Cold rolled steel hardness electromagnetic nondestructive testing method based on stack type auto-encoder
Through the combination of stack self-encoder and multi-magnetic detection technology, the problems of nonlinear coupling and electromagnetic interference in cold-rolled steel hardness detection are solved, and high-precision non-destructive detection is achieved, which improves detection efficiency and reduces costs.
Patent Information
- Application Number
- CN202510383690.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-08-15
AI Technical Summary
The prior art is difficult to achieve high-precision electromagnetic non-destructive detection in cold-rolled steel production, especially in hardness detection, the problem of insufficient detection accuracy and low efficiency caused by complex electromagnetic interference environments.
The stacked autoencoder (SAE) network is used to extract the electromagnetic signals in depth, and the martensite content of the material is obtained by combining multi-magnetic detection technology and electron backscattering diffraction (EBSD), establishing the correlation between hardness and electromagnetic characteristics, and realizing non-destructive detection.
The digitization and rapid evaluation of the hardness of ferromagnetic steel materials has been achieved, which improves detection efficiency and reduces costs.
Smart Images
Figure CN120489822A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of electromagnetic nondestructive testing of ferromagnetic materials, and in particular to an electromagnetic nondestructive testing method for cold-rolled steel hardness based on a stacked autoencoder. Background Art
[0002] In recent years, with the rapid development of intelligent manufacturing, electromagnetic-based nondestructive testing (NDT) has been widely used in the field of metal material performance evaluation due to its non-contact and high efficiency. Traditional cold-rolled steel hardness testing relies primarily on sampling destructive testing (such as the Vickers hardness tester) or linear regression models based on electromagnetic parameters. The former has problems with material integrity and low detection efficiency, while the latter has difficulty in achieving higher detection accuracy due to the nonlinear coupling relationship between the cold-rolled steel microstructure and the electromagnetic signal.
[0003] Existing electromagnetic detection technologies mostly use eddy current detection or magnetic Barkhausen noise method, and establish prediction models by measuring the empirical relationship between single electromagnetic parameters such as magnetic permeability and coercive force and hardness. However, fluctuations in process parameters such as rolling force and cooling rate during cold rolling will lead to enhanced magnetic anisotropy of the material, causing the electromagnetic response signal to exhibit significant non-stationary characteristics. At the same time, the cross-influence of multiple physical field factors such as surface residual stress and grain size distribution causes traditional feature extraction methods to have information loss problems. Studies have shown that when the hardness of the material is in the range of 400-600HV, the prediction error of the conventional BP neural network model can reach ±25HV, which is difficult to meet the needs of high-precision online detection.
[0004] Current research focuses on time-frequency domain signal fusion analysis, such as the use of wavelet packet transforms combined with support vector machines (SVMs). However, these methods still rely on manually designed features and have limited ability to represent complex electromagnetic signatures. While deep learning methods have achieved breakthroughs in image detection, they face challenges with gradient vanishing and overfitting when processing high-noise electromagnetic signals. In particular, the complex electromagnetic interference environment of cold rolling production lines results in signal-to-noise ratios (SNRs) of collected signals generally below 15dB, posing a significant challenge to the feature extraction capabilities of shallow neural networks.
[0005] Notably, the stacked autoencoder (SAE) effectively extracts deep nonlinear features from electromagnetic signals through a layer-by-layer greedy training strategy. Recent experimental data show that in bearing hardness testing, a three-layer SAE network suppresses high-frequency noise by approximately 40% compared to traditional filters, and achieves a feature dimension compression rate of over 80%. This provides a new technical approach for addressing signal denoising and feature abstraction in cold-rolled steel hardness testing, but its application in continuous cold-rolled strip production scenarios has not yet been seen. Summary of the Invention
[0006] The technical problem to be solved by the present invention is to address the defects involved in the background technology and provide an electromagnetic non-destructive testing method for cold-rolled steel hardness based on a stacked autoencoder. By studying the correlation between electromagnetic characteristics and microstructure and hardness, the feasibility of the electromagnetic non-destructive testing method for detecting hardness is verified. The stacked autoencoder is used to extract deep features from the electromagnetic signal to achieve rapid detection of the surface hardness of cold-rolled steel.
[0007] The present invention adopts the following technical solutions to solve the above technical problems:
[0008] The electromagnetic nondestructive testing method for cold-rolled steel hardness based on stacked autoencoders includes the following steps:
[0009] Step 1), obtaining the martensite content of the cold-rolled steel sample;
[0010] Step 1.1), cut a small piece from the cold-rolled steel sample for making the EBSD sample, and use the remaining part as the electromagnetic detection sample;
[0011] Step 1.2), mechanically polishing the cut small pieces to make them of appropriate size and smooth surface;
[0012] Step 1.3), electrolytically polishing the mechanically polished small piece using an electrolyte, and obtaining an EBSD sample after cleaning and drying;
[0013] In step 1.4), the EBSD sample is placed in a scanning electron microscope equipped with EBSD, and the data is processed using Channel 5 software to obtain the martensite content of the EBSD sample, which is used as the martensite content of the cold-rolled steel sample;
[0014] Step 2) Using an HVS-10 instrument, perform N Vickers hardness tests on the EBSD sample, and take the average value as the hardness of the cold-rolled steel sample, where N is a preset first threshold;
[0015] Step 3) Based on the multi-magnetic detection equipment, the electromagnetic nondestructive testing is performed on the electromagnetic detection sample using the MBN technology and the MIP technology at the same time to obtain the MBN original signal and the MIP original signal of the electromagnetic detection sample; the electromagnetic characteristics of the electromagnetic detection sample are extracted from the MBN original signal and the MIP original signal of the electromagnetic detection sample, and the electromagnetic characteristics are used as the electromagnetic characteristics of the cold-rolled steel sample; and the MBN original signal and the MIP original signal of the electromagnetic detection sample are used as the MBN original signal and the MIP original signal of the cold-rolled steel sample;
[0016] Step 4) Repeat steps 1) to 3) M times, where M is a preset second threshold, to obtain MBN raw signals, MIP raw signals, electromagnetic characteristics, hardness, and martensite content of M cold-rolled steel samples, and make a training data set to eliminate electromagnetic signal errors caused by unexpected factors;
[0017] Step 5), dividing the hardness of the cold-rolled steel into P continuous hardness intervals and representing them respectively using one-hot encoding, where P is a preset third threshold;
[0018] Step 6), using the MBN original signal and the MIP original signal of the cold-rolled steel as input and the one-hot encoding of the hardness range of the cold-rolled steel as output, a stacked autoencoder is built and trained based on the training dataset;
[0019] Step 7), realize online non-destructive testing of cold-rolled steel hardness based on the trained stacked autoencoder.
[0020] As a further optimization scheme of the cold-rolled steel hardness electromagnetic nondestructive testing method based on the stacked autoencoder of the present invention, when setting the scanning electron microscope equipment in step 1.4), the acceleration voltage is 20kV, the working distance is 16mm, and the tilt angle is 70°.
[0021] As a further optimization scheme of the cold-rolled steel hardness electromagnetic nondestructive testing method based on the stacked autoencoder of the present invention, the N is preferably set to 5, and the loading time of the Vickers hardness test is 15s.
[0022] As a further optimization scheme of the cold-rolled steel hardness electromagnetic nondestructive testing method based on the stacked autoencoder of the present invention, P is preferably taken as 4, and the four consecutive hardness intervals are 95HV-130HV, 130HV-170HV, 170HV-220HV, and >220HV, and the unique hot encodings are 0001, 0010, 0100, and 1000, respectively.
[0023] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0024] The present invention realizes non-destructive testing of the hardness of ferromagnetic steel materials, and provides a digital and rapid evaluation method for surface hardness testing of ferromagnetic steel materials, thereby improving testing efficiency and reducing testing costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A schematic diagram of the software interface of a multi-magnetic detection device in an embodiment of the present invention;
[0026] Figure 2 (a) Figure 2 (b) are schematic diagrams of signal butterfly of MBN and MIP in the embodiment of the present invention;
[0027] Figure 3 Schematic diagram of the relationship between martensite volume fraction and hardness in an embodiment of the present invention;
[0028] Figure 4 (a) Figure 4 (b) Figure 4 (c) Figure 4 (d) are schematic diagrams of the relationship between MBN root mean square, MBN peak-to-peak value, MBN coercive magnetic field, MIP coercive magnetic field and hardness in the embodiments of the present invention;
[0029] Figure 5 This is the process of mining hidden features by the stacked autoencoder in the implementation case of the present invention. DETAILED DESCRIPTION
[0030] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings:
[0031] The present invention can be implemented in many different forms and should not be considered to be limited to the embodiments described herein. On the contrary, these embodiments are provided to make this disclosure thorough and complete and will fully convey the scope of the invention to those skilled in the art. In the accompanying drawings, components are enlarged for clarity.
[0032] To achieve the above technical effects, the present invention proposes an electromagnetic nondestructive testing method for cold-rolled steel hardness based on a stacked autoencoder, comprising the following steps:
[0033] Step 1), obtaining the martensite content of the cold-rolled steel sample;
[0034] Step 1.1), cut a small piece from the cold-rolled steel sample for making the EBSD sample, and use the remaining part as the electromagnetic detection sample;
[0035] Step 1.2), mechanically polishing the cut small pieces to make them of appropriate size and smooth surface;
[0036] Step 1.3), electrolytically polishing the mechanically polished small piece using an electrolyte, and obtaining an EBSD sample after cleaning and drying;
[0037] In step 1.4), the EBSD sample is placed in a scanning electron microscope equipped with EBSD, and the data is processed using Channel 5 software to obtain the martensite content of the EBSD sample, which is used as the martensite content of the cold-rolled steel sample.
[0038] The scanning electron microscope was set up with an accelerating voltage of 20 kV, a working distance of 16 mm, and a tilt angle of 70°.
[0039] Step 2) Use an HVS-10 instrument to perform N Vickers hardness tests on the EBSD sample, and take the average value as the hardness of the cold-rolled steel sample, where N is a preset first threshold.
[0040] The N is preferably 5, and the loading time of the Vickers hardness test is 15 s.
[0041] Step 3) Based on the multi-magnetic detection equipment, the electromagnetic non-destructive testing is performed on the electromagnetic detection sample using the MBN technology and the MIP technology at the same time to obtain the MBN original signal and the MIP original signal of the electromagnetic detection sample; the electromagnetic characteristics of the electromagnetic detection sample are extracted from the MBN original signal and the MIP original signal of the electromagnetic detection sample, and are used as the electromagnetic characteristics of the cold-rolled steel sample; at the same time, the MBN original signal and the MIP original signal of the electromagnetic detection sample are used as the MBN original signal and the MIP original signal of the cold-rolled steel sample.
[0042] During testing, varying alternating magnetic fields are applied to the ferromagnetic material, from zero field excitation to positive saturation and back to zero field excitation, then increasing to negative saturation and back to zero field excitation. Throughout this repeated magnetization cycle, different microstructures undergo distinct magnetization processes. The cumulative effect of magnetic domain movement causes a change in the material's magnetic permeability, deepening the degree of magnetization, and ultimately shifting the overall magnetization curve. The signals obtained during this process can reflect the magnetic properties and microstructural characteristics of different ferromagnetic materials.
[0043] The software interface of the multi-magnetic detection device is as follows Figure 1 The probe inherits two micro-electromagnetic nondestructive testing technologies and selects 21 micro-electromagnetic features to characterize the magnetic parameters of cold-rolled steel.
[0044] In MBN technology, a high-amplitude, low-frequency sinusoidal current is fed into a yoke coil wrapped around a U-shaped magnetic yoke. To ensure that the signal of irreversible magnetic domain motion is detected by the receiving coil, the applied magnetization amplitude is sufficient to excite the ferromagnetic material being detected to a saturation level.
[0045] The detected MBN signal is processed by a combination of bandpass filter, low-pass / high-pass filter, and amplified, and also includes post-amplification and signal smoothing rectification. The MBN butterfly curve is as follows Figure 2 (a) shows the numerically transformed and smoothed MBN amplitude as the ordinal axis, with the excitation amplitude as the horizontal axis. From this, the maximum amplitude, MMAX, is derived as a test statistic. Accordingly, the magnetic field strength at MMAX is assigned to the test statistic, HCM. The expansion of the profile is evaluated at 25%, 50%, and 75% of MMAX (DH25M, DH50M, and DH75M). An additional test statistic is MMEAN, which is the mean value of the profile over a specific period.
[0046] During the reorganization of magnetic domains, Bloch walls are displaced, occurring as discrete jumps. Bloch walls exhibit different motion characteristics due to variations in microstructure. These microstructural changes can be reflected in the properties of the MBN. MMAX test statistics can be used to quantitatively assess finishing conditions, such as deep hardness and surface hardness.
[0047] Unlike the MBN method, in the MIP technique, high- and low-frequency sinusoidal currents are necessary to obtain information about the reversible domain motion. As with the MBN method, high-amplitude, low-frequency (10 Hz) excitation of the U-shaped yoke generates hysteresis loops in the material. Simultaneously, a low-amplitude (milliampere level), high-frequency (5 kHz) sinusoidal current is fed into the transmitter coil, similar to the MFEC method, to generate small asymmetric hysteresis loops that overlap with the main hysteresis curve.
[0048] Similar to MBN, the maximum amplitude (UMAX) of the MIP is extracted as an important feature. The magnetic field intensity at UMAX is also derived as a statistical parameter. In addition, the curve extensions of 25%, 50% and 75% (DH25U, DH50U and DH75U) and the average UMEAN of the time period are also used as MIP features. MIP can be used to characterize the material properties near the surface (surface hardening). The shell depth information comes from the amplitude of the UMAX signal received from the core structure, and the hardness information can be obtained from the related forced field intensity HCU. The stress state information is quantitatively described by the curve extension (DH25U, DH50U and DH75U).
[0049] Step 4) Repeat steps 1) to 3) M times, where M is a preset second threshold, to obtain MBN raw signals, MIP raw signals, electromagnetic characteristics, hardness, and martensite content of M cold-rolled steel samples, and make a training data set to eliminate electromagnetic signal errors caused by unexpected factors;
[0050] from Figure 3 It can be seen that hardness increases with the increase in the volume fraction of martensite. Compared to other structural parameters (such as austenite), martensite has more dislocations and interfaces, which hinder the movement of dislocations. When the amount of martensite in the material's crystal structure increases, the material's hardness also increases. In addition, the formation of martensite will lead to the generation of residual stress fields within the crystal, thereby increasing the hardness of the material. Therefore, it can be seen that there is a correlation between the electromagnetic characteristics of cold-rolled steel and its hardness.
[0051] MBN and MIP characteristics change with hardness, as shown in 4(a), Figure 4 (b) Figure 4 (c) Figure 4(d) It can be seen that with the increase of hardness, the MBN peak, RMS and MIP coercive force magnetic field decrease monotonically, while the MBN coercive force magnetic field shows an increasing trend. Figure 3 It can be found that increasing the martensite content leads to an increase in hardness, so the change in electromagnetic characteristics with hardness is essentially caused by changes in martensite. When the MVF increases, the increase in dislocation density hinders the movement of magnetic domains, thus affecting the hardness and electromagnetic properties of the material.
[0052] Step 5) The hardness of the cold-rolled steel is divided into P continuous hardness intervals and represented by one-hot encoding, where P is a preset third threshold.
[0053] P is preferably set to 4. The four consecutive hardness intervals and their corresponding one-hot encodings are shown in the following table:
[0054]
[0055] Step 6), using the MBN original signal and the MIP original signal of the cold-rolled steel as input and the one-hot encoding of the hardness range of the cold-rolled steel as output, a stacked autoencoder is built and trained based on the training dataset;
[0056] Build a stacked autoencoder, and set the input sample as the MBN and MIP original signal within one cycle. Through the encoding and decoding process, extract and learn richer feature information from the MBN and MIP original signals to enhance the accuracy and generalization ability of hardness grading. For the mining of hidden features such as Figure 5 shown.
[0057] Step 7), realize online non-destructive testing of cold-rolled steel hardness based on the trained stacked autoencoder.
[0058] In order to better evaluate the accuracy of the model prediction classification of the input samples, the accuracy (Precision), recall (Recall), and F1 score are used as evaluation indicators of the model prediction. The calculation formula is as follows:
[0059]
[0060] Where TP represents the number of samples correctly predicted as positive examples by the model, that is, the model correctly classifies the true positive samples as positive; TN represents the number of samples correctly predicted as negative examples by the model, that is, the model correctly classifies the true negative samples as negative; FP represents the number of samples that the model incorrectly predicts as negative samples, that is, the model mistakenly identifies the negative class as positive; FN represents the number of samples that the model incorrectly predicts as positive samples, that is, the model mistakenly identifies the positive class as negative; C is the total number of categories.
[0061] The cold-rolled steel electromagnetic prediction data is evaluated in combination with the actual value, as shown in the following table:
[0062]
[0063] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art and, unless defined as such, will not be interpreted in an idealized or overly formal sense.
[0064] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. An electromagnetic nondestructive testing method for cold-rolled steel hardness based on a stacked autoencoder, characterized in that: The following steps are involved: Step 1), obtaining the martensite content of the cold-rolled steel sample; In step 1.1, a small piece for making EBSD specimens was cut from the cold-rolled steel sample, and the remaining piece was used as the EM detection specimen. Step 1.2) Mechanically polish the cut pieces to make them of appropriate size and smooth surface; In step 1.3, the small piece after mechanical polishing is electrolytically polished using an electrolyte, and after cleaning and drying, an EBSD sample is obtained; In step 1.4, the EBSD sample is placed in a scanning electron microscope equipped with EBSD. The data is processed using Channel 5 software to obtain the martensite content of the EBSD sample, which is used as the martensite content of the cold-rolled steel sample. Step 2), using an HVS-10 instrument to perform N Vickers hardness tests on the EBSD sample, and taking the average value as the hardness of the cold-rolled steel sample, where N is a preset first threshold; Step 3), based on the multi-magnetic detection equipment, the electromagnetic non-destructive testing is performed on the electromagnetic detection sample using the MBN technology and the MIP technology at the same time to obtain the MBN original signal and the MIP original signal of the electromagnetic detection sample; the electromagnetic characteristics of the electromagnetic detection sample are extracted from the MBN original signal and the MIP original signal of the electromagnetic detection sample, and are used as the electromagnetic characteristics of the cold-rolled steel sample; and the MBN original signal and the MIP original signal of the electromagnetic detection sample are used as the MBN original signal and the MIP original signal of the cold-rolled steel sample; Step 4) repeats steps 1) to 3) M times, where M is a preset second threshold, and obtains MBN original signals, MIP original signals, electromagnetic characteristics, hardness, and martensite content of M cold-rolled steel samples, which are made into a training data set to eliminate electromagnetic signal errors caused by unexpected factors; Step 5), the hardness of the cold-rolled steel is divided into P continuous hardness intervals and represented by one-hot encoding, where P is a preset third threshold; Step 6) Using the MBN raw signal and the MIP raw signal of the cold-rolled steel as input and the one-hot encoding of the hardness range of the cold-rolled steel as output, a stacked autoencoder is built and trained based on the training dataset; Step 7) Based on the trained stacked autoencoder, online non-destructive testing of cold-rolled steel hardness is achieved.
2. The cold-rolled steel hardness electromagnetic nondestructive testing method based on stacked autoencoders according to claim 1, characterized in that: When setting up the scanning electron microscope in step 1.4), use an accelerating voltage of 20 kV, a working distance of 16 mm, and a tilt angle of 70°.
3. The cold-rolled steel hardness electromagnetic nondestructive testing method based on stacked autoencoders according to claim 1, characterized in that: The N is preferably set to 5, and the loading time of the Vickers hardness test is 15 s.
4. The cold-rolled steel hardness electromagnetic nondestructive testing method based on stacked autoencoders according to claim 1 is characterized in that: P is preferably set to 4, and the four consecutive hardness intervals are 95HV-130HV, 130HV-170HV, 170HV-220HV, and >220HV, and the one-hot encodings are 0001, 0010, 0100, and 1000 respectively.