic memory signal, and the Z-channel metal magnetic memory signal to obtain the preprocessed multi-channel metal magnetic memory signal.

[0031] An array of magnetic sensors is deployed above the wire rope detection path, moving at a constant speed for detection. The raw signals collected by the sensors are amplified and filtered by a signal conditioning circuit, and then transmitted to a computer via a data acquisition card to simultaneously acquire the X-channel, Y-channel, and Z-channel magnetic memory signals of the wire rope.

[0032] The X-channel's magnetic memory signal primarily reflects the tangential magnetic field component, used to detect stress gradient changes at defect edges. The Y-channel's magnetic memory signal primarily reflects the normal magnetic field component, with its zero point corresponding to stress concentration regions. The Z-channel's magnetic memory signal supplements spatial magnetic field information, further enhancing the three-dimensional characterization of defects.

[0033] A steel wire rope with a diameter of 8 mm and a structure of 6 strands by 19 wires was selected as the target metal component. This steel wire rope contained defects of varying degrees, including artificially created broken wires and wear. In the actual testing process, an array of magnetic sensors was deployed above the steel wire rope.

[0034] A multi-step preprocessing operation was performed on the acquired X-channel, Y-channel, and Z-channel metallic magnetic memory signals. First, an adaptive median filtering algorithm was used to remove impulse noise from the signals. Next, wavelet thresholding was used to suppress environmental electromagnetic noise interference. Finally, signal standardization was performed to normalize the signal amplitude to a uniform range, eliminating the influence of signal amplitude differences under different detection environments. The resulting clean and stable preprocessed multi-channel metallic magnetic memory signals provide a high-quality data foundation for subsequent feature extraction.

[0035] Based on the fundamental theory of ferromagnetism, in the natural Earth's magnetic field environment, the inhomogeneity of the internal structure of metallic materials leads to high stress energy under external stress. According to the principle of energy minimization, to offset the increase in stress energy and minimize the total free energy of the system, the magnetic domain walls in the dislocation accumulation region undergo irreversible rearrangement. This process is mainly manifested as an increase in magnetoelastic energy, which in turn induces a spontaneous leakage magnetic field within the target metallic component with a strength significantly higher than the Earth's magnetic field. Due to the influence of various internal friction mechanisms within the metal, the local stress concentration state of the target metallic component will still be preserved even after the dynamic load is removed. Metal magnetic memory defect detection is based on this principle. By measuring the distribution of the spontaneous leakage magnetic field generated in the stress concentration region of the target metallic component under working load, its magnetic field characteristics are extracted, thereby enabling the analysis and evaluation of the local stress state and defect existence of the target metallic component.

[0036] The relationship between the leakage magnetic field and the change in mechanical stress on the target metal component is shown in Equation (1): Formula (1); in, It is a leakage magnetic field. The magnetoelastic coupling coefficient is... For external magnetic field environment, This is the irreversible component of the magnetoelastic effect, which depends on mechanical stress, external magnetic field strength, and temperature. For changes in mechanical stress, This is the permeability of free space. For example, The value of can be 4π × 10 -7 .

[0037] The largest variation in the scattered leakage magnetic field, i.e., the tangential component of the leakage magnetic field, is formed in the stress and deformation concentration region. (x) has a maximum value, while the normal component (y) changes sign and has a zero point.

[0038] Optionally, the multi-channel time-domain features include peak-to-peak value, peak value, gap factor, and signal gradient; the multi-channel frequency-domain features include wavelet packet energy entropy and wavelet packet scale entropy; the step of extracting multi-channel time-domain features and multi-channel frequency-domain features based on the multi-channel metallic magnetic memory signal to obtain multi-channel enhancement features includes: extracting the peak-to-peak value, peak value, gap factor, and signal gradient of each of the X, Y, and Z channels based on the preprocessed multi-channel metallic magnetic memory signal; extracting the wavelet packet energy entropy and wavelet packet scale entropy of each of the X, Y, and Z channels based on the preprocessed multi-channel metallic magnetic memory signal; and obtaining the multi-channel enhancement features based on the peak-to-peak value, peak value, gap factor, signal gradient, wavelet packet energy entropy, and wavelet packet scale entropy.

[0039] Based on the intrinsic correlation mechanism between the magnetic memory signal of metals and defects, and combined with the mechanical behavior of materials, this application conducted in-depth mining of multi-dimensional features. Specifically, four types of time-domain features and two types of frequency-domain features were extracted. The four types of time-domain features include: peak value, peak-peak value, clearance factor, and signal gradient; the two types of frequency-domain features include: wavelet packet energy entropy (E1) and wavelet packet scale entropy (S1). Therefore, the 18-dimensional features, namely the four types of time-domain features and the two types of frequency-domain features, are specifically manifested in the X, Y, and Z channels, thus comprehensively characterizing the magnetic memory response characteristics of defects. For example, the peak value in the X channel.

[0040] Specifically, the 18 features are: X-peak, Y-peak, Z-peak, X-peak-peak, Y-peak-peak, Z-peak-peak, X-clearance factor, Y-clearance factor, Z-clearance factor, X-gradient, Y-gradient, Z-gradient, X-clearance factor, Y-clearance factor, Z-clearance factor, X-gradient, Y-gradient, Z-gradient, X-wavelet packet energy entropy (X-E1), Y-clearance factor, Z-clearance factor, X-gradient, Y-gradient, Z-gradient, X-wavelet packet scale entropy (X-S1), Y-clearance factor, Z-clearance factor, X-gradient, Y-gradient, Z-gradient, and Z-gradient.

[0041] Multi-channel time-domain characteristics can reflect signal intensity variations, extreme value characteristics, and distribution patterns, and are directly related to the material mechanical behavior caused by defects. Multi-channel time-domain characteristics include peak-to-peak value, peak value, gap factor, and signal gradient.

[0042] The peak value corresponds to the maximum degree of plastic deformation in the core region of the defect, allowing for precise location of the defect center. The calculation is shown in formula (2): Formula (2); in, Peak value, Given a time-domain sequence of a channel's magnetic memory signal, where t is the sampling time. It represents the maximum absolute value of the signal.

[0043] Peak-to-peak value reflects the intensity of stress concentration at the defect location; the more pronounced the stress concentration, the larger the peak-to-peak value. The calculation is shown in formula (3): Formula (3); in, Peak-to-peak value Given a time-domain sequence of a channel's magnetic memory signal, where t is the sampling time. The maximum value of the absolute value of the signal. It is the minimum absolute value of the signal.

[0044] The clearance factor reflects the uneven stress distribution in the defect region. The calculation is shown in formula (4): Formula (4); in, The difference in signal amplitude between adjacent points The reference amplitude of the magnetic memory signal in the normal region is used, and the experimental calibration is based on the peak mean of the signal in the normal region.

[0045] The signal gradient corresponds to the stress gradient at the defect edge, enabling accurate identification of the defect boundary. (Signal gradient) The calculation is shown in formula (5): Formula (5); in, The difference in signal amplitude between adjacent points Let be the physical location coordinates corresponding to the i-th sampling point. Let be the physical location coordinates corresponding to the (i+1)th sampling point. Let be the physical location coordinates corresponding to the i-th sampling point. This represents the amplitude of the magnetic memory signal at the (i+1)th sampling point. This represents the amplitude of the magnetic memory signal at the i-th sampling point. This represents the sampling interval.

[0046] Multi-channel frequency domain features are extracted through wavelet packet decomposition, reflecting the distribution characteristics of signal frequency components and related to mechanical behaviors such as grain fragmentation and lattice distortion within the material. These multi-channel frequency domain features include wavelet packet energy entropy and wavelet packet scale entropy.

[0047] Wavelet energy entropy can quantify the uniformity of signal energy distribution across different frequency bands. In defective regions, the energy distribution dispersion increases due to the increased high-frequency components. The calculation is shown in formula (6): Formula (6); Where L is the total number of subbands in the wavelet packet decomposition, L=8; For sub-band index, This represents the energy percentage of the k-th subband.

[0048] Wavelet scale entropy reflects the degree of disorder in the frequency scale of a signal. Plastic deformation and stress concentration in materials can disrupt the frequency stability of magnetic signals, leading to an increase in scale entropy. The calculation is shown in formula (7): Formula (7); Where m is the wavelet packet decomposition layer number, which can be from 1 to 3 layers; It represents the percentage of total energy allocated to the m-th layer of decomposition.

[0049] Optionally, the method further includes: acquiring historical multi-channel enhancement features; performing dimensionality reduction on the multi-channel enhancement features based on principal component analysis to obtain dimensionality-reduced multi-channel enhancement features; and clustering the dimensionality-reduced multi-channel enhancement features based on K-means clustering to generate initial cluster labels. The initial cluster labels are smoothed using a moving average algorithm to generate the training labels.

[0050] Optionally, the method further includes: calculating and ranking the contribution of each feature in the historical multi-channel enhancement features in defect identification based on the out-of-bag data permutation importance analysis method, and obtaining a ranking result; based on the ranking result, selecting features with a contribution higher than a preset contribution from the historical multi-channel enhancement features to obtain the predetermined optimal feature subset.

[0051] In some embodiments, based on the out-of-bag data permutation importance analysis method, the 18-dimensional multi-channel enhancement features are sorted in descending order of contribution, and the top 10 features with the highest contribution are selected to obtain the optimal feature subset.

[0052] Specifically, the top 10 features, ranked from highest to lowest contribution, are: Y-gradient, X-gradient, Z-gradient, Y-peak-peak, X-peak-peak, Z-peak-peak, X-clearance factor, X-peak, Z-clearance factor, and X-E1 wavelet packet energy.

[0053] K-means clustering is performed with a set number of clusters K, and the optimal clustering result is selected after 5 iterations. For example, K can be 2, meaning that the multi-channel metal magnetic memory signal is divided into two types through clustering.

[0054] Defect clusters are identified and labeled based on different types of saliency indicators to generate initial pseudo-labels, which serve as training labels for subsequent random forest models. A moving average method is used to smooth the training labels to filter out isolated noise points and improve the continuity and reliability of the labels.

[0055] An out-of-bag data permutation importance analysis method is used to calculate the contribution of each feature in the multi-channel enhancement features to the defect identification process. Based on the contribution ranking results, the feature set is optimized, prioritizing the retention of core effective features. In this embodiment, 10 features are retained to form the optimal feature subset.

[0056] Optionally, the method further includes: constructing an initial random forest model; using the predetermined optimal feature subset as input features, using the training labels corresponding to the predetermined optimal feature subset as supervision labels, performing supervised learning training on the initial random forest model, and in the process of supervised learning training, performing global optimization of the hyperparameters of the initial random forest model based on the Bayesian optimization algorithm to obtain the pre-trained random forest model.

[0057] In the process of training the random forest model, the training labels generated by clustering and principal component analysis are used as the input features of the training data. Bayesian optimization is used to automatically search for and determine the hyperparameters of the random forest model, avoiding the limitation of traditional parameter optimization methods that are prone to getting trapped in local optima. The resulting random forest model can correct the missed and false judgments that may exist in the preprocessing stage and output high-precision and high-reliability defect identification results.

[0058] Optionally, the method further includes: determining the spatial location information of the defect in the target metal component based on the defect identification result; and outputting the spatial location information of the defect.

[0059] Based on the defect identification results, an inspection report is generated, which includes quantitative performance indicators of the model such as defect identification accuracy, precision, recall, and F1 score. Simultaneously, a visualization function is provided, overlaying defect interval markers onto the original three-channel signal time sequence diagram to clearly present the defect distribution, providing an intuitive and reliable reference for subsequent maintenance and repair work.

[0060] Table 1 compares the performance of the threshold-based random forest model in the prior art and the random forest model of this application in terms of accuracy, precision, recall, and F1 score. Experimental data shows that the random forest model of this application outperforms the model in terms of accuracy, precision, recall, and F1 score. The accuracy is improved to 98.16%, the precision to 96.97%, and the recall to 97.56%, reducing the risk of missed defect detection. The overall F1 score is also improved to 0.9726. These comparison results fully verify that the multi-dimensional feature enhancement, training label generation based on unsupervised clustering, and optimized random forest model adopted in this application can effectively improve the accuracy of defect identification of target metal components.

[0061] Table 1. Performance comparison results of the random forest model in this application and existing technologies. Figure 2 This is a schematic diagram illustrating the data flow of the method provided in the embodiments of this application. For example... Figure 2 As shown, in the signal acquisition stage, multi-channel metallic magnetic memory signals are collected by sensors. Subsequently, in the feature extraction stage, time-domain and frequency-domain features are extracted from the multi-channel metallic magnetic memory signals, and then fused to generate multi-channel enhanced features. In the defect identification stage, feature dimensionality reduction and initial clustering are performed sequentially to generate training labels, which are then input into a random forest model for intelligent decision-making. The process ends with the result output, where the random forest model outputs the final defect identification result. This flowchart clearly reveals the data flow from the original multi-channel metallic magnetic memory signals to the defect identification result.

[0062] Figure 3 A schematic diagram of a feature-enhanced metal magnetic memory signal defect identification system provided in this application embodiment is shown below. Figure 3 As shown, this application proposes a metal magnetic memory signal defect identification system based on feature enhancement. The system 300 includes: a signal acquisition module 310, a feature extraction module 320, and a defect identification module 330. The signal acquisition module 310 is used to acquire the multi-channel metal magnetic memory signal of the target metal component; In some embodiments, the signal acquisition module 310 includes a signal acquisition unit and a signal preprocessing unit.

[0063] The signal acquisition unit consists of an array of magnetic sensors, a signal conditioning circuit, and a data acquisition card. It is responsible for acquiring the metal magnetic memory signals of the X channel, Y channel, and Z channel of the wire rope, and transmitting the acquired signals to the processing unit.

[0064] The signal preprocessing unit incorporates an adaptive median filtering algorithm, a wavelet threshold denoising algorithm, and a signal standardization processing program. These are used to preprocess the acquired X-channel, Y-channel, and Z-channel metallic magnetic memory signals, remove various noise interferences, and output clean and stable preprocessed multi-channel metallic magnetic memory signals.

[0065] The feature extraction module 320 is used to extract multi-channel time-domain features and multi-channel frequency-domain features based on the multi-channel metal magnetic memory signal to obtain multi-channel enhanced features; and to select corresponding target features from the multi-channel enhanced features based on a pre-determined optimal feature subset to obtain the optimal feature vector. The defect identification module 330 is used to input the optimal feature vector into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

[0066] Optionally, the system further includes an output module; the output module is used to determine the spatial location information of the defect in the target metal component based on the defect identification result; and output the spatial location information of the defect.

[0067] The output module can visualize the spatial coordinates of the defects in the multi-channel metal magnetic memory signal, as well as the performance indicators of the model, making it convenient for staff to view.

[0068] This application proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement a method for identifying defects in metal magnetic memory signals based on feature enhancement.

[0069] In some embodiments, this application also provides a hardware-level structure for a computer device, which includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-described feature-enhanced metal magnetic memory signal defect identification method.

[0070] It should be noted that the descriptions of the above system embodiments and computer-readable storage medium embodiments are similar to the descriptions of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the system embodiments and computer-readable storage medium embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0071] It should be noted that, in the embodiments of this application, if the above-mentioned feature-enhanced metal magnetic memory signal defect identification method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0072] Correspondingly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps in any of the feature-enhanced metal magnetic memory signal defect identification methods described in the above embodiments. Correspondingly, embodiments of this application also provide a computer program product, which, when executed by a processor of an electronic device, is used to implement the steps in any of the feature-enhanced metal magnetic memory signal defect identification methods described in the above embodiments.

[0073] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0074] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0076] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.

[0077] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0078] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0079] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for identifying defects in metal magnetic memory signals based on feature enhancement, characterized in that, An application is made to a feature-enhanced metal magnetic memory signal defect identification system; the system includes: a signal acquisition module, a feature extraction module, and a defect identification module; the method includes: The signal acquisition module is used to acquire multi-channel metal magnetic memory signals of the target metal component; Using the feature extraction module, multi-channel time-domain features and multi-channel frequency-domain features are extracted based on the multi-channel metallic magnetic memory signal to obtain multi-channel enhanced features; based on a pre-determined optimal feature subset, corresponding target features are selected from the multi-channel enhanced features to obtain the optimal feature vector; Using the defect identification module, the optimal feature vector is input into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

2. The method according to claim 1, characterized in that, The multi-channel metal magnetic memory signal includes an X-channel metal magnetic memory signal, a Y-channel metal magnetic memory signal, and a Z-channel metal magnetic memory signal; acquiring the multi-channel metal magnetic memory signal of the target metal component includes: An array of magnetic sensors is deployed on the surface of the target metal component, and the array of magnetic sensors is moved at a constant speed to detect the metal magnetic memory signal of the X channel, the metal magnetic memory signal of the Y channel, and the metal magnetic memory signal of the Z channel. The metal magnetic memory signals of the X channel, Y channel, and Z channel are preprocessed to obtain the preprocessed multi-channel metal magnetic memory signals.

3. The method according to claim 2, characterized in that, The multi-channel time-domain features include peak-to-peak value, peak value, gap factor, and signal gradient; the multi-channel frequency-domain features include wavelet packet energy entropy and wavelet packet scale entropy; based on the multi-channel metallic magnetic memory signal, multi-channel time-domain features and multi-channel frequency-domain features are extracted respectively to obtain multi-channel enhancement features, including: Based on the preprocessed multi-channel metallic magnetic memory signal, the peak-to-peak value, peak value, gap factor, and signal gradient of each channel in the X channel, Y channel, and Z channel are extracted respectively. Based on the preprocessed multi-channel metallic magnetic memory signal, the wavelet packet energy entropy and wavelet packet scale entropy of each channel in the X channel, Y channel and Z channel are extracted respectively. The multi-channel enhancement features are obtained based on the peak-to-peak value, the peak value, the gap factor, the signal gradient, the wavelet packet energy entropy, and the wavelet packet scale entropy.

4. The method according to claim 1, characterized in that, The method further includes: Acquire historical multi-channel enhanced features; Based on the principal component analysis method, the multi-channel enhancement features are dimensionality reduced to obtain the dimensionality-reduced multi-channel enhancement features; Based on the K-means clustering method, the dimensionality-reduced multi-channel enhanced features are clustered to generate initial cluster labels; The initial cluster labels are smoothed using a moving average algorithm to generate the training labels.

5. The method according to claim 4, characterized in that, The method further includes: Based on the out-of-bag data permutation importance analysis method, the contribution of each feature in the historical multi-channel enhancement features to defect identification is calculated and ranked to obtain the ranking result; Based on the ranking results, features with a contribution higher than a preset contribution are selected from the historical multi-channel enhancement features to obtain the predetermined optimal feature subset.

6. The method according to claim 5, characterized in that, The method further includes: Construct an initial random forest model; The predetermined optimal feature subset is used as input features, and the training labels corresponding to the predetermined optimal feature subset are used as supervision labels to perform supervised learning training on the initial random forest model. During the supervised learning training process, the hyperparameters of the initial random forest model are globally optimized based on the Bayesian optimization algorithm to obtain the pre-trained random forest model.

7. The method according to claim 1, characterized in that, The method further includes: Based on the defect identification results, the spatial location information of the defects in the target metal component is determined; Output the spatial location information of the defect.

8. A metal magnetic memory signal defect identification system based on feature enhancement, characterized in that, The system includes: a signal acquisition module, a feature extraction module, and a defect identification module; The signal acquisition module is used to acquire the multi-channel metal magnetic memory signal of the target metal component; The feature extraction module is used to extract multi-channel time-domain features and multi-channel frequency-domain features based on the multi-channel metal magnetic memory signal to obtain multi-channel enhanced features; and to select corresponding target features from the multi-channel enhanced features based on a pre-determined optimal feature subset to obtain the optimal feature vector. The defect identification module is used to input the optimal feature vector into a pre-trained random forest model to obtain the defect identification result for the target metal component; wherein, the pre-trained random forest model is obtained by generating training labels through an unsupervised clustering algorithm based on unlabeled historical multi-channel enhanced features, and training based on the training labels.

9. The system according to claim 8, characterized in that, The system also includes an output module; The output module is used to determine the spatial location information of the defect in the target metal component based on the defect identification result; and output the spatial location information of the defect.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, is used to implement the method as described in any one of claims 1-7.

Images