Online magnesia carbon brick quality inspection method and system based on knocking detection

Abstract

The invention discloses a magnesia carbon brick online quality inspection method and system based on knocking detection, and relates to the field of acoustic detection, a reciprocating swing knocking mechanism driven by a stepping motor is adopted to carry out fixed-point knocking on the surface of a magnesia carbon brick according to a preset interval, and multichannel knocking sound wave signals are synchronously collected through a pickup array; time domain, frequency domain and time-frequency domain features are extracted from the preprocessed sound wave signals, and a fusion feature vector is constructed after normalization processing; and inputting the fusion feature vector into an attention mechanism enhanced extreme learning machine model, and outputting the defect type and confidence of the magnesia carbon brick. According to the invention, visual identification and self-adaptive regulation and control technologies are combined, real-time detection can be completed without stripping a production process, the detection rate is greatly improved, and continuous operation requirements of an automatic production line are met. Characteristic parameters of magnesia carbon brick materials are deeply fused into the whole process of detection parameter regulation and control, signal processing and model construction, and accurate recognition and grading of different types of defects are achieved by means of an intelligent model with an attention mechanism enhanced.

Description

Technical Field

[0001] This invention relates to the field of acoustic testing technology, and in particular to an online quality inspection method and system for magnesia-carbon bricks based on impact testing. Background Technology

[0002] Magnesia-carbon bricks, as a key refractory material in the metallurgical industry, are widely used in the lining of high-temperature smelting equipment such as converters and ladles. Their internal defects directly affect the service life of the equipment and the safety of smelting. Existing quality inspection methods for magnesia-carbon bricks are mainly divided into three categories: manual tapping inspection, offline instrument inspection, and traditional acoustic inspection.

[0003] Manual tapping inspection relies on operator experience and identifies defects by differences in the sound of tapping. This method is highly subjective, inconsistent, and makes it difficult to quantify defect parameters. Furthermore, it cannot meet the high-efficiency inspection requirements of automated production lines. Offline instrument inspection, such as industrial CT scans, can accurately acquire defect information, but it is inefficient, costly, and requires removing samples from the production line for inspection. This makes it impossible to achieve real-time online quality inspection and fails to meet the quality control requirements of large-scale continuous production.

[0004] While traditional acoustic impact testing has the potential for online applications, it suffers from significant technical limitations. Its testing parameters are mostly fixed, failing to dynamically adjust to the differences in the material properties of magnesia-carbon bricks, resulting in poor adaptability to different batches and specifications. The signal processing lacks targeted optimization, is susceptible to noise interference, and suffers from insufficient accuracy in extracting effective defect features. Furthermore, existing technologies do not establish a deep correlation between material properties, acoustic response, and defect states, resulting in weak defect identification models that struggle to accurately distinguish between different types of defects such as delamination, porosity, and air pockets. Moreover, the test results cannot effectively feed back into the production line for iterative parameter optimization, leading to a disconnect between quality inspection and production.

[0005] In summary, existing quality inspection methods for magnesia-carbon bricks suffer from problems such as low efficiency, poor adaptability, insufficient precision, and lack of closed-loop capability, making it difficult to meet the demands of modern magnesia-carbon brick production lines for efficient, accurate, and intelligent quality inspection. There is an urgent need for an online quality inspection solution that integrates material properties and intelligent detection technology. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an online quality inspection method and system for magnesia-carbon bricks based on impact testing. The technical solution adopted is as follows: The online quality inspection method for magnesia-carbon bricks based on impact testing includes the following steps: Step 1: Collect magnesia-carbon brick samples of different sizes, surface morphologies, and defects such as delamination, looseness, and pores. Use industrial CT scans to determine the location and size of defects, associate the samples with production process parameters, and establish an labeled sample library. Step 2: Obtain the real-time specifications of the magnesia-carbon brick to be tested through the visual recognition module, and automatically adjust the hammer parameters, striking speed and microphone array spacing of the striking mechanism. Step 3: A reciprocating pendulum striking mechanism driven by a stepper motor is used to strike the surface of the magnesium carbon brick at preset intervals, and multi-channel striking sound wave signals are collected synchronously through a microphone array. Step 4: Optimize the number of decomposition layers and penalty parameters of variational mode decomposition using a genetic algorithm, screen effective IMF components, and obtain the preprocessed acoustic signal through wavelet packet decomposition and threshold denoising; extract time-domain, frequency-domain and time-frequency-domain features from the preprocessed acoustic signal, and construct a fused feature vector after normalization. Step 5: Input the fused feature vectors into the attention mechanism-enhanced Extreme Learning Machine model and output the defect type and confidence level of the magnesium-carbon brick.

[0007] Optionally, step 6 is also included: constructing an acoustic response mapping model based on finite element simulation, converting the recognition results into a two-dimensional visual image, and dynamically adjusting the recognition threshold in combination with the real-time detection accuracy. Step 7: Feed the quality inspection results back to the production line control system to trigger the non-conforming product sorting mechanism, and at the same time send the test data back to the sample library to achieve iterative optimization of model parameters.

[0008] Optionally, step 1 includes the following sub-steps: Step 11: Select magnesia-carbon brick samples covering different production processes, batches, and specifications, and use industrial CT equipment to scan the samples to obtain the type, three-dimensional location coordinates, and geometric dimension information of their internal defects, which will be used as the true values ​​for defect labeling. Step 12: Perform a standardized tapping sound wave signal acquisition experiment on the sample, and record the sound wave signal, tapping position coordinates and environmental parameters at each tapping point. Step 13: Establish an associated database, with each record associated with the following information of a sample: production process parameters, defect information set annotated by industrial CT, acoustic signal dataset collected by impact experiment, and environmental parameters, forming an labeled sample library for model training and validation.

[0009] Optionally, in step 2, the visual recognition module uses deep learning to acquire the geometric dimensions and surface roughness of the magnesia-carbon brick in real time. ; The formula for calculating the microphone array spacing D is: ; in The longitudinal wave velocity in magnesia-carbon bricks. , For dynamic Young's modulus, Density; To set the center frequency for the tapping excitation, This is the layout coefficient; The adjustment of the striking speed v is based on: ; in H is the reference speed, and H is the thickness of the brick.

[0010] Optionally, the motion control equations for the stepper motor-driven reciprocating pendulum striking mechanism are as follows: ; in For striking force, For the acoustic impedance of magnesia-carbon bricks, satisfy ; The instantaneous speed of the hammer head. This is the acoustic impedance mismatch coefficient. denoted as the material attenuation coefficient, and d as the defect depth.

[0011] Optionally, in step 3, the strategy for fixed-point tapping is as follows: based on the rectangular outline of the magnesia-carbon brick, a grid-based planning method is used to determine the tapping point array, and the tapping point avoids the edge of the brick by a set distance; the microphone array includes a main microphone located directly above the tapping point and at least two auxiliary microphones located on the side of the brick, and the main and auxiliary microphones are triggered synchronously with the data acquisition card to ensure the timing consistency of the multi-channel signals.

[0012] Optionally, step 4 includes the following sub-steps: Step 41, define the fitness function F of the genetic algorithm as a weighted sum of envelope entropy and correlation coefficient: ; in, This is the sum of the envelope entropies of all IMF components after VMD decomposition. This represents the average correlation coefficient between each IMF component and the original signal. and These are the weighting coefficients; Step 42: Perform VMD decomposition on the original signal using optimized parameters to obtain K IMF components; Step 43: Calculate the correlation coefficient between each IMF component and the original signal. Set threshold ,reserve The IMF component is the valid component; Step 44: Perform wavelet packet decomposition on the selected valid IMF components, and use an improved threshold function for denoising. for: , ; ; in, These are wavelet coefficients. is the threshold, and a and b are adjustment parameters used to achieve a smooth transition of the threshold.

[0013] Optionally, in step 4, the extracted features include time-domain features, frequency-domain features, and time-frequency-domain features; Time-domain characteristics include short-time energy, root mean square, waveform factor, and peak factor; Frequency domain features include spectral centroid, spectral variance, and power spectral entropy; Time-frequency domain features include the energy proportion of each subband in wavelet packet decomposition and wavelet packet entropy; After Z-score normalization of all extracted features, principal component analysis is used for dimensionality reduction and fusion to form the final fused feature vector.

[0014] Optionally, in step 6, the acoustic response mapping model is constructed based on multi-physics coupling simulation of magnesia-carbon bricks, and the governing equations include thermal, mechanical, and acoustic coupling effects. The governing equations are: ; in It is the gradient operator. Let C be the displacement field and C be the stiffness tensor. It is the stiffness tensor C and the displacement gradient tensor The double dot product, Where is the coefficient of thermal expansion, and T is the temperature field. For temperature gradient, This is the impact load.

[0015] The online quality inspection system for magnesia-carbon bricks is used to realize an online quality inspection method for magnesia-carbon bricks based on impact testing. The system includes a material parameter measurement module, an adaptive control unit, a signal acquisition and processing unit, a multiphysics simulation engine, an intelligent decision-making unit, and a data closed-loop management unit. The material parameter measurement module includes a visual recognition module, an adjustable striking mechanism, and a microphone array, used to acquire the density ρ and dynamic Young's modulus E of magnesia-carbon bricks online. n The parameters are achieved using a combination of a non-contact ultrasonic thickness gauge and a density sensor. The adaptive control unit is communicatively connected to the material parameter measurement module and automatically adjusts the detection parameters based on the material parameters. The signal acquisition and processing unit is communicatively connected to the material parameter measurement module to perform online signal denoising, VMD decomposition, and feature extraction. The multiphysics simulation engine runs an acoustic response mapping model; The intelligent decision-making unit is equipped with an extreme learning machine model with enhanced attention mechanism to identify and determine the type and level of defects; The data closed-loop management unit enables the return of detection data, incremental updates of the model, and coordinated control of the production line.

[0016] In summary, the present invention has at least one of the following beneficial technical effects: This invention provides an online quality inspection method and system for magnesia-carbon bricks based on impact testing. Employing an online inspection architecture that combines visual recognition and adaptive control technology, it enables real-time inspection without requiring separation from the production process, significantly improving inspection speed and adapting to the continuous operation requirements of automated production lines. Regarding inspection accuracy, the method deeply integrates the material characteristics of magnesia-carbon bricks into the entire process of inspection parameter control, signal processing, and model construction, optimizing acoustic signal acquisition and feature extraction. Utilizing an attention-enhanced intelligent model, it achieves accurate identification and classification of different types of defects, reducing the false positive rate.

[0017] In terms of adaptability, by dynamically adjusting the striking parameters and the spacing of the microphones, the system can accommodate material differences in magnesia-carbon bricks of different specifications and batches, thereby improving its robustness to fluctuations in the production process. Regarding production collaboration, relying on multiphysics simulation and closed-loop data management, the system enables visualization of test results and coordinated control of the production line. Test data feeds back into the sample library and model iterations, forming a closed-loop mechanism of quality inspection-production-optimization, which helps improve the overall production quality of magnesia-carbon bricks. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the online quality inspection method for magnesia-carbon bricks based on impact testing according to the present invention. Figure 2 This is a schematic diagram of the architecture of the online quality inspection system for magnesium-carbon bricks according to a specific embodiment of the present invention; Figure 3 This is a schematic diagram comparing the average recognition accuracy of magnesia-carbon bricks at various tilt angles using different detection methods according to specific embodiments of the present invention. Figure 4 This is a schematic diagram showing a multi-dimensional comparison of the efficiency of adjusting the striking angle and the compatibility of the tilt angle of magnesia-carbon bricks in a specific embodiment of the present invention. Detailed Implementation

[0019] The present invention will be further described in detail below with reference to the accompanying drawings.

[0020] This invention discloses an online quality inspection method and system for magnesia-carbon bricks based on impact testing.

[0021] Reference Figures 1-4 Example 1, an online quality inspection method for magnesia-carbon bricks based on impact testing, includes the following steps: Step 1: Collect magnesia-carbon brick samples of different sizes, surface morphologies, and defects such as delamination, looseness, and pores. Use industrial CT scans to determine the location and size of defects, associate the samples with production process parameters, and establish an labeled sample library. Step 2: Obtain the real-time specifications of the magnesia-carbon brick to be tested through the visual recognition module, and automatically adjust the hammer parameters, striking speed and microphone array spacing of the striking mechanism. Step 3: A reciprocating pendulum striking mechanism driven by a stepper motor is used to strike the surface of the magnesium carbon brick at preset intervals, and multi-channel striking sound wave signals are collected synchronously through a microphone array. Step 4: Optimize the number of decomposition layers and penalty parameters of variational mode decomposition using a genetic algorithm, screen effective IMF components, and obtain the preprocessed acoustic signal through wavelet packet decomposition and threshold denoising; extract time-domain, frequency-domain and time-frequency-domain features from the preprocessed acoustic signal, and construct a fused feature vector after normalization. Step 5: Input the fused feature vectors into the attention mechanism-enhanced Extreme Learning Machine model and output the defect type and confidence level of the magnesium-carbon brick.

[0022] Example 2 also includes step 6, which involves constructing an acoustic response mapping model based on finite element simulation, converting the recognition results into a two-dimensional visual image, and dynamically adjusting the recognition threshold in conjunction with the real-time detection accuracy. Step 7: Feed the quality inspection results back to the production line control system to trigger the non-conforming product sorting mechanism, and at the same time send the test data back to the sample library to achieve iterative optimization of model parameters.

[0023] Example 3, step 1 includes the following sub-steps: Step 11: Select magnesia-carbon brick samples covering different production processes, batches, and specifications, and use industrial CT equipment to scan the samples to obtain the type, three-dimensional location coordinates, and geometric dimension information of their internal defects, which will be used as the true values ​​for defect labeling. Step 12: Perform a standardized tapping sound wave signal acquisition experiment on the sample, and record the sound wave signal, tapping position coordinates and environmental parameters at each tapping point. Step 13: Establish an associated database, with each record associated with the following information of a sample: production process parameters, defect information set annotated by industrial CT, acoustic signal dataset collected by impact experiment, and environmental parameters, forming an labeled sample library for model training and validation.

[0024] By adopting the above technical solution, step 1 constructs an annotated sample library to provide a benchmark for subsequent detection, uses industrial CT scans to obtain accurate defect information, and establishes the correspondence between defect status and acoustic signals and production processes. Essentially, it achieves quantitative calibration and data association of defect features, providing high-quality sample support for model training.

[0025] Step 2 uses visual recognition to obtain brick specifications and dynamically adjusts detection parameters. The core principle is to adapt to the geometry and material properties of different bricks, ensuring the efficiency of impact energy transmission and the integrity of sound wave acquisition, and avoiding signal distortion caused by fixed parameters.

[0026] Step 3 employs a stepper motor-driven reciprocating pendulum striking mechanism. Through fixed-point striking and multi-channel signal acquisition, the principle is to ensure the consistency of striking action and the comprehensiveness of signal acquisition. The main and auxiliary microphones work together to capture sound wave signals from different propagation paths, reducing the loss of information from a single acquisition angle.

[0027] The reciprocating pendulum striking mechanism is designed for online, targeted striking of magnesia-carbon bricks, as detailed below: Power source module: The core is a two-phase hybrid stepper motor, equipped with a high-precision driver and encoder, with stepless speed adjustment from 0-500 r / min and output torque of 5-20 N·m. The motor is fixed to the frame via a shock-absorbing base.

[0028] The transmission conversion assembly consists of a crank-connecting rod mechanism, a swing shaft, and a bearing housing, enabling the conversion of rotary motion into reciprocating oscillation.

[0029] The crank length is adjustable, the connecting rod is connected via a ball joint, and the swing shaft is equipped with a mechanical limit block and a photoelectric sensor, allowing for adjustable swing angle.

[0030] The striking assembly features a replaceable hammerhead with a built-in miniature force sensor.

[0031] The swing arm is made of carbon fiber composite material and is fixed to the swing shaft by a shrink sleeve.

[0032] The positioning and synchronization components include an incremental encoder at the end of the swing shaft, forming a dual positioning system with the motor encoder. A high-speed photoelectric switch is also included for synchronous triggering with the microphone array.

[0033] A disc spring buffer mechanism is provided between the hammer head and the swing arm, and the buffer stiffness is adjustable.

[0034] The bearing housing is installed via a cross slide, supporting fine-tuning of the X / Y plane and tilt adjustment to ensure that the hammer head strikes the brick perpendicularly.

[0035] The machine is mounted across the production line via a gantry frame, the height of which is adjustable to accommodate magnesia-carbon bricks of different thicknesses.

[0036] A stepper motor drives a crank to rotate, which in turn drives a swing arm to swing back and forth via a connecting rod. An encoder provides real-time position feedback, controlling the motor to precisely stop and strike at a preset striking point. A force sensor works in conjunction with the motor speed to dynamically adjust the striking force, and a synchronous trigger interface ensures that the timing of the striking and sound wave acquisition is consistent, adapting to the testing needs of magnesia-carbon bricks with different characteristics.

[0037] Step 4 optimizes the variational mode decomposition parameters using a genetic algorithm. The core of this step is to use the algorithm's adaptive capability to find the optimal signal decomposition scheme, accurately separate the effective signal from the noise, and then further suppress interference components and retain the core signal features related to the defects through wavelet packet decomposition and threshold denoising.

[0038] Step 5 extracts multi-dimensional features and constructs a fused feature vector. The principle is that different dimensional features can reflect defect characteristics from different perspectives. Time-domain features reflect the signal amplitude variation pattern, frequency-domain features reflect the signal frequency distribution differences, and time-frequency domain features take into account both time and frequency dimensions. Normalization processing can eliminate the influence of feature dimensions and improve the model's recognition stability. The attention-enhanced Extreme Learning Machine model works by strengthening the weights of defect-sensitive features through an attention mechanism, suppressing interference from irrelevant features, and relying on the rapid learning capability of the Extreme Learning Machine to achieve efficient and accurate output of defect type and confidence level.

[0039] Simulation mapping and closed-loop optimization construct a theoretical model through multiphysics simulation and dynamically iterate it using actual detection data to improve the system's detection robustness and production synergy. Finite element simulation constructs an acoustic response mapping model, which is essentially based on the thermo-mechanical-acoustic coupling effect. It simulates the acoustic response laws of magnesia-carbon bricks under different defect states, temperature environments, and impact loads, establishing a mapping relationship between theoretical acoustic signals and defect states, providing theoretical support for the visualization of detection results.

[0040] The identification results are transformed into a two-dimensional visual image. The principle is to use a simulation model to convert abstract acoustic signal features into an intuitive image of defect distribution, facilitating operators to quickly locate the defect's position and extent. The identification threshold is dynamically adjusted based on real-time detection accuracy. The core principle is to use feedback data to correct the judgment criteria, reducing the misjudgment rate caused by material batch differences and environmental interference, thus achieving adaptive optimization of the identification criteria.

[0041] By linking quality inspection results with the production line control system, the system enables rapid sorting of non-conforming products. At the same time, the inspection data is sent back to the sample library to supplement new sample information and defect characteristics, driving iterative updates of model parameters. This allows the inspection model to continuously adapt to changes in the production process, forming a virtuous cycle of inspection-production-optimization and improving the overall production quality control level.

[0042] By systematically collecting samples and correlating multi-dimensional data, a sample library that is both representative and complete is created, providing a reliable data foundation for subsequent detection and model training. Step 11 selects multiple batches and specifications of samples and calibrates defects using industrial CT scans. The core is to utilize the penetration and precision of industrial CT to obtain the true three-dimensional information of the defects, which serves as the ground truth for defect annotation, avoiding the subjective errors of manual annotation and ensuring the accuracy of sample annotation.

[0043] Step 12 involves conducting standardized impact experiments and recording multi-dimensional data. The principle is to control variables such as impact force, location, and environment to ensure the consistency and comparability of acoustic signal acquisition, eliminate interference from experimental variables on signal characteristics, and ensure that the acquired acoustic signals accurately reflect the correlation between defect states and material properties. Step 13 establishes a relational database. The principle is to integrate multi-dimensional data such as production process parameters, defect information, acoustic signals, and environmental parameters to explore the influence of different factors on the acoustic properties and defect states of magnesia-carbon bricks. This provides comprehensive feature input for model training and improves the model's adaptability to different production scenarios and defect types.

[0044] In Example 4, in step 2, the visual recognition module uses deep learning to acquire the geometric dimensions and surface roughness of the magnesia-carbon brick in real time. ; The formula for calculating the microphone array spacing D is: ; in The longitudinal wave velocity in magnesia-carbon bricks. , For dynamic Young's modulus, Density; To set the center frequency for the tapping excitation, This is the layout coefficient; The adjustment of the striking speed v is based on: in H is the reference speed, and H is the thickness of the brick.

[0045] Example 5: The motion control equations for the stepper motor-driven reciprocating pendulum striking mechanism are as follows: ; in For striking force, For the acoustic impedance of magnesia-carbon bricks, satisfy ; The instantaneous speed of the hammer head. This is the acoustic impedance mismatch coefficient. denoted as the material attenuation coefficient, and d as the defect depth.

[0046] In Example 6, step 3, the strategy for fixed-point tapping is as follows: based on the rectangular outline of the magnesia-carbon brick, a grid-based planning method is used to determine the tapping point array, and the tapping point avoids the edge of the brick by a set distance; the microphone array includes a main microphone located directly above the tapping point and at least two auxiliary microphones located on the side of the brick. The main and auxiliary microphones are triggered synchronously with the data acquisition card to ensure the timing consistency of multi-channel signals.

[0047] By adopting the above technical solution, the core of the formula for calculating the spacing of the microphone array is to determine a reasonable spacing by combining the longitudinal wave velocity of the magnesia-carbon brick with the impact center frequency. The longitudinal wave velocity is determined by the dynamic Young's modulus and density, directly reflecting the internal density and mechanical properties of the material. The layout coefficient can be flexibly adjusted according to the detection accuracy requirements. This formula enables the microphone to cover the effective propagation range of the sound wave, ensuring the capture of complete defect-related sound wave signals and avoiding signal loss due to excessive spacing or data redundancy due to insufficient spacing.

[0048] The formula for adjusting the striking speed works on the principle of balancing striking energy with the characteristics of the brick. The greater the brick thickness, the higher the density, and the smaller the dynamic Young's modulus, the higher the required striking speed is to ensure sufficient energy to excite sound waves. Based on a reference speed, the striking speed is adapted to different brick characteristics through coupled calculations of material and geometric parameters. This avoids both excessively low speeds that result in weak sound signals and excessively high speeds that damage the brick or generate interference signals.

[0049] The motion control of the striking mechanism optimizes the energy transfer efficiency of the striking action through acoustic impedance matching, thereby improving the sensitivity of the acoustic wave response to defects. Acoustic impedance is the material's resistance to sound wave propagation, which is directly related to density and longitudinal wave velocity. The control equation of the striking mechanism incorporates acoustic impedance parameters, essentially adapting the hammer striking force to the acoustic properties of the magnesia-carbon brick, reducing the reflection loss of striking energy at the contact surface, and maximizing the energy transfer to the interior of the brick.

[0050] In the equation, the acoustic impedance mismatch coefficient quantifies the degree of energy reflection at the contact surface between the hammer and the brick, the material attenuation coefficient reflects the energy loss of sound waves propagating inside the brick, and the defect depth affects the intensity of sound wave reflection and attenuation. By integrating these parameters and dynamically adjusting the striking force, the striking output can be optimized according to the brick material characteristics and potential defect states, so that defects of different depths and types can elicit significant sound wave responses, while avoiding excessive striking that could damage the brick, thus balancing detection effectiveness and brick integrity.

[0051] The striking strategy and microphone array layout improve the consistency and comprehensiveness of the acoustic signal through standardized striking and multi-channel synchronous acquisition, providing high-quality data for subsequent defect identification. The principle of the rectangular contour-based gridded striking dot matrix is ​​to achieve uniform coverage detection on the brick surface, ensuring no blind spots. Edge areas are avoided because stress concentration effects exist at edges, and the acoustic waves generated by striking are easily distorted by boundary interference, affecting the accuracy of defect judgment.

[0052] The layout design of the main and auxiliary microphones is based on the principle of capturing sound wave signals from multiple angles. The main microphone collects the direct sound wave directly above the impact point, while the auxiliary microphone collects reflected and transmitted sound waves propagating from the sides. Sound wave signals from different paths can complement each other to reflect the internal defect state of the brick. The main and auxiliary microphones are triggered synchronously with the data acquisition card. The core principle is to ensure the consistency of the time base of the multi-channel signals, avoiding signal feature misalignment caused by timing deviations, and providing a reliable guarantee for subsequent multi-channel signal fusion analysis and defect localization.

[0053] Example 7, step 4 includes the following sub-steps: Step 41, define the fitness function F of the genetic algorithm as a weighted sum of envelope entropy and correlation coefficient: ; in, This is the sum of the envelope entropies of all IMF components after VMD decomposition. This represents the average correlation coefficient between each IMF component and the original signal. and These are the weighting coefficients; Step 42: Perform VMD decomposition on the original signal using optimized parameters to obtain K IMF components; Step 43: Calculate the correlation coefficient between each IMF component and the original signal. Set threshold ,reserve The IMF component is the valid component; Step 44: Perform wavelet packet decomposition on the selected valid IMF components, and use an improved threshold function for denoising. for: , ; ; in, These are wavelet coefficients. is the threshold, and a and b are adjustment parameters used to achieve a smooth transition of the threshold.

[0054] By adopting the above technical solution, step 41 designs the fitness function and optimizes the parameters using a genetic algorithm. The principle is to construct the optimization objective by combining signal complexity and correlation indicators. The envelope entropy reflects the complexity of the IMF components; the smaller the value, the more regular the components and the higher the proportion of effective signal. The mean correlation coefficient reflects the correlation between the components and the original signal; the higher the value, the more the components contain the core information of the original signal. The weighted sum form of the fitness function can balance the requirements of both. The genetic algorithm iteratively searches for the optimal decomposition level and penalty parameters, essentially adaptively finding the most suitable decomposition scheme for the current acoustic signal, so that VMD decomposition can fully separate signal and noise while avoiding feature loss due to over-decomposition.

[0055] Step 42 utilizes optimized parameters for VMD decomposition. The principle is to improve the targeting and accuracy of the decomposition based on optimal parameters. The number of decomposition layers determines the dimensions of the signal being split, while the penalty parameter affects the compactness and noise suppression capability of the decomposition. The synergistic optimization of both allows the IMF components obtained from the decomposition to more accurately correspond to signal components in different frequency ranges, providing a high-quality foundation for subsequent effective component selection.

[0056] Step 43 filters effective components using correlation coefficients. The principle is to eliminate redundant components with weak correlation to the original signal. These components are mostly noise or irrelevant interference signals. Retaining components with correlation coefficients higher than the threshold can reduce the interference of invalid information on subsequent processing, focus on the core signals that carry the defect characteristics, and improve processing efficiency and accuracy.

[0057] Step 44 employs an improved threshold function for noise reduction. The core principle is to address the problems of signal abrupt changes in traditional hard threshold functions and constant deviations in soft threshold functions. Adjusting parameters allows for a smooth threshold transition, preserving the detailed characteristics of the acoustic signal to the greatest extent possible while suppressing noise. When the absolute value of the wavelet coefficients is higher than the threshold, a smoothing formula is used to avoid abrupt changes in signal amplitude; when it is lower than the threshold, it is directly set to zero to eliminate noise. Ultimately, a balance is achieved between noise suppression and signal fidelity, ensuring that the denoised signal accurately reflects the acoustic characteristics corresponding to the internal defects of the magnesia-carbon brick.

[0058] In Example 8, step 4 extracts features including time-domain features, frequency-domain features, and time-frequency-domain features; Time-domain characteristics include short-time energy, root mean square, waveform factor, and peak factor; Frequency domain features include spectral centroid, spectral variance, and power spectral entropy; Time-frequency domain features include the energy proportion of each subband in wavelet packet decomposition and wavelet packet entropy; After Z-score normalization of all extracted features, principal component analysis is used for dimensionality reduction and fusion to form the final fused feature vector.

[0059] In Example 9, in step 6, the acoustic response mapping model is constructed based on the multi-physics coupling simulation of magnesium-carbon bricks. The governing equations include thermal, mechanical, and acoustic coupling effects, and the governing equations are: ; in It is the gradient operator. Let C be the displacement field and C be the stiffness tensor. It is the stiffness tensor C and the displacement gradient tensor The double dot product, Where is the coefficient of thermal expansion, and T is the temperature field. For temperature gradient, This is the impact load.

[0060] By adopting the above technical solutions, time-domain feature extraction focuses on the changes in signal amplitude and shape. Short-time energy reflects local energy fluctuations in the signal and can accurately capture sudden changes in sound wave energy caused by defects. Root mean square quantization measures the average level of signal amplitude, reflecting the degree of energy attenuation during sound wave propagation. Waveform factor and peak factor reflect the distortion characteristics of the signal waveform. The presence of defects will change the sound wave propagation path, leading to a decrease in waveform regularity. These two types of factors can effectively characterize this difference.

[0061] Frequency domain feature extraction revolves around the frequency distribution characteristics of the signal. The spectral centroid reflects the frequency location where the signal energy is concentrated. Defects can cause the frequency components of the sound wave to shift, resulting in regular changes in the spectral centroid. The spectral variance quantifies the dispersion of the frequency distribution, reflecting the difference in the impact of defects on sound waves of different frequencies. The power spectral entropy reflects the disorder of the frequency distribution. The presence of defects increases the complexity of the frequency distribution, causing the power spectral entropy value to rise.

[0062] Combining time and frequency domain features with information from both time and frequency dimensions, and based on the sub-band energy ratio of wavelet packet decomposition, the energy changes in different frequency ranges can be accurately located. The attenuation effect of defects on sound waves of specific frequencies is more significant, and the defect features can be locked by the difference in sub-band energy ratio. Wavelet packet entropy quantifies the complexity of the time-frequency space, further enhancing the signal distortion features caused by defects.

[0063] Z-Score normalization aims to eliminate dimensional differences between features, preventing the model from becoming overly biased towards a particular feature due to significant variations in feature values, and ensuring balanced weights across all features during model training. Principal component analysis (PCA) dimensionality reduction and fusion focuses on eliminating redundant information between features, retaining the core components that best reflect differences in defects. This simplifies feature vector dimensions, improves model computational efficiency, enhances generalization ability, and constructs a fused feature vector that is both representative and concise.

[0064] The acoustic response mapping model is constructed based on the multi-physics coupling effect, accurately simulating the acoustic response law of magnesia-carbon bricks under actual working conditions. This provides theoretical support for the visualization of test results and threshold adjustment, and solves the problem of the disconnect between single-physics simulation and actual working conditions.

[0065] In the actual production and testing process, magnesia-carbon bricks are not only subjected to impact loads. Changes in the temperature field will cause thermal expansion of the brick body, which will generate internal stress and affect the mechanical properties of the material and the propagation law of sound waves. Heat, force and sound are interrelated and mutually influential. Multiphysics coupling simulation can completely reproduce this complex process.

[0066] The governing equations construct a multi-field coupling relationship from the perspective of mechanical equilibrium. The left-hand side reflects the change of inertial force of the brick under the action of sound waves, characterizing the dynamic characteristics of the acoustic response. The first term on the right-hand side describes the influence of the elastic deformation of the brick on the displacement field through the stiffness tensor, reflecting the relationship between the mechanical field and the displacement field. The second term converts the temperature field change into additional stress through the coefficient of thermal expansion, reflecting the effect of the temperature field on the mechanical properties. The third term is the impact load, which serves as the excitation source of the acoustic response, connecting the force field and the sound field.

[0067] This control equation establishes a quantitative correlation between the temperature field, impact load, brick displacement field, and acoustic response, accurately simulating the sound wave propagation law under different defect states, temperature environments, and impact parameters. This ensures that the simulated acoustic response closely matches the actual detection signal, providing a theoretical basis for visualizing the identification results. It also provides simulation support for dynamically adjusting the identification threshold, further improving the accuracy and robustness of the detection system.

[0068] Example 10: Online quality inspection system for magnesia-carbon bricks, used to implement an online quality inspection method for magnesia-carbon bricks based on impact testing. The system includes a material parameter measurement module, an adaptive control unit, a signal acquisition and processing unit, a multiphysics simulation engine, an intelligent decision-making unit, and a data closed-loop management unit. The material parameter measurement module includes a visual recognition module, an adjustable striking mechanism, and a microphone array, used to acquire the density ρ and dynamic Young's modulus E of magnesia-carbon bricks online. n The parameters are achieved using a combination of a non-contact ultrasonic thickness gauge and a density sensor. The adaptive control unit is communicatively connected to the material parameter measurement module and automatically adjusts the detection parameters based on the material parameters. The signal acquisition and processing unit is communicatively connected to the material parameter measurement module to perform online signal denoising, VMD decomposition, and feature extraction. The multiphysics simulation engine runs an acoustic response mapping model; The intelligent decision-making unit is equipped with an extreme learning machine model with enhanced attention mechanism to identify and determine the type and level of defects; The data closed-loop management unit enables the return of detection data, incremental updates of the model, and coordinated control of the production line.

[0069] This invention relates to an online quality inspection system for magnesia-carbon bricks used in converters. The method and system enable efficient and accurate quality inspection of magnesia-carbon bricks of different specifications and defect types. It also verifies the effect of optimized tapping angle on improving inspection performance. The specific implementation process is as follows: Sample settings: The total sample size is 350, covering 10 specifications and involving 5 production process batches, to ensure the representativeness of the samples; Defect types and quantities: 50 defect-free bricks, 80 delamination bricks with crack widths of 0.1 to 2 mm, 80 loose bricks with porosity of 5 to 20%, 90 pores with pore diameters of 0.05 to 1 mm, and 50 newly added inclined magnesia-carbon bricks with an inclination of 0.5 to 5 degrees to simulate the posture deviation caused by the offset of the conveyor rollers in the production line. Defect identification: All samples were scanned by industrial CT with a resolution of 0.01 mm to obtain defect type, three-dimensional location coordinates, and geometric dimension information, which served as the true basis for defect identification. Sample association information: Records the production process parameters and environmental parameters of each sample, which are used for sample library construction and model training.

[0070] Environmental conditions: The real-time temperature of the production line is 25 to 80 degrees Celsius and the ambient noise is 60 to 85 decibels, which is consistent with the actual production conditions. No additional constant temperature and humidity environment is built to ensure the practicality of the test results. Equipment configuration: Visual recognition module: adopts a convolutional neural network model, with a size measurement error of no more than 0.1 mm, a surface roughness measurement range of 0.1 to 10 micrometers, and a tilt detection error of no more than 0.1 degrees; The striking mechanism features a two-phase hybrid stepper motor equipped with a high-precision driver and encoder, offering stepless speed adjustment from 0 to 500 rpm and output torque from 5 to 20 Nm. The cross slide supports ±3 degrees of tilt adjustment and ±5 mm of X / Y plane position fine-tuning. The hammerhead diameter ranges from 8 to 20 mm and can be made of carbide or polyurethane. It incorporates a built-in miniature force sensor with a measurement range of 0 to 500 N and an accuracy of ±1%. Microphone array: It contains one main microphone and two auxiliary microphones. The main microphone is located directly above the impact point, and the auxiliary microphones are located on the side of the brick. The sampling frequency is 48 kHz, and the timing error of the synchronous trigger with the data acquisition card does not exceed 10 microseconds. Intelligent decision-making unit: Equipped with an extreme learning machine model enhanced with attention mechanism, trained for 500 iterations, with a learning rate of 0.01, and attention weight coefficients determined through optimization using sample library data.

[0071] Comparison objects: traditional manual tapping test, traditional acoustic tapping test, and industrial CT offline test; traditional manual tapping test is performed by operators with more than 5 years of experience, traditional acoustic tapping test uses fixed parameters and the tapping angle is not adjustable, and industrial CT offline test is an industry-recognized accurate test method. Evaluation indicators include: defect identification accuracy, single-block inspection time, tilted brick adaptation rate, and frequency of manual intervention. Among them, the defect identification accuracy is statistically classified according to defect type, which comprehensively verifies the accuracy and convenience of the present invention.

[0072] Sample library construction: 350 samples were selected, and defect true value information was obtained through industrial CT scanning. A defect annotation library was established, which includes key parameters such as defect type, three-dimensional coordinates, and size. Under standardized conditions, a tapping experiment was conducted on the sample. The tapping mechanism of the present invention was used to tap at a fixed angle of 90 degrees according to a gridding strategy, and the sound wave signal, position coordinates and environmental parameters of each tapping point were recorded. Establish an associated database, with each record containing the associated sample's production process parameters, CT defect true values, acoustic signal dataset, and environmental parameters, forming an labeled sample library for model training and validation.

[0073] Detection parameters are adaptively adjusted: The visual recognition module collects the geometric dimensions, surface roughness, and tilt of the magnesia-carbon brick to be inspected in real time and transmits them to the adaptive control unit. Adjusting detection parameters: Based on the information obtained online by the material parameter measurement module, the spacing of the microphone array and the striking speed are adjusted; for tilted bricks, the adaptive control unit drives the cross slide to correct the tilt angle of the striking mechanism to a vertical state, and at the same time, the hammer material is adjusted, with hard alloy used for high-density bricks and polyurethane used for bricks with fragile surfaces.

[0074] Acoustic signal acquisition: Tapping strategy: Based on the rectangular outline of magnesia-carbon bricks, a grid-planned tapping dot matrix is ​​adopted. The grid spacing is adjusted according to the brick size. The tapping points avoid the brick edge by 10 mm to avoid signal distortion caused by edge stress concentration. Signal acquisition: The stepper motor drives the reciprocating pendulum striking mechanism to strike at preset points. The main and auxiliary microphones synchronously acquire multi-channel sound wave signals. The acquisition time is 50 milliseconds for each striking point. The data acquisition card converts the signal into a digital signal and then transmits it to the signal acquisition and processing unit.

[0075] Signal preprocessing and feature extraction: Signal optimization: The number of decomposition layers and penalty parameters of variational mode decomposition are optimized using a genetic algorithm to screen effective IMF components; Denoising: Wavelet packet decomposition is performed on the effective IMF components, and a modified threshold function is used for denoising; Feature construction: Extract time-domain features, frequency-domain features, and time-frequency-domain features. Time-domain features include short-time energy, root mean square, waveform factor, and peak factor. Frequency-domain features include spectral centroid, spectral variance, and power spectral entropy. Time-frequency-domain features include the energy proportion of each sub-band in wavelet packet decomposition and wavelet packet entropy. After normalization, dimensionality reduction is performed using principal component analysis to construct a fused feature vector.

[0076] Defect identification and result output: The extreme learning machine model is enhanced by fusion feature vector input and attention mechanism. The model strengthens the defect-sensitive features through attention weights and outputs the defect type and confidence level. The defect types include no defects, delamination, looseness and porosity. The confidence threshold is initially set to 0.85. When the output confidence reaches or exceeds 0.85, the defect type is directly determined. When it is below 0.85, multiphysics simulation calibration is triggered.

[0077] Simulation calibration and closed-loop optimization: Multiphysics simulation: Based on the multiphysics coupling control equation, an acoustic response mapping model is constructed. The material parameters, temperature field and impact load of the brick to be tested are input to simulate the sound wave propagation law and correct the recognition results. Visualization and Threshold Adjustment: The recognition results are converted into two-dimensional visual images, and the location and extent of defects are marked; the confidence threshold is dynamically adjusted in combination with the real-time detection accuracy. Closed-loop management: Quality inspection results are fed back to the production line control system, and non-conforming products trigger the sorting mechanism; at the same time, the test data is sent back to the sample library to achieve incremental iterative optimization of model parameters.

[0078] Defect identification accuracy verification: Based on the blind test results of 350 samples, the defect identification accuracy of the system of this invention and the traditional solution were compared. The data are shown in Table 1: Table 1 brick tilt Testing Plan Defect-free identification accuracy accuracy of split fracture identification Loose identification accuracy Pore ​​identification accuracy Average accuracy False positive rate 0 degrees (horizontal) tapping at a fixed angle 90% 85% 82% 78% 83.75% 16.25% 0 degrees (horizontal) Traditional acoustic percussion testing 88% 80% 72% 63% 75.75% 24.25% 0 degrees (horizontal) Manual tapping test 82% 75% 68% 55% 70% 30% 0 degrees (horizontal) This invention system 99% 98% 97% 99% 98.25% 1.75% 1 degree tilt tapping at a fixed angle 82% 76% 70% 65% 73.25% 26.75% 1 degree tilt This invention system 99% 97% 96% 98% 97.5% 2.5% 3-degree tilt tapping at a fixed angle 75% 68% 62% 58% 65.75% 34.25% 3-degree tilt This invention system 98% 96% 95% 97% 96.5% 3.5% In the scenario of horizontal bricks, the average accuracy of the system of this invention reaches 98.25%, which is 14.5 percentage points higher than that of fixed-angle tapping and 22.5 percentage points higher than that of traditional acoustic detection. The core advantage lies in the accurate capture of defect signals by multi-dimensional feature fusion and attention mechanism model. In scenarios involving tilted bricks, the accuracy of tapping at a fixed angle drops sharply as the tilt increases, reaching only 65.75% at a 3-degree tilt. However, this invention maintains an accuracy of over 96.5% through adaptive vertical tapping, and its accuracy in identifying tiny pores is far superior to that of the fixed-angle solution, avoiding sound wave signal distortion and energy loss caused by angle deviation.

[0079] Using magnesia-carbon bricks with dimensions of 200 mm × 200 mm × 300 mm and an inclination of 2 degrees as the test object, the full-process test indicators for each scheme are shown in Table 2: Table 2 Evaluation Dimensions Fixed-angle tapping (traditional method) Industrial CT offline inspection Manual tapping test This invention system Performance improvement (vs. fixed angle) Single block detection time 15 seconds 30 minutes 30 seconds 3 seconds 80% Frequency of human intervention Each piece needs to be manually checked for tilt and adjusted. Each piece requires manual transport and positioning. Each piece requires manual inspection. 0 times 100% Tilt adaptation range Only supports 0 degrees (horizontal). Only supports 0 degrees (horizontal). No clear scope of compatibility 0 to 5 degrees tilt ＞500% Continuous production compatibility rate (100 units) 62% 0% 75% 100% 38% The system of this invention takes only 3 seconds to inspect a single block, which is fully compatible with the conveying speed of 2 meters per minute on the production line. It can realize continuous online quality inspection, shortening the time by 80% compared with fixed angle tapping and improving the efficiency by 99% compared with industrial CT offline inspection. Magnesia-carbon bricks in the production line often have slight tilting due to wear of the conveyor rollers and stacking deviation. This invention achieves 100% compatibility through visual recognition and automatic angle adjustment of the cross slide. Ordinary operators can start working after 1 hour of training, which greatly reduces labor costs and the operating threshold.

[0080] The core technology supporting the tapping angle optimization in this embodiment stems from the structural design and control logic of this invention: Hardware foundation: The cross slide assembly of the striking mechanism supports tilt adjustment and micron-level fine-tuning to ensure the accuracy of angle adjustment; Control logic: The visual recognition module detects the brick tilt in real time, and the adaptive control unit outputs adjustment commands based on the detection data. In coordination with the motion control equation of the stepper motor, it realizes closed-loop control of tilt detection, angle correction and impact force optimization. Signal and model synergy: Vertical striking ensures efficient transmission of sound wave energy along the longitudinal direction of the brick, reducing lateral dispersion and increasing the amplitude of defect-related sound wave signals by 30% to 50%. This provides high-quality data support for subsequent feature extraction and model recognition, ultimately achieving a balance between detection accuracy and convenience.

[0081] Appendix Figure 3 This is a comparison chart of defect identification accuracy under different brick inclination angles; the horizontal axis represents the brick inclination angle, in degrees, ranging from 0 to 5 degrees, with 1-degree intervals; the vertical axis represents the average identification accuracy, in percentages, ranging from 60% to 100%, with 5% intervals. The two broken lines represent fixed-angle tapping and the adaptive vertical tapping of the present invention, respectively. The fixed-angle tapping corresponds to the solid line, and the adaptive vertical tapping of the present invention corresponds to the dashed line. Key data points are marked: 83.75% vs. 98.25% at 0 degrees and 65.75% vs. 96.5% at 3 degrees, highlighting the accuracy advantage of the present invention in tilted scenarios. At the same time, the critical point of 1 degree angle deviation is marked to reflect the performance mutation characteristics of the fixed-angle solution.

[0082] Appendix Figure 4 This is a comparison chart of the tapping angle adjustment efficiency and the adaptation rate; the horizontal axis represents the detection scheme, including fixed-angle tapping and the adaptive tapping of this invention; the left vertical axis represents the adjustment time in seconds, ranging from 0 to 15 seconds, with 3-second intervals; the right vertical axis represents the adaptation rate in percentages, ranging from 0 to 100%, with 20% intervals. The thick bars on the left represent adjustment time, approximately 13.5 seconds for a fixed angle and approximately 0.4 seconds for this invention; the thin bars on the right represent the adaptation rate, 62% for a fixed angle and 100% for this invention; the production line speed threshold of 3 seconds per piece is marked, visually demonstrating the efficiency advantage and adaptability of this invention, meeting the needs of continuous production.

[0083] Magnesia-carbon brick online quality inspection system: The system integrates various functional modules through an industrial IoT platform. The specific composition and workflow are as follows: The material parameter measurement module consists of a vision recognition module, an adjustable striking mechanism, and a microphone array. It is equipped with a non-contact ultrasonic thickness gauge and a density sensor to acquire parameters such as the density and dynamic Young's modulus of magnesia-carbon bricks online and transmit them to the adaptive control unit in real time.

[0084] The adaptive control unit communicates with the material parameter measurement module. Based on the acquired material parameters and geometric dimensions, it automatically adjusts the detection parameters such as the pickup array spacing, striking speed, and striking force to ensure detection adaptability.

[0085] The signal acquisition and processing unit is connected to the material parameter measurement module to synchronously acquire multi-channel acoustic signals, perform online signal denoising, VMD decomposition and feature extraction, and output fused feature vectors to the intelligent decision-making unit.

[0086] The multiphysics simulation engine runs the acoustic response mapping model, receives the recognition results from the intelligent decision-making unit, generates a two-dimensional visualized defect image, and provides a simulation basis for adjusting the recognition threshold.

[0087] The intelligent decision-making unit is equipped with an extreme learning machine model with enhanced attention mechanism. After inputting the fused feature vector, it outputs the defect type and level of the magnesia-carbon brick, and the judgment result is transmitted to the data closed-loop management unit.

[0088] The data closed-loop management unit receives the judgment results, drives the production line control system to trigger the non-conforming product sorting mechanism, and at the same time transmits the test data back to the sample library to realize incremental updates of model parameters, forming a closed-loop collaboration between quality inspection and production.

[0089] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. An online quality inspection method for magnesia-carbon bricks based on impact testing, characterized in that, Includes the following steps: Step 1: Collect magnesia-carbon brick samples of different sizes, surface morphologies, and defects such as delamination, looseness, and pores. Use industrial CT scans to determine the location and size of defects, associate the samples with production process parameters, and establish an labeled sample library. Step 2: Obtain the real-time specifications of the magnesia-carbon brick to be tested through the visual recognition module, and automatically adjust the hammer parameters, striking speed and microphone array spacing of the striking mechanism. Step 3: A reciprocating pendulum striking mechanism driven by a stepper motor is used to strike the surface of the magnesium carbon brick at preset intervals, and multi-channel striking sound wave signals are collected synchronously through a microphone array. Step 4: Optimize the number of decomposition layers and penalty parameters of variational mode decomposition using a genetic algorithm, screen effective IMF components, and obtain the preprocessed acoustic signal through wavelet packet decomposition and threshold denoising; extract time-domain, frequency-domain and time-frequency-domain features from the preprocessed acoustic signal, and construct a fused feature vector after normalization. Step 5: Input the fused feature vectors into the attention mechanism-enhanced Extreme Learning Machine model and output the defect type and confidence level of the magnesium-carbon brick.

2. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 1, characterized in that, It also includes step 6, which involves constructing an acoustic response mapping model based on finite element simulation, converting the recognition results into a two-dimensional visual image, and dynamically adjusting the recognition threshold in conjunction with the real-time detection accuracy. Step 7: Feed the quality inspection results back to the production line control system to trigger the non-conforming product sorting mechanism, and at the same time send the test data back to the sample library to achieve iterative optimization of model parameters.

3. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 2, characterized in that, Step 1 includes the following sub-steps: Step 11: Select magnesia-carbon brick samples covering different production processes, batches, and specifications, and use industrial CT equipment to scan the samples to obtain the type, three-dimensional location coordinates, and geometric dimension information of their internal defects, which will be used as the defect annotation true values. Step 12: Perform a standardized tapping sound wave signal acquisition experiment on the sample, and record the sound wave signal, tapping position coordinates and environmental parameters at each tapping point. Step 13: Establish an associated database, with each record associated with the following information of a sample: production process parameters, defect information set annotated by industrial CT, acoustic signal dataset collected by impact experiment, and environmental parameters, forming an labeled sample library for model training and validation.

4. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 2, characterized in that, In step 2, the visual recognition module uses deep learning to acquire the geometric dimensions and surface roughness of the magnesia-carbon bricks in real time. ; The formula for calculating the microphone array spacing D is: ; in The longitudinal wave velocity in magnesia-carbon bricks. , For dynamic Young's modulus, Density; To set the center frequency for the tapping excitation, This is the layout coefficient; The adjustment of the striking speed v is based on: ; in H is the reference speed, and H is the thickness of the brick.

5. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 4, characterized in that, The motion control equations for the stepper motor-driven reciprocating pendulum striking mechanism are as follows: ； in For striking force, For the acoustic impedance of magnesia-carbon bricks, satisfy ; The instantaneous speed of the hammer head. Let be the acoustic impedance mismatch coefficient, and be the instantaneous velocity of the hammerhead. denoted as the material attenuation coefficient, and d as the defect depth.

6. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 5, characterized in that, In step 3, the strategy for fixed-point tapping is as follows: based on the rectangular outline of the magnesia-carbon brick, a grid-based planning method is used to determine the tapping point array, and the tapping point avoids the edge of the brick by a set distance; the microphone array includes a main microphone located directly above the tapping point and at least two auxiliary microphones located on the side of the brick. The main and auxiliary microphones are triggered synchronously with the data acquisition card to ensure the timing consistency of the multi-channel signals.

7. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 6, characterized in that, Step 4 includes the following sub-steps: Step 41, define the fitness function F of the genetic algorithm as a weighted sum of envelope entropy and correlation coefficient: ； in, This is the sum of the envelope entropies of all IMF components after VMD decomposition. This represents the average correlation coefficient between each IMF component and the original signal. β and β are weighting coefficients; Step 42: Perform VMD decomposition on the original signal using optimized parameters to obtain K IMF components; Step 43: Calculate the correlation coefficient between each IMF component and the original signal. Set threshold ,reserve The IMF component is the valid component; Step 44: Perform wavelet packet decomposition on the selected valid IMF components, and use an improved threshold function for denoising. for: ； ； in, These are wavelet coefficients. is the threshold, and a and b are adjustment parameters used to achieve a smooth transition of the threshold.

8. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 7, characterized in that, In step 4, the extracted features include time-domain features, frequency-domain features, and time-frequency-domain features; Time-domain characteristics include short-time energy, root mean square, waveform factor, and peak factor; Frequency domain features include spectral centroid, spectral variance, and power spectral entropy; Time-frequency domain features include the energy proportion of each subband in wavelet packet decomposition and wavelet packet entropy; After Z-score normalization of all extracted features, principal component analysis is used for dimensionality reduction and fusion to form the final fused feature vector.

9. The online quality inspection method for magnesia-carbon bricks based on impact testing according to claim 8, characterized in that, In step 6, the acoustic response mapping model is constructed based on multiphysics coupling simulation of magnesia-carbon bricks. The governing equations include thermal, mechanical, and acoustic coupling effects, and the governing equations are: ; in It is the gradient operator. Let C be the displacement field and C be the stiffness tensor. It is the stiffness tensor C and the displacement gradient tensor The double dot product, Where is the coefficient of thermal expansion, and T is the temperature field. For temperature gradient, This is the impact load.

10. An online quality inspection system for magnesia-carbon bricks, characterized in that, To implement the online quality inspection method for magnesia-carbon bricks based on impact detection as described in claim 9, the system includes a material parameter measurement module, an adaptive control unit, a signal acquisition and processing unit, a multiphysics simulation engine, an intelligent decision-making unit, and a data closed-loop management unit. The material parameter measurement module includes a visual recognition module, an adjustable striking mechanism, and a microphone array, used to acquire the density ρ and dynamic Young's modulus E of magnesia-carbon bricks online. n The parameters are achieved using a combination of a non-contact ultrasonic thickness gauge and a density sensor. The adaptive control unit is communicatively connected to the material parameter measurement module and automatically adjusts the detection parameters based on the material parameters. The signal acquisition and processing unit is communicatively connected to the material parameter measurement module to perform online signal denoising, VMD decomposition, and feature extraction. The multiphysics simulation engine runs an acoustic response mapping model; The intelligent decision-making unit is equipped with an extreme learning machine model with enhanced attention mechanism to identify and determine the type and level of defects; The data closed-loop management unit enables the return of detection data, incremental updates of the model, and coordinated control of the production line.

Images