A method and system for crack detection during a process of diamond sheet mining

Abstract

This invention relates to the field of diamond sheet mining technology, specifically to a method and system for crack detection during the diamond sheet mining process. The method includes: collecting vibration intensity, impact frequency, and processing load from the mining equipment; generating a state control factor based on a normalized baseline interval of crack-free conditions; adjusting the exposure time, acquisition frame rate, number of consecutive image frames, oblique incident light source, and imaging magnification according to the component changes of the state control factor to obtain a multi-scale response imaging sequence; determining a synchronous sampling segment based on the reciprocal of the impact frequency, sampling delay, and the state control factor; and extracting the vibration energy peak value, acoustic emission amplitude, and acoustic emission radio frequency band energy.

Description

Technical Field

[0001] This invention relates to the field of diamond sheet mining technology, and in particular to a method and system for detecting cracks in the diamond sheet mining process. Background Technology

[0002] During the mining, cutting, or separation of diamond sheet materials, they are typically subjected to the combined effects of impact loads from mining equipment, continuous vibration, and fluctuations in processing loads. Due to the high hardness and brittleness of diamond sheets, they are prone to developing microcracks, edge cracks, or crack propagation under unstable operating conditions. If these cracks are not identified in time during the mining process, they may lead to diamond sheet breakage, reduced mining quality, and affect the operational stability of the equipment.

[0003] Existing crack detection methods typically employ image recognition, vibration monitoring, or acoustic emission monitoring to determine crack conditions. Among these, image-based detection methods can reflect the morphology and location of cracks. However, when the vibration intensity, impact frequency, and processing load of mining equipment are constantly changing, fixed exposure times, acquisition frame rates, imaging magnification, or incident light angles are difficult to adapt to different working conditions, easily leading to blurred crack edges, unstable texture features, or difficulty in identifying micro-cracks.

[0004] While some existing technologies combine vibration or acoustic emission signals to aid in crack condition assessment, they typically focus on fusing results from single or multiple signals, lacking a mechanism to adjust acquisition parameters, sampling timing, and feature reliability based on the current mining conditions. When impact frequency, vibration intensity, or processing load changes, the response time, signal strength, and feature stability of different sensor signals may vary. If a fixed sampling window or fixed fusion weight is still used, inaccurate matching between image features, vibration features, and acoustic emission features can easily occur, thus affecting the assessment of crack location, size, and risk level.

[0005] Therefore, there is still a need in the existing technology for a crack detection method in the diamond sheet mining process that can adjust image acquisition parameters in conjunction with the dynamic changes in vibration intensity, impact frequency and processing load of mining equipment, determine the synchronous sampling segment of multi-source signals, and perform fusion identification based on the credibility of image, vibration and acoustic emission characteristics under the current working conditions, thereby improving the accuracy of crack detection and risk assessment under complex mining conditions. Summary of the Invention

[0006] The purpose of this invention is to address the lack of a mechanism in the prior art to adjust the acquisition parameters, sampling timing, and feature reliability in conjunction with the current mining status, and to propose a crack detection method for diamond sheet mining.

[0007] To achieve the objective, the present invention employs the following technical solution: a method for crack detection in the diamond sheet mining process, comprising: S1, collecting vibration intensity, impact frequency, and processing load of the mining equipment, and generating a state control factor after normalization based on a reference interval of crack-free working conditions; S2, adjusting the exposure time, acquisition frame rate, number of consecutive image frames, oblique incident light source, and imaging magnification according to the component changes of the state control factor to obtain a multi-scale response imaging sequence; S3, determining the synchronous sampling segment based on the reciprocal of the impact frequency, sampling delay, and state control factor, and extracting the vibration energy peak value, acoustic emission amplitude, and acoustic emission radio frequency band energy; S4 S1. Based on the changes in crack edge and vibration energy peak, and the changes in crack texture and acoustic emission amplitude, a structural difference is formed, and a confidence weight is determined to generate a fused feature vector. S5. The fused feature vector and multi-scale response imaging sequence are input into the crack identification model, causing the state control factor to generate state adjustment weights and jointly adjust each feature channel, outputting the crack location, size, and initial risk level. S6. Based on the changes in crack size, state control factor, and acoustic emission radio frequency band energy within a continuous acquisition cycle, a risk assessment value and a corrected risk level are determined, and the state control factor for the next acquisition cycle is corrected accordingly. This invention also provides a crack detection system for the diamond sheet mining process for executing the method.

[0008] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention first collects the vibration intensity, impact frequency, and processing load of the mining equipment, and generates a state control factor based on a benchmark interval without cracks, ensuring that subsequent detection parameters match the changes in the current mining conditions. Second, it adjusts the exposure time, acquisition frame rate, number of consecutive image frames, oblique incident light source, and imaging magnification according to different components of the state control factor. Specifically, shortening the exposure time and increasing the acquisition frame rate when the impact intensity increases helps reduce image blurring caused by vibration or impact; increasing the number of consecutive image frames when the impact frequency increases helps improve the temporal coverage of the crack change process; and adjusting the oblique incident light source and increasing the imaging magnification when the load coupling component increases helps enhance the recognizability of crack edges and texture details. Furthermore, it determines a synchronous sampling segment based on the reciprocal of the impact frequency, the sampling delay, and the state control factor, enabling feature extraction of images, vibration, and acoustic emission signals within similar operating conditions, thereby reducing... Feature mismatch caused by sampling time-series bias is addressed. Based on this, the structural differences between crack edge variation and vibration energy peak variation, and crack texture variation and acoustic emission amplitude variation are used to determine the credibility weights of image, vibration, and acoustic emission features. This allows the fusion process to reduce the interference of low-credibility features on the recognition results. Subsequently, the crack recognition model uses a state control factor to generate state adjustment weights, which, together with the credibility weights, adjust each feature channel, thereby improving the adaptability of crack location, size, and initial risk level judgments to dynamic mining conditions. Finally, the risk assessment value is determined based on the crack size variation, state control factor variation, and acoustic emission radio frequency band energy variation within a continuous acquisition cycle, and the state control factor for the next acquisition cycle is corrected accordingly. This forms a feedback correction mechanism between acquisition cycles, reducing the impact of operating condition fluctuations on continuous detection results and improving the reliability of crack recognition and the stability of risk assessment in complex mining processes. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] This invention provides a method for crack detection in the diamond sheet mining process, comprising the following steps: S1, collecting vibration intensity, impact frequency, and processing load of the mining equipment, and normalizing the values ​​of the samples collected under low, medium, and high load conditions when no cracks appear on the diamond sheet, generating a state control factor that includes impact intensity component, impact frequency component, and load coupling component; wherein, the state control factor includes impact intensity component, impact frequency component, and load coupling component; S2. Calculate the periodic average of the values ​​of the impact intensity component, impact frequency component, and load coupling component within two adjacent image acquisition cycles to obtain the intensity trend component, frequency trend component, and load trend component, respectively. When the intensity trend component increases, shorten the exposure time of the imaging system and increase the acquisition frame rate. When the frequency trend component increases, increase the number of consecutive image frames per unit acquisition cycle. When the load trend component increases, adjust the light source illumination angle to an inclined incident angle relative to the normal of the diamond sheet surface and increase the imaging magnification to create differences in crack edge shadows and enhance the distinguishability of crack boundaries and background textures. Determine the impact acquisition analysis segment based on the reciprocal of the impact frequency and the image acquisition frame rate, and divide the impact acquisition analysis segment into the impact initiation stage, impact peak stage, and impact decay stage according to the changing trend of the state control factor within the impact acquisition analysis segment. For adjacent impact acquisition analysis segments... The mean value of each component in the state control factor of the image acquisition cycle is calculated, and the mean values ​​of each component are weighted to obtain the trend control factor. The trend control factor is the mean value of the total state value of the state control factor within the corresponding image acquisition cycle. When the trend control factor increases within two consecutive image acquisition cycles, it is determined to be the impact initiation stage. When the trend control factor reaches its peak value within the impact acquisition analysis segment and decreases in the next image acquisition cycle, the image acquisition cycle corresponding to the peak value is determined to be the impact peak stage. When the trend control factor decreases within two consecutive image acquisition cycles, it is determined to be the impact decay stage. In the impact initiation stage, the acquisition frame rate is increased; in the impact peak stage, the exposure time is shortened; and in the impact decay stage, the illumination angle of the light source is adjusted to an inclined incident angle relative to the normal of the diamond sheet surface. A multi-scale response imaging sequence is generated for the same detection area in a stage sequence. S3. Determine the synchronous sampling segment based on the reciprocal of the impact frequency, the sampling delay, and the state control factor. Extract the vibration energy peak value, acoustic emission amplitude, and acoustic emission radio frequency band energy. Obtain the vibration energy peak value change and acoustic emission amplitude change based on the vibration energy peak value and acoustic emission amplitude in adjacent synchronous sampling periods. S4. The crack edge variation is obtained from the changes in crack boundary length, boundary gray-level gradient, and the number of discontinuous points within adjacent image acquisition cycles. The crack texture variation is obtained from the changes in gray-level distribution, texture direction, and local contrast in the crack area. The crack edge variation, crack texture variation, vibration energy peak value variation, and acoustic emission amplitude variation are all normalized based on the value ranges of samples acquired under low, medium, and high load conditions when the diamond sheet has no cracks. The difference between the normalized crack edge variation and the vibration energy peak value variation is calculated to obtain the first structural difference. The normalized crack texture variation is then compared with the acoustic emission amplitude variation. The difference is calculated to obtain the second structural difference quantity; the confidence weights of image, vibration, and acoustic emission features are determined based on the first and second structural difference quantities; the first and second reference ranges are determined by the normal fluctuation ranges of the corresponding structural difference quantities under low, medium, and high load conditions, respectively; when the first structural difference quantity falls within the first reference range, the vibration feature is used as a reliable feature consistent with the crack edge change, and the weight of the vibration feature in the fused feature vector is increased; when the second structural difference quantity falls within the second reference range, the acoustic emission feature is used as a reliable feature consistent with the crack texture change, and the weight of the acoustic emission feature in the fused feature vector is increased; when the crack edge change falls within the second reference range, the image, vibration, and acoustic emission features are ... edge texture change, and the weight of the acoustic emission feature in the fused feature vector is increased; when the crack edge change falls within the second reference range, the image, vibration, and acoustic emission features are used as a reliable feature consistent with the crack texture change, and the weight of the acoustic emission feature When the changes in crack edge and crack texture fall within the image change baseline range for at least three consecutive image acquisition cycles, the weight of image features in the fused feature vector is increased. Before generating the fused feature vector, based on the synchronous sampling cycle corresponding to the image acquisition cycle as the baseline, historical baseline intervals for vibration energy peak value, acoustic emission amplitude, crack edge change, and crack texture change are established using vibration intensity, impact frequency, and processing load combination samples collected under low, medium, and high load conditions when the diamond sheet has no cracks. When a feature deviates from its corresponding historical baseline interval, while the other three features remain within their respective historical baseline intervals, the weight of the feature deviating from the historical baseline interval in the fused feature vector is reduced, and the feature is retained for subsequent synchronous sampling periods for continuous trend judgment. When the feature continues to deviate from its corresponding historical baseline interval in subsequent synchronous sampling periods, its weight in the fused feature vector calculation is restored to suppress crack misjudgment caused by single-mode instantaneous impact interference. The fused feature vector is generated based on the crack edge change, crack texture change, vibration energy peak change, acoustic emission amplitude change, first structural difference, second structural difference, and the confidence weights of image, vibration, and acoustic emission features. S5. The fused feature vector and multi-scale response imaging sequence are input into the crack recognition model. The crack recognition model includes an image feature channel, a vibration feature channel, an acoustic emission feature channel, a state adjustment layer, and a recognition output layer. The state adjustment layer maps the state control factor to initial state adjustment weights acting on the image feature channel, vibration feature channel, and acoustic emission feature channel, respectively. Based on the signed change before normalization, the change direction of the crack edge is compared with the change of the peak vibration energy within the same acquisition period, and the change direction of the crack texture is compared with the change of the acoustic emission amplitude within the same acquisition period. When both the change of the crack edge and the change of the peak vibration energy increase or decrease, the initial state adjustment weights of the image feature channel and the vibration feature channel are increased and corrected. When both the change of the crack texture and the change of the acoustic emission amplitude increase or decrease, the initial state adjustment weights of the image feature channel and the acoustic emission feature channel are increased and corrected. The corrected state adjustment weights and the confidence weights are applied together to the corresponding feature channels, and the recognition output layer outputs the crack location, crack size, and initial risk level. S6. Determine the risk assessment value and correct the risk level based on the changes in crack size, state control factor, and acoustic emission radio frequency band energy within the continuous acquisition cycle, and provide feedback to correct the state control factor for the next acquisition cycle.

[0012] Specifically, this embodiment illustrates a crack detection method in the diamond sheet mining process, focusing on how the state control factor controls the imaging parameters, and how the changes in crack edge, crack texture, vibration energy peak value, and acoustic emission amplitude form the first and second structural differences.

[0013] The object of inspection is a diamond sheet during the mining process. The inspection equipment includes an imaging device, a light source adjustment device, a vibration sensor, an acoustic emission sensor, and a processor. The imaging device is positioned facing the surface of the diamond sheet to be inspected, used to acquire images of the diamond sheet surface; the light source adjustment device is used to adjust the illumination angle of the light source relative to the normal of the diamond sheet surface; the vibration sensor is mounted on the frame, clamping mechanism, or support structure near the diamond sheet of the mining equipment, used to acquire vibration signals from the mining equipment; the acoustic emission sensor is located on the support structure of the diamond sheet or near the diamond sheet, used to acquire acoustic emission signals generated by crack initiation or propagation.

[0014] The unit of vibration intensity is m / s². 2Or g, the impact frequency is in Hz, the processing load is expressed as a percentage of the equipment's rated load or the load sensor output value, the image acquisition period is in ms, the exposure time is in μs or ms, the acquisition frame rate is in fps, the light source illumination angle is in °, the imaging magnification is in times, the acoustic emission amplitude is expressed in mV or dB, when expressed in dB, the preset reference amplitude is used as the 0dB benchmark, the crack length and crack width are in mm, and the acoustic emission radio frequency band energy can be obtained by integrating the square of the acoustic emission signal amplitude within the preset frequency band, which is a dimensionless value after normalization. The normalized state control factor, structural difference quantity, confidence weight, state adjustment weight, and risk assessment value are all dimensionless values.

[0015] Within the same testing batch, the processor calculates the same type of data using the same unit and the same reference range. Vibration intensity is expressed in m / s². 2 The unit of measurement is either g or g; the acoustic emission amplitude is used in the calculation of the change in mV or dB, and when dB is used, the same reference amplitude is used for conversion; the crack length, crack width and crack boundary length are all measured in mm after pixel calibration when used for risk assessment, and the pixel value is only used for image preprocessing and candidate region tracking.

[0016] In this specification, "acoustic emission radio frequency band energy" refers to the energy of the acoustic emission signal within the preset frequency band in the synchronous sampling segment; "state adjustment weight" is the weight obtained by mapping the state control factor through the state adjustment layer and applied to each feature channel of the crack identification model; "confidence weight" is the fusion weight of image features, vibration features, and acoustic emission features determined based on the first structural difference and the second structural difference; "image acquisition cycle" is the cycle in which the imaging device completes one or a set of continuous image acquisitions; "synchronous sampling cycle" is the sampling cycle of vibration signals and acoustic emission signals aligned with the corresponding image acquisition cycle time; and "acquisition cycle" includes the corresponding image acquisition cycle and its synchronous sampling cycle unless otherwise specified.

[0017] To ensure consistency in terminology across different embodiments, "no cracks found" in this specification refers to a baseline state where no cracks were detected through manual review or high-precision offline detection; "crack edge variation" and "crack texture variation" correspond to image edge variation and texture variation, respectively, and the terms "crack edge variation" or "crack texture variation" with different meanings will no longer be used separately.

[0018] I. Establishment of Benchmark Samples and Benchmark Intervals Before formal testing, diamond sheets without cracks were selected as baseline samples, and sample data were collected under low, medium, and high load conditions. Low load conditions were 20%-40% of the equipment's rated load, medium load conditions were 40%-70% of the rated load, and high load conditions were 70%-90% of the rated load. Conditions below 20% or above 90% were considered abnormal conditions exceeding the baseline range and recorded separately or a new baseline range was established. At least three sets of samples were collected under each load condition, and each set of samples was collected continuously for at least 30 seconds.

[0019] Under each load condition, the processor collects vibration intensity, impact frequency, processing load, diamond sheet surface image, vibration signal, and acoustic emission signal. Based on the sample data without cracks, the processor establishes reference intervals for vibration intensity, impact frequency, processing load, peak vibration energy, acoustic emission amplitude, crack edge variation, and crack texture variation. In the absence of cracks, the crack edge variation and crack texture variation are calculated using candidate dark texture regions, background texture disturbance regions, or false detection candidate regions obtained after image preprocessing, to characterize the normal fluctuation range of image edges and textures in the crack-free state.

[0020] The baseline interval can be determined by using the minimum to maximum values ​​of the sample data under the corresponding operating conditions, or by adding or subtracting a preset multiple of the standard deviation from the sample mean. It is preferable to use the mean plus or minus two standard deviations to determine the baseline interval, in order to reduce the impact of occasional noise on the baseline interval.

[0021] The first reference range of the first structural difference and the second reference range of the second structural difference are determined by the normal fluctuation range of the first and second structural differences calculated from samples without cracks under low, medium, and high load conditions, respectively. Preferably, the processor calculates the mean and standard deviation of the structural differences under the corresponding conditions, and uses the mean plus or minus twice the standard deviation as the corresponding reference range.

[0022] The image change reference range is jointly constituted by the crack edge change reference range and the crack texture change reference range. When the crack edge change is within the crack edge change reference range and the crack texture change is within the crack texture change reference range, the processor determines that the image change within this image acquisition period falls within the image change reference range.

[0023] II. Generation of State Control Factors During the mining inspection process, the processor collects the current vibration intensity, impact frequency, and processing load according to the set image acquisition cycle. The image acquisition cycle can be set to 20-200ms, the vibration signal sampling frequency can be set to 5-20kHz, and the acoustic emission signal sampling frequency is not less than twice the upper limit of the acoustic emission filter frequency band, and can be set to 600kHz-1MHz.

[0024] The processor, based on whether the current processing load is low, medium, or high, calls the corresponding reference interval for that condition and normalizes the current vibration intensity, impact frequency, and processing load. The normalization process is as follows: if the current data is below the lower limit of the corresponding reference interval, the normalized value is 0; if the current data is above the upper limit of the corresponding reference interval, the normalized value is 1; if the current data is within the corresponding reference interval, it is converted to a value between 0 and 1 according to its position within the reference interval.

[0025] The normalized vibration intensity is used to generate the impact intensity component, and the normalized impact frequency is used to generate the impact frequency component. Let the normalized vibration intensity be I, the normalized impact frequency be F, and the normalized machining load be P, then the impact intensity component... =I, impact frequency component =F, load coupling component It is used to characterize the amplified effect of processing load on the impact state.

[0026] The processor's state control factor is formed by the impact intensity component C1, the impact frequency component C2, and the load coupling component C3. When the total state value is needed, use... Where a, b, and c are parameter weights, and a + b + c = 1; when there are no historical correction records, a = b = c = 1 / 3. When used for trend judgment and imaging parameter adjustment, the processor uses each component of the state control factor; when used for determining the synchronous sampling segment length, calculating the risk assessment value, and determining the corrected risk level, the processor uses the total state value Cs.

[0027] In the process of generating the state control factor, the load coupling component is used to reflect the amplification relationship between the combined effect of the processing load and the vibration intensity and impact frequency. In this embodiment, the load coupling component is determined by multiplying the normalized processing load by the average of the normalized vibration intensity and the normalized impact frequency. In the numerical embodiment, 0.60×(0.60+0.60) / 2 yields 0.36, which is consistent with the definition.

[0028] The change in the state control factor is used to characterize the degree of change in mining state between adjacent acquisition cycles. As a preferred method, the processor compares the total state value of the current acquisition cycle with the total state value of the previous acquisition cycle and takes the absolute value of the difference as the change in the state control factor; alternatively, the changes in the three state components can be calculated separately and weighted according to the current parameter weights.

[0029] III. Adjustment of Imaging Parameters by State Control Factors The processor calculates the periodic average of the impact intensity component, impact frequency component, and load coupling component within two adjacent image acquisition cycles, respectively, to obtain the intensity trend component, frequency trend component, and load trend component. If the trend component of the current cycle increases by more than a preset trend threshold compared to the trend component of the previous cycle, it is determined that the trend component has increased. The preset trend threshold can be set between 0.02 and 0.05.

[0030] When the intensity trend component increases, it indicates that the impact intensity of the mining equipment is increasing. The processor controls the imaging device to shorten the exposure time and increase the acquisition frame rate. The exposure time can be shortened from 1ms to 50-500μs, and the acquisition frame rate can be increased from 50fps to 100-500fps. Shortening the exposure time reduces image ghosting caused by mining vibration, and increasing the acquisition frame rate increases the number of effective images per unit time.

[0031] When the frequency trend component increases, it indicates an increase in the number of impacts per unit time. The processor then increases the number of consecutive image frames per acquisition cycle. This number can be increased from 3 frames to 5-10 frames per acquisition cycle. By increasing the number of consecutive image frames, the same detection area can be acquired multiple times within one impact cycle, thereby reducing the probability of missing the short-term crack development process.

[0032] The number of consecutive image frames within a unit acquisition cycle is matched with the acquisition frame rate and the image acquisition cycle. When a shorter image acquisition cycle is set and multiple consecutive image frames need to be acquired, the processor synchronously increases the acquisition frame rate, or extends a single unit acquisition cycle into a continuous acquisition window covering multiple adjacent exposure times, so that the number of consecutive image frames, the acquisition frame rate, and the acquisition cycle satisfy the time coverage relationship.

[0033] When the load trend component increases, it indicates that the stress state of the diamond sheet is enhanced. The processor controls the light source adjustment device to adjust the light source illumination angle to a tilted incident angle of 20°-60° relative to the normal of the diamond sheet surface, and increases the imaging magnification to 1.2-3 times the original magnification. The tilted incident light source can create shadow differences at the crack edge, and increasing the imaging magnification can enhance the distinguishability between the crack boundary and the background texture.

[0034] When two or three of the intensity trend component, frequency trend component, and load trend component increase simultaneously, the processor simultaneously executes the corresponding imaging parameter adjustment actions. If the imaging device or light source adjustment device has a hardware limit, the processor prioritizes shortening the exposure time and increasing the acquisition frame rate, then increases the number of consecutive image frames, and finally adjusts the light source illumination angle and imaging magnification.

[0035] After adjustment, the imaging device acquires multi-scale response imaging sequences of the same detection area under different exposure times, different acquisition frame rates, different consecutive image frames, different light source illumination angles, and different imaging magnifications. Here, multi-scale includes time scale, illumination scale, and imaging magnification scale.

[0036] IV. Image Preprocessing and Crack Candidate Region Determination The processor performs grayscale conversion, noise filtering, brightness equalization, crack edge enhancement, and crack texture enhancement on images in the multi-scale response imaging sequence. Noise filtering can use median filtering or Gaussian filtering; crack edge enhancement can use the Sobel operator, Canny operator, or Laplacian operator; crack texture enhancement can use local contrast enhancement, gray-level co-occurrence matrix feature enhancement, or directional filtering enhancement.

[0037] After image preprocessing, the processor determines crack candidate regions based on grayscale gradient, local contrast, and thin, elongated dark texture regions. Crack candidate regions can be obtained by threshold segmentation combined with connected component filtering; when no real crack is detected, crack candidate regions include candidate dark texture regions, background texture disturbance regions, or false detection candidate regions, where the aspect ratio, area, and grayscale gradient of the connected components are used to exclude isolated noise points and non-crack textures.

[0038] V. Obtaining the Change in Crack Edge During adjacent image acquisition cycles, the processor performs boundary tracking on the crack candidate region and extracts the crack boundary length, boundary gray-level gradient, and number of boundary discontinuity points.

[0039] The crack boundary length is the length of the boundary line segment of the crack candidate region after pixel calibration, and the unit can be mm or expressed in pixels in image coordinates. The boundary gray-level gradient is the average value of the gray-level change intensity on both sides of the crack boundary, and the unit can be gray level / pixel. The number of boundary discontinuity points is the number of interruption positions in the crack boundary where the continuity is lower than the preset judgment condition, and the unit is points.

[0040] The processor calculates the changes in crack boundary length, boundary grayscale gradient, and the number of discontinuous points between the current image acquisition cycle and the previous image acquisition cycle, and then weights these three factors to form the crack edge change. The weighting coefficients for these three factors can be set equally or pre-defined based on the sample training results. Each factor has a weight of one-third.

[0041] VI. Obtaining the Changes in Crack Texture The processor extracts grayscale distribution, texture orientation, and local contrast within the crack candidate region. Grayscale distribution represents the grayscale difference between the crack region and the background region; texture orientation represents the change in the main direction of the crack region relative to the background texture direction; and local contrast represents the local brightness difference within the crack region.

[0042] Gray-level distribution can be obtained from the gray-level histogram of the crack candidate region, texture direction can be obtained from the direction filter, gradient direction statistics or gray-level co-occurrence matrix, and local contrast can be calculated from local windows of 5×5 pixels, 7×7 pixels or 9×9 pixels.

[0043] The processor calculates the changes in grayscale distribution, texture direction, and local contrast between the current image acquisition cycle and the previous image acquisition cycle, and then weights these three factors to form the crack texture change. The weighting coefficients for these three factors can be set equally or pre-defined based on the sample training results. Each factor has a weight of one-third.

[0044] VII. Determination of Synchronous Sampling Segment and Extraction of Sensor Signal Features During the image acquisition cycle, the processor simultaneously acquires vibration and acoustic emission signals. The processor determines the impact period based on the current impact frequency, which is the reciprocal of the impact frequency. Using the image exposure center moment of the current image acquisition cycle as a reference moment, and combining it with the exposure delay of the imaging device, the sampling delay of the vibration sensor, and the sampling delay of the acoustic emission sensor, the processor determines the synchronous sampling segment for the vibration and acoustic emission signals.

[0045] The exposure delay of the imaging device, the sampling delay of the vibration sensor, and the sampling delay of the acoustic emission sensor can be obtained in advance through equipment calibration. During calibration, the processor sends synchronous trigger signals to the imaging device, vibration sensor, and acoustic emission sensor, and records the time difference between the actual response time and the trigger time of each device, using this time difference as the delay amount of the corresponding device.

[0046] Specifically, the synchronous sampling segment for vibration signals is centered at the image exposure center time plus the vibration sampling delay, and the synchronous sampling segment for acoustic emission signals is centered at the image exposure center time plus the acoustic emission sampling delay. The length of the synchronous sampling segment is determined based on the total state value of the impact period and the state control factor. When the total state value of the state control factor is low, the length of the synchronous sampling segment can be 0.5-1 times one impact period; when the total state value of the state control factor is high, the length of the synchronous sampling segment can be increased to 1-2 times one impact period.

[0047] In this embodiment, the sampling delay is used as a timestamp compensation amount. If the device calibration result indicates that the timestamp of a certain sensor is lagging behind the exposure center time of the image, the processor compensates for the lag on the sensor time axis according to the lag amount, so that the physical response time of the sensor sampling segment corresponds to the exposure center time of the image; if the device system has completed timestamp correction, the corrected unified timestamp is directly used to determine the synchronous sampling segment.

[0048] The processor performs bandpass filtering on the vibration signal, with the filtering frequency band set to 10Hz-2kHz. It then calculates the vibration energy curve within the synchronous sampling segment, which is formed by the integral of the squared amplitude or the root mean square value of the vibration signal within a sliding time window. The peak vibration energy is the maximum value of the vibration energy curve within the synchronous sampling segment. The processor compares the peak vibration energy of the current synchronous sampling period with the peak vibration energy of the previous synchronous sampling period to obtain the change in peak vibration energy.

[0049] The processor performs bandpass filtering on the acoustic emission signal, with the filtering frequency band set to 20-300kHz. The acoustic emission signal sampling frequency is no less than 600kHz. Within the synchronous sampling segment, it extracts the acoustic emission amplitude and acoustic emission radio frequency band energy. The acoustic emission radio frequency band energy is the integral of the square of the amplitude of the acoustic emission signal within the 20-300kHz frequency band within the synchronous sampling segment, which is normalized to a dimensionless value after normalization to the corresponding reference interval. The acoustic emission amplitude can be the peak amplitude, root mean square amplitude, or envelope peak amplitude of the acoustic emission signal within the synchronous sampling segment. When the acoustic emission amplitude is expressed in dB, the dB value is obtained by converting the current amplitude to a preset reference amplitude. The envelope peak amplitude is preferred. The processor compares the acoustic emission amplitude of the current synchronous sampling period with the acoustic emission amplitude of the previous synchronous sampling period to obtain the change in acoustic emission amplitude.

[0050] The change in acoustic emission radio frequency band energy is obtained by comparing the normalized acoustic emission radio frequency band energy of the current synchronous sampling period with the normalized acoustic emission radio frequency band energy of the previous synchronous sampling period. As a preferred method, the processor takes the absolute value of the difference between the two as the change in acoustic emission radio frequency band energy; when it is necessary to determine the crack propagation trend, the difference in band sign is also retained.

[0051] 8. Normalization Process The changes in crack edge, crack texture, peak vibration energy, and acoustic emission amplitude were all normalized based on the corresponding reference intervals established for diamond sheets without cracks under low, medium, and high load conditions.

[0052] If the current processing load is low, the processor calls the corresponding reference range under low load conditions; if the current processing load is medium, the processor calls the corresponding reference range under medium load conditions; if the current processing load is high, the processor calls the corresponding reference range under high load conditions.

[0053] After normalization, the changes in crack edge, crack texture, peak vibration energy, and acoustic emission amplitude are all converted to values ​​between 0 and 1. If the current data exceeds the upper limit of the corresponding reference interval, the normalized value is 1; if it is below the lower limit of the corresponding reference interval, the normalized value is 0.

[0054] IX. Formation of the First and Second Structural Differences The processor calculates the difference between the normalized crack edge variation and the normalized vibration energy peak variation to obtain a first structural difference. The first structural difference is preferably the absolute difference between the two. This first structural difference represents the degree of consistency between the crack edge variation and the vibration response variation. A smaller first structural difference indicates a more consistent relationship between the crack edge variation and the vibration energy peak variation.

[0055] The processor calculates the difference between the normalized crack texture variation and the normalized acoustic emission amplitude variation to obtain a second structural difference. The second structural difference is preferably the absolute difference between the two. This second structural difference represents the degree of consistency between the crack texture variation and the acoustic emission response variation. A smaller second structural difference indicates a more consistent relationship between the crack texture variation and the acoustic emission amplitude variation.

[0056] Based on sample data of crack-free diamond sheets under low, medium, and high load conditions, the processor establishes a first reference range for the first structural difference and a second reference range for the second structural difference. The first reference range is used to determine whether there is consistency between changes in crack edge and changes in vibration response; the second reference range is used to determine whether there is consistency between changes in crack texture and changes in acoustic emission response.

[0057] 10. Determining Credibility Weights and Generating Fusion Feature Vectors The processor sets initial confidence weights for image features, vibration features, and acoustic emission features. These initial confidence weights can be set equally, meaning they are the same.

[0058] When the first structural difference falls within the first reference range, the processor determines that the vibration feature is consistent with the crack edge change and increases the confidence weight of the vibration feature in the fused feature vector; when the first structural difference exceeds the first reference range, the processor decreases the confidence weight of the vibration feature in the fused feature vector.

[0059] When the second structural difference falls within the second reference range, the processor determines that the acoustic emission feature is consistent with the crack texture change and increases the confidence weight of the acoustic emission feature in the fused feature vector; when the second structural difference exceeds the second reference range, the processor decreases the confidence weight of the acoustic emission feature in the fused feature vector.

[0060] When the changes in crack edge and crack texture fall within the image change baseline range for at least three consecutive image acquisition cycles, the processor determines that the image acquisition quality and candidate crack texture representation are stable within the consecutive cycles, and increases the credibility weight of image features in the fused feature vector.

[0061] Each adjustment of the confidence weight can range from 10% to 30% of the original weight. When multiple weight adjustment conditions are met simultaneously, the processor completes the corresponding weight adjustments and then normalizes the confidence weights of image features, vibration features, and acoustic emission features so that the sum of the three confidence weights is 1.

[0062] Increasing or decreasing the confidence weight does not directly change the original image, vibration, or acoustic emission data, but only alters the input contribution of the corresponding modal features to the fused feature vector. The processor can set a lower and upper limit for each confidence weight, preferably a lower limit of 0.05-0.10 and an upper limit of 0.70-0.85; after each adjustment, normalization is performed so that the sum of the three confidence weights equals 1.

[0063] The processor fuses image features, vibration features, and acoustic emission features according to normalized confidence weights to generate a fused feature vector. The fused feature vector includes crack edge variation, crack texture variation, vibration energy peak variation, acoustic emission amplitude variation, acoustic emission radio frequency band energy, first structural difference, second structural difference, and corresponding confidence weights.

[0064] XI. Crack Identification Output The processor inputs the multi-scale response imaging sequence and the fused feature vector into the crack identification model. The crack identification model determines the location of the crack candidate region based on the multi-scale response imaging sequence, judges the consistency of the multimodal response of the crack candidate region based on the fused feature vector, and outputs the crack location, crack size, and initial risk level.

[0065] Crack identification models can employ convolutional neural networks, support vector machines, random forest models, or other classification or regression models capable of receiving image sequences and fused feature vectors to output crack location, crack size, and risk level. As one possible implementation, the multi-scale response imaging sequence is uniformly scaled to an M×N pixel image sequence and input into the image feature channel, while the fused feature vector is input into the feature channel. Training samples include samples without cracks, manually labeled crack samples, and historical detection samples collected under different load conditions. Labels include at least the crack candidate box location, crack length, crack width, and initial risk level. When using a convolutional neural network, the image feature channel outputs the candidate box center coordinates, candidate box width and height, and crack mask features. The feature channel also outputs multimodal response consistency features. Training losses include candidate box regression loss, crack length and width regression loss, and risk level classification cross-entropy loss. When using a support vector machine or random forest model, the image preprocessing module first outputs the crack candidate box location and size features, and then the model outputs the risk level based on the fused feature vector.

[0066] As a specific model training method, the image data, vibration data and acoustic emission data in the training samples are associated according to the acquisition cycle number and the synchronous sampling segment number. The image labels are determined by manually annotated crack candidate boxes, crack center lines and crack endpoints. The crack length and crack width are determined by the actual dimensions after calibration. The risk level label is determined by joint verification of crack size changes, acoustic emission radio frequency band energy changes and historical risk records.

[0067] Crack location can be represented using image coordinates or the actual coordinates of the calibrated diamond sheet surface; crack size includes crack length and crack width, in mm; the risk level output by the crack identification model is used as the initial risk level, and the risk assessment value R is used to generate a corrected risk level within a continuous acquisition cycle. As one possible implementation, a risk assessment value R less than 0.35 is corrected to low risk, R greater than or equal to 0.35 and less than 0.70 is corrected to medium risk, and R greater than or equal to 0.70 is corrected to high risk; the threshold can also be calibrated based on samples without cracks, manually labeled crack samples, and historical detection results.

[0068] In this embodiment, the state control factor is used not only for the adaptive adjustment of the imaging device's exposure time, acquisition frame rate, number of consecutive image frames, light source illumination angle, and imaging magnification, but also for the determination of synchronous sampling segments; the first structural difference quantity and the second structural difference quantity are used to establish the consistency relationship between image features, vibration features, and acoustic emission features, thereby improving the stability and anti-interference ability of crack detection during diamond sheet mining.

[0069] The state control factor, multi-scale response imaging sequence, first structural difference quantity, second structural difference quantity, and fused feature vector formed in this embodiment can be used as inputs for subsequent impact stage division, crack identification model state adjustment, and risk assessment feedback correction.

[0070] 12. Handling of Historical Reference Intervals and Single-Mode Anomalies Before generating the fused feature vector, the processor establishes historical reference intervals for vibration energy peak value, acoustic emission amplitude, crack edge variation, and crack texture variation based on samples collected from crack-free diamond sheets under low, medium, and high load conditions. Each historical reference interval includes a lower limit, an upper limit, and a load condition identifier. When only one of the four features deviates from its corresponding historical reference interval, while the other three features remain within their respective historical reference intervals, the processor reduces the confidence weight of the mode to which the anomalous feature belongs in the fused feature vector and stores the original value, normalized value, acquisition period, and deviation direction of the feature in a trend judgment cache. If the feature recovers to its corresponding historical reference interval in the next acquisition period, it is marked as transient interference; if the feature continuously deviates from its corresponding historical reference interval for two or more consecutive acquisition periods, its confidence weight in the fused feature vector calculation is restored to avoid long-term suppression of the true crack development signal.

[0071] XIII. Impact Phase Division and Phased Data Collection The processor determines the impact acquisition and analysis segment based on the reciprocal of the impact frequency and the image acquisition frame rate. Preferably, the processor uses the image exposure center time... Centered on the data, extending forward and backward by L / 2 time lengths respectively, the time range of the impact acquisition and analysis segment is... Where L is the length of the impact acquisition and analysis segment. Let the impact frequency be f, then the impact period T = 1 / f, and 0.5T ≤ L ≤ 2T. The processor calculates the periodic mean of each component in the state control factor of adjacent image acquisition cycles within the impact acquisition and analysis segment, and weights the mean of each component to obtain the trend control factor; the trend control factor can be the periodic mean of the total state value Cs of the state control factor within the corresponding image acquisition cycle. When the trend control factor increases for two consecutive image acquisition cycles, it is determined to be the impact initiation stage; when the trend control factor reaches its peak and decreases in the next image acquisition cycle, the period corresponding to the peak is determined to be the impact peak stage; when the trend control factor decreases for two consecutive image acquisition cycles, it is determined to be the impact attenuation stage. The processor increases the acquisition frame rate in the impact initiation stage, shortens the exposure time in the impact peak stage, adjusts the light source to an inclined incident angle relative to the normal of the diamond sheet surface in the impact attenuation stage, and generates a multi-scale response imaging sequence according to the stage sequence.

[0072] In real-time acquisition scenarios, the peak impact phase can be confirmed after the next image acquisition cycle is completed. That is, when the processor detects that the current trend control factor has reached the peak value within the peak acquisition analysis segment, and the trend control factor decreases in the next image acquisition cycle, the processor marks the image acquisition cycle corresponding to the previous peak value as the peak impact phase.

[0073] XIV. State Adjustment Weights and Crack Identification Model The crack recognition model includes image feature channels, vibration feature channels, acoustic emission feature channels, a state adjustment layer, and a recognition output layer. The state adjustment layer first maps the state control factor C=[C1,C2,C3] to image initial state adjustment weights, vibration initial state adjustment weights, and acoustic emission initial state adjustment weights, and then normalizes these three to a sum of 1. As one possible implementation, the state adjustment weights for each channel can be calculated using a normalized exponential function. Let the state adjustment weight of the i-th feature channel be... The linear mapping value of the i-th feature channel is ,but: Calculate, where i represents the image feature channel, vibration feature channel, or acoustic emission feature channel. The linear mapping value of the i-th feature channel. It can be represented as C1 is the impact strength component, C2 is the impact frequency component, and C3 is the load coupling component. , , and The parameters are obtained through model training or device calibration. The processor then uses the unnormalized signed change to determine the direction of change: the signed change is obtained by subtracting the previous period's feature value from the current period's feature value; when the signed changes of the crack edge change and the vibration energy peak change are both positive or both negative, the state adjustment weights of the image feature channel and the vibration feature channel are increased; when the signed changes of the crack texture change and the acoustic emission amplitude change are both positive or both negative, the state adjustment weights of the image feature channel and the acoustic emission feature channel are increased. The processor corrects the initial state adjustment weights based on the change direction comparison results to obtain the final state adjustment weights, and applies the final state adjustment weights and the confidence weights together to the corresponding feature channels; as a preferred method, the state adjustment weights of the corresponding modes are multiplied by the confidence weights and then normalized to obtain the channel input weights. The recognition output layer outputs the crack location, crack size, and risk level based on the weighted multimodal features.

[0074] The state adjustment weights and the confidence weights act on different targets. The state adjustment weights reflect the adjustment requirements of the current mining state for each feature channel, while the confidence weights reflect the consistency and reliability of image, vibration, and acoustic emission features under the current operating conditions. The processor applies both to the corresponding feature channel and normalizes the result to obtain the final input contribution for that channel.

[0075] XV. Risk Assessment Values ​​and Feedback Corrections The processor determines the risk assessment value based on the crack size propagation term, acoustic emission radio frequency band energy response term, and state disturbance term within the continuous acquisition cycle. The crack size propagation term D is determined by the normalized crack length variation. and normalized crack width variation get, The acoustic emission radio frequency band energy response term E is obtained from the normalized change in acoustic emission radio frequency band energy, or it can be obtained by combining the change in acoustic emission amplitude. The state disturbance term S is obtained from the normalized change in state control factor. The risk assessment value R = A × D + B × E + C × S, where A, B, and C are risk weights, and A + B + C = 1. By default, A, B, and C are all 1 / 3. When the change in crack length and crack width in the current acquisition cycle is greater than the corresponding change in the previous acquisition cycle, the processor increases A and B and decreases C. When the change in state control factor increases and the change in crack length and crack width in the current acquisition cycle is not greater than the corresponding change in the previous acquisition cycle, the processor decreases C to reduce the impact of equipment impact disturbance on crack risk assessment. The processor generates a corrected risk level based on the adjusted risk assessment value and corrects the parameter weights of vibration intensity, impact frequency, and processing load in the state control factor, so that the corrected state control factor participates in the imaging parameter adjustment, synchronous sampling segment determination, and crack identification model calculation in the next acquisition cycle.

[0076] The state disturbance term is obtained from the change in the state control factor, preferably using the normalized change in the state control factor. If the state control factor increases significantly in the current acquisition period, but the changes in crack length and crack width do not increase synchronously, the processor determines that the change mainly originates from equipment impact or processing load disturbance, and reduces the proportion of the state disturbance term in the risk assessment value; if the state control factor increases and the changes in crack size and acoustic emission radio frequency band energy increase synchronously, the proportion of the state disturbance term is retained or increased.

[0077] The risk weights are adjusted based on the changes over consecutive acquisition cycles. When the changes in crack length and crack width in the current acquisition cycle are both greater than the corresponding changes in the previous acquisition cycle, the processor increases the weights of the crack size propagation term and the acoustic emission radio frequency band energy response term; when only the change in the state control factor increases while the change in crack size does not increase, the processor decreases the weight of the state disturbance term. After each adjustment, the processor normalizes the risk weights and the state control factor parameter weights to ensure that the sum of the weights is 1.

[0078] The present invention also provides a crack detection system for diamond sheet mining process, including an imaging device, a light source adjustment device, a vibration sensor, an acoustic emission sensor, a processor, and a memory; the vibration sensor is used to collect vibration signals, the acoustic emission sensor is used to collect acoustic emission signals, the imaging device is used to collect images of the diamond sheet surface, and the light source adjustment device is used to change the irradiation angle. The memory is used to store benchmark samples, historical benchmark intervals, crack identification model parameters, and risk assessment records of continuous acquisition cycles when the diamond sheet has no cracks under low load, medium load, and high load conditions. The processor is used to normalize and weight the vibration intensity, impact frequency and processing load based on the value range of the sample collected when the diamond sheet has no cracks, and generate a state control factor including impact intensity component, impact frequency component and load coupling component. The processor then controls the exposure time, acquisition frame rate, number of consecutive image frames, light source illumination angle and imaging magnification according to the component changes of the state control factor. The processor is also used to determine the synchronous sampling segment, extract image, vibration and acoustic emission features, form a first structural difference quantity and a second structural difference quantity, generate a confidence weight based on the first structural difference quantity and the second structural difference quantity, and fuse image features, vibration features and acoustic emission features according to the confidence weight to form a fused feature vector. The processor is also used to generate state adjustment weights for the image feature channel, vibration feature channel and acoustic emission feature channel in the crack identification model based on the state control factor. The state adjustment weights are used to adjust the input contribution of the corresponding feature channel in the crack identification model, and to enable the fused feature vector and the state control factor to participate in the crack identification model calculation, and output the crack location, crack size and initial risk level. The crack size change includes the crack length change and the crack width change; when the risk level increases and at least one of the crack length change and crack width change increases, the processor increases the adjustment range of imaging parameters, the coverage range of synchronous sampling segment and the adjustment range of feature channel weight in the next acquisition cycle, and the increase is 10%-30% based on the current range.

[0079] Specifically, the imaging device is positioned facing the diamond sheet surface to be inspected, used to acquire images of the diamond sheet surface. The light source adjustment device is used to change the illumination angle of the light source relative to the normal to the diamond sheet surface, and can also adjust the light intensity. A vibration sensor is mounted on the mining equipment frame, clamping mechanism, or support structure near the diamond sheet, used to acquire vibration signals during the mining process. An acoustic emission sensor is positioned on the diamond sheet support structure, clamping mechanism, or near the diamond sheet, used to acquire acoustic emission signals generated during crack initiation or crack propagation. A processor is connected to the imaging device, light source adjustment device, vibration sensor, acoustic emission sensor, and memory, respectively, and is used to perform data acquisition control, feature extraction, weight calculation, model recognition, and feedback adjustment. The memory is used to store benchmark samples, historical benchmark intervals, crack recognition model parameters, and risk assessment records for continuous acquisition cycles.

[0080] The reference samples stored in the memory are data collected under low, medium, and high load conditions when the diamond sheet has not developed cracks. Low, medium, and high load conditions can be defined as a percentage of the equipment's rated load; for example, low load is 20%-40% of the rated load, medium load is 40%-70%, and high load is 70%-90%. The load range is a preferred embodiment and is not intended to limit the scope of protection of this invention. The reference samples include at least vibration intensity, impact frequency, processing load, peak vibration energy, acoustic emission amplitude, acoustic emission radio frequency band energy, crack edge variation, and crack texture variation.

[0081] Historical reference intervals are established based on reference samples taken when no cracks were observed. These intervals include historical reference intervals for vibration energy peak values, acoustic emission amplitude, acoustic emission radio frequency band energy, crack edge variation, and crack texture variation. Each historical reference interval is stored separately for low-load, medium-load, and high-load conditions, and includes an upper limit, lower limit, and corresponding condition identifier. Historical reference intervals are used to determine whether any abnormalities have occurred in the image, vibration, or acoustic emission characteristics within the current acquisition period.

[0082] The crack identification model parameters are stored in memory. The crack identification model includes an image feature channel, a vibration feature channel, an acoustic emission feature channel, a state adjustment layer, and an identification output layer. The image feature channel is used to process the diamond sheet surface image and extract crack edge features and crack texture features; the vibration feature channel is used to process the vibration signal and extract the vibration energy peak value and the change in vibration energy peak value; the acoustic emission feature channel is used to process the acoustic emission signal and extract the acoustic emission amplitude, the change in acoustic emission amplitude, and the acoustic emission radio frequency band energy; the state adjustment layer is used to generate state adjustment weights based on the state control factor; the identification output layer is used to output the crack location, crack size, and initial risk level.

[0083] Risk assessment records for continuous acquisition cycles are stored in memory. Each risk assessment record includes at least the acquisition cycle number, acquisition time, state control factor, crack location, crack length, crack width, crack length variation, crack width variation, acoustic emission radio frequency band energy variation, risk assessment value, initial risk level, revised risk level, confidence weight, state adjustment weight, and feature channel weight adjustment range. Risk assessment records are used to compare crack changes and risk changes between the current acquisition cycle and the previous acquisition cycle.

[0084] During the inspection process, the processor receives vibration signals from vibration sensors, acoustic emission signals from acoustic emission sensors, and images of the diamond sheet surface from imaging devices. The processor also acquires the current processing load. This current processing load can be obtained from the load percentage output by the mining equipment control system, from load sensors, or calculated from drive current, drive torque, or feed pressure.

[0085] The processor retrieves the corresponding low-load, medium-load, or high-load reference sample value range from memory based on the current processing load condition, and performs normalized weighted processing on the current vibration intensity, impact frequency, and processing load. After normalization, the processor generates a state control factor including an impact intensity component, an impact frequency component, and a load coupling component. The impact intensity component is obtained from the normalized vibration intensity, the impact frequency component is obtained from the normalized impact frequency, and the load coupling component is determined jointly by the normalized processing load, the normalized vibration intensity, and the normalized impact frequency.

[0086] The processor controls the imaging device and the light source adjustment device based on the changes in the state control factor components. When the impact intensity component increases, the processor controls the imaging device to shorten the exposure time and increase the acquisition frame rate; when the impact frequency component increases, the processor increases the number of consecutive image frames per acquisition cycle; when the load coupling component increases, the processor controls the light source adjustment device to change the light source illumination angle and increase the imaging magnification. Through this control, the imaging device obtains a multi-scale response imaging sequence.

[0087] The processor determines the synchronous sampling segment based on the reciprocal of the impact frequency, the image exposure center time, the sampling delay of the vibration sensor, and the sampling delay of the acoustic emission sensor. The synchronous sampling segment ensures that the image frame, vibration signal, and acoustic emission signal correspond in time. Within the synchronous sampling segment, the processor extracts the vibration energy peak value, acoustic emission amplitude, and acoustic emission radio frequency band energy, and obtains the changes in vibration energy peak value, acoustic emission amplitude, and acoustic emission radio frequency band energy based on the data from the current acquisition cycle and the previous acquisition cycle.

[0088] The processor extracts image features from the multi-scale response imaging sequence. These image features include crack edge variation and crack texture variation. Crack edge variation can be obtained from changes in crack boundary length, boundary gray-level gradient, and the number of discontinuous points on the boundary; crack texture variation can be obtained from changes in gray-level distribution within the crack region, texture direction, and local contrast.

[0089] The processor calculates the difference between the crack edge variation and the vibration energy peak variation to form a first structural difference; it also calculates the difference between the crack texture variation and the acoustic emission amplitude variation to form a second structural difference. The first structural difference represents the degree of consistency between image edge variation and vibration response variation, while the second structural difference represents the degree of consistency between image texture variation and acoustic emission response variation.

[0090] The processor generates confidence weights based on the first and second structural differences. These confidence weights include image feature confidence weights, vibration feature confidence weights, and acoustic emission feature confidence weights. When the first structural difference falls within a first reference range, the processor increases the vibration feature confidence weight; when the second structural difference falls within a second reference range, the processor increases the acoustic emission feature confidence weight; and when the crack edge variation and crack texture variation both fall within the image variation reference range for at least three consecutive image acquisition cycles, the processor increases the image feature confidence weight. The processor normalizes the image feature confidence weight, vibration feature confidence weight, and acoustic emission feature confidence weight to ensure their sum equals 1.

[0091] The processor fuses image features, vibration features, and acoustic emission features based on confidence weights to form a fused feature vector. The fused feature vector includes at least the crack edge variation, crack texture variation, vibration energy peak variation, acoustic emission amplitude variation, acoustic emission radio frequency band energy, first structural difference, second structural difference, and corresponding confidence weights.

[0092] The processor generates state adjustment weights for the image feature channel, vibration feature channel, and acoustic emission feature channel in the crack recognition model based on the state control factor. These state adjustment weights are used to adjust the input contribution of the corresponding feature channel in the crack recognition model. Preferably, the state adjustment layer inputs the state control factor into the weight mapping function to obtain the state adjustment weights for the image feature channel, vibration feature channel, and acoustic emission feature channel, and then normalizes these weights so that their sum is 1.

[0093] The feature channel weight adjustment magnitude represents the degree of adjustment to the input contribution of the corresponding feature channel. This adjustment magnitude can be the adjustment magnitude of the state adjustment weight, or it can be the adjustment magnitude of the final input contribution of the corresponding feature channel after the combined effect of the state adjustment weight and the confidence weight. Preferably, the processor multiplies the confidence weight and the state adjustment weight of the corresponding channel and normalizes the result to obtain the final input contribution of the corresponding feature channel.

[0094] The processor enables the fused feature vector and state control factor to jointly participate in the crack recognition model calculation. The crack recognition model obtains multimodal fusion features based on the fused feature vector, obtains state adjustment weights based on the state control factor, and adjusts the input contributions of the image feature channel, vibration feature channel, and acoustic emission feature channel through the state adjustment weights. The recognition output layer outputs the crack location, crack size, and initial risk level based on the adjusted feature channels.

[0095] The crack location can be represented using image coordinates or the actual coordinates of the calibrated diamond sheet surface. Crack dimensions include crack length and crack width, both in millimeters (mm). Crack size variations include changes in crack length and crack width. The crack length variation is obtained by comparing the crack length in the current acquisition cycle with that in the previous acquisition cycle, and the crack width variation is obtained by comparing the crack width in the current acquisition cycle with that in the previous acquisition cycle.

[0096] The processor determines an initial or revised risk level based on the risk assessment value. The risk assessment value can be obtained by weighting a crack size propagation term, an acoustic emission frequency band energy response term, and a state disturbance term. Specifically, the crack size propagation term is obtained from the changes in crack length and crack width, the acoustic emission frequency band energy response term is obtained from the changes in acoustic emission frequency band energy, and the state disturbance term is obtained from the changes in the state control factor. The processor compares the risk assessment value with a preset risk level threshold to obtain the initial or revised risk level. The preset risk level threshold can be determined based on samples without cracks, manually labeled crack samples, and historical detection results.

[0097] When the processor determines, based on the risk assessment records of consecutive acquisition cycles, that the risk level of the current acquisition cycle is higher than that of the previous acquisition cycle, and that at least one of the changes in crack length and crack width has increased, the processor increases the adjustment range of imaging parameters, the coverage area of ​​the synchronous sampling segment, and the adjustment range of feature channel weights in the next acquisition cycle. The increase range is 10%-30% based on the current range.

[0098] Specifically, increasing the adjustment range of imaging parameters in the next acquisition cycle includes at least one of the following: increasing the acquisition frame rate adjustment range, increasing the continuous image frame count adjustment range, increasing the imaging magnification adjustment range, increasing the light source illumination angle adjustment range, or increasing the exposure time shortening range. Increasing the coverage of the synchronous sampling segment refers to expanding the sampling time window of the vibration signal and acoustic emission signal relative to the image exposure center moment. Increasing the feature channel weight adjustment range refers to increasing the adjustment range of the state adjustment weight, or increasing the adjustment range of the final input contribution of the corresponding feature channel after the combined effect of the state adjustment weight and the confidence weight.

[0099] When the risk level does not increase during the current acquisition cycle, or when neither the change in crack length nor the change in crack width increases, the processor does not execute enhanced feedback actions, or maintains the adjustment range of imaging parameters, the coverage range of the synchronous sampling segment, and the adjustment range of feature channel weights for the current acquisition cycle. If the risk level decreases or remains stable over multiple consecutive acquisition cycles, the processor can gradually restore the adjustment range of imaging parameters, the coverage range of the synchronous sampling segment, and the adjustment range of feature channel weights to the baseline level.

[0100] Through this embodiment, the system can store benchmark samples, historical benchmark intervals, crack identification model parameters, and risk assessment records in the memory, and complete the generation of state control factors, adjustment of imaging parameters, determination of synchronous sampling segments, generation of structural difference quantity, generation of confidence weight, generation of fusion feature vector, generation of state adjustment weight, determination of risk level, and feedback adjustment through the processor, thereby realizing continuous detection of crack location, crack size, and risk level in the diamond sheet mining process.

[0101] In the complete numerical embodiment, the reference range under a certain load condition is a vibration intensity of 5-15 m / s². 2 Impact frequency 20-60Hz, processing load 40%-70%, current collected value is vibration intensity 11m / s² 2Given an impact frequency of 44Hz and a processing load of 58%, the values ​​are: I = (11-5) / (15-5) = 0.60, F = (44-20) / (60-20) = 0.60, P = (58-40) / (70-40) = 0.60, C1 = 0.60, C2 = 0.60, C3 = 0.60 × (0.60 + 0.60) / 2 = 0.36, and the default total state value is Cs = (0.60 + 0.60 + 0.36) / 3 = 0.52. If the crack length in the previous cycle was 1.20 mm and the width was 0.05 mm, and the crack length in the current cycle is 1.35 mm and the width is 0.07 mm, and the preset normalized upper limit for crack length variation is 0.30 mm and the normalized upper limit for crack width variation is 0.05 mm, then the normalized crack length variation ΔL = (1.35 - 1.20) / 0.30 = 0.50, the normalized crack width variation ΔW = (0.07 - 0.05) / 0.05 = 0.40, and the crack size expansion term D = (0.50 + 0.40) / 2 = 0.45. If the normalized variation of acoustic emission radio frequency band energy E = 0.55, the state disturbance term S = 0.30, and the risk weights are adjusted to A = 0.40, B = 0.40, and C = 0.20 due to the increase in both crack length and crack width variation,... The risk assessment value R = 0.40 × 0.45 + 0.40 × 0.55 + 0.20 × 0.30 = 0.46. Since 0.35 ≤ R < 0.70, the processor determines the corrected risk level of the continuous acquisition cycle as medium risk and feeds back this risk assessment value for the parameter weight correction of the state control factor in the next acquisition cycle.

[0102] In the numerical embodiment, the vibration intensity, impact frequency, and processing load are all within the medium load condition reference range, and the normalized result is 0.60. The load coupling component is determined by the average of the normalized processing load, normalized vibration intensity, and normalized impact frequency, and is therefore 0.36. The current periodic risk assessment value is 0.46, corresponding to a medium risk level, which is consistent with the preset risk level threshold.

[0103] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may use the disclosed technical content to make changes or modifications to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention shall still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for detecting cracks in the diamond sheet mining process, characterized in that, Includes the following steps: S1. Collect the vibration intensity, impact frequency and processing load of the mining equipment, and generate a state control factor containing impact intensity component, impact frequency component and load coupling component after normalization based on the benchmark interval of the crack-free working condition. S2. When the impact intensity component increases, shorten the exposure time and increase the acquisition frame rate; when the impact frequency component increases, increase the number of consecutive image frames; when the load coupling component increases, adjust the tilted incident light source and increase the imaging magnification to obtain a multi-scale response imaging sequence. S3. Determine the synchronous sampling segment based on the reciprocal of the impact frequency, the sampling delay, and the state control factor, and extract the vibration energy peak, acoustic emission amplitude, and acoustic emission radio frequency band energy. S4. The first structural difference is formed by the change in crack edge and the change in peak vibration energy, and the second structural difference is formed by the change in crack texture and the change in acoustic emission amplitude. Based on this, the credibility weights of image, vibration and acoustic emission features are determined, and a fused feature vector is generated. S5. Input the fused feature vector and multi-scale response imaging sequence into the crack identification model, so that the state control factor generates state adjustment weights, and together with the confidence weights, adjusts each feature channel to output the crack location, size and initial risk level. S6. Determine the risk assessment value and correct the risk level based on the changes in crack size, state control factor, and acoustic emission radio frequency band energy within the continuous acquisition cycle, and provide feedback to correct the state control factor for the next acquisition cycle.

2. The method for crack detection in the diamond sheet mining process according to claim 1, characterized in that, In S4, the crack edge variation is obtained from the changes in crack boundary length, boundary gray-level gradient, and the number of discontinuous points within adjacent image acquisition cycles; the crack texture variation is obtained from the changes in gray-level distribution, texture direction, and local contrast in the crack region; the crack edge variation, crack texture variation, vibration energy peak variation, and acoustic emission amplitude variation are all normalized based on the value range of samples acquired under low, medium, and high load conditions when the diamond sheet has no cracks; the difference between the normalized crack edge variation and the vibration energy peak variation is calculated to obtain the first structural difference; the difference between the normalized crack texture variation and the acoustic emission amplitude variation is calculated to obtain the second structural difference.

3. The method for crack detection in the diamond sheet mining process according to claim 2, characterized in that, In step S4, the confidence weights of image, vibration, and acoustic emission features are determined based on the first and second structural difference quantities. The first and second reference ranges are determined by the normal fluctuation ranges of the corresponding structural difference quantities under low, medium, and high load conditions, respectively. When the first structural difference quantity falls within the first reference range, the vibration feature is taken as a reliable feature consistent with the crack edge change, and the weight of the vibration feature in the fused feature vector is increased. When the second structural difference quantity falls within the second reference range, the acoustic emission feature is taken as a reliable feature consistent with the crack texture change, and the weight of the acoustic emission feature in the fused feature vector is increased. When the crack edge change quantity and the crack texture change quantity both fall within the image change reference range for at least three consecutive image acquisition cycles, the weight of the image feature in the fused feature vector is increased.

4. The method for crack detection in the diamond sheet mining process according to claim 2, characterized in that, In step S4, before generating the fused feature vector, based on the synchronous sampling period corresponding to the image acquisition period, historical reference intervals are established for vibration intensity, impact frequency, and processing load combination samples collected under low, medium, and high load conditions when the diamond sheet has no cracks. These intervals include peak vibration energy, acoustic emission amplitude, crack edge variation, and crack texture variation. When one of these features deviates from its corresponding historical reference interval in the current synchronous sampling period, while the other three features remain within their respective historical reference intervals, the weight of the feature deviating from the historical reference interval in the fused feature vector is reduced, and this feature is retained for subsequent synchronous sampling periods for continuous trend judgment. When this feature continues to deviate from its corresponding historical reference interval in subsequent synchronous sampling periods, its weight in the fused feature vector calculation is restored to suppress crack misjudgment caused by single-mode instantaneous impact interference.

5. The method for crack detection in the diamond sheet mining process according to claim 1, characterized in that, In S2, the state control factor includes an impact intensity component, an impact frequency component, and a load coupling component; the values ​​of the impact intensity component, the impact frequency component, and the load coupling component in two adjacent image acquisition cycles are calculated by periodic mean calculation to obtain the intensity trend component, the frequency trend component, and the load trend component, respectively. When the intensity trend component increases, shorten the exposure time of the imaging system and increase the acquisition frame rate; When the frequency trend component increases, the number of consecutive image frames per unit acquisition cycle is increased; when the load trend component increases, the illumination angle of the light source is adjusted to an oblique incident angle relative to the normal of the diamond sheet surface, and the imaging magnification is increased to form the difference in shadow at the crack edge and enhance the distinguishability of the crack boundary and background texture.

6. The method for crack detection in the diamond sheet mining process according to claim 1, characterized in that, In step S2, the impact acquisition analysis segment is determined based on the reciprocal of the impact frequency and the image acquisition frame rate. The impact initiation stage, impact peak stage, and impact decay stage are divided according to the changing trend of the state control factor within the impact acquisition analysis segment. The period mean of each component in the state control factor of adjacent image acquisition cycles within the impact acquisition analysis segment is calculated, and the trend control factor is obtained by weighting the mean of each component. The trend control factor is the period mean of the total state value of the state control factor within the corresponding image acquisition cycle. When the trend control factor increases within two consecutive image acquisition cycles, it is determined to be the initial stage of the impact. When the trend control factor reaches its peak during the impact acquisition analysis period and decreases in the next image acquisition cycle, the image acquisition cycle corresponding to the peak value is determined as the impact peak stage. When the trend control factor decreases within two consecutive image acquisition cycles, it is determined to be the impact attenuation stage. The acquisition frame rate is increased in the impact initiation stage, the exposure time is shortened in the impact peak stage, and the light source illumination angle is adjusted to an inclined incident angle relative to the normal of the diamond sheet surface in the impact attenuation stage. Multi-scale response imaging sequences are generated for the same detection area in a stage sequence.

7. The method for crack detection in the diamond sheet mining process according to claim 3, characterized in that, In step S5, the crack identification model includes an image feature channel, a vibration feature channel, an acoustic emission feature channel, a state adjustment layer, and an identification output layer. The state adjustment layer maps the state control factor to initial state adjustment weights that act on the image feature channel, vibration feature channel, and acoustic emission feature channel, respectively. Based on the signed change before normalization, the change direction of the crack edge is compared with the change of the vibration energy peak value within the same acquisition period, and the change direction of the crack texture is compared with the change of the acoustic emission amplitude within the same acquisition period. When both the change of the crack edge and the change of the vibration energy peak value increase or decrease, the initial state adjustment weights of the image feature channel and the vibration feature channel are increased and corrected. When both the change of the crack texture and the change of the acoustic emission amplitude increase or decrease, the initial state adjustment weights of the image feature channel and the acoustic emission feature channel are increased and corrected. The corrected state adjustment weights and the confidence weights are applied together to the corresponding feature channels, and the identification output layer outputs the crack location, crack size, and initial risk level.

8. The method for crack detection in the diamond sheet mining process according to claim 1, characterized in that, In step S6, the risk assessment value is determined by the crack size propagation term, the acoustic emission radio frequency band energy response term, and the state disturbance term within a continuous acquisition cycle. The crack size propagation term is obtained from the crack length change and crack width change, the acoustic emission radio frequency band energy response term is obtained from the acoustic emission radio frequency band energy change, and the state disturbance term is obtained from the state control factor change. When the crack length change and crack width change in the current acquisition cycle are both greater than the corresponding changes in the previous acquisition cycle, the proportion of the crack size propagation term and the acoustic emission radio frequency band energy response term in the risk assessment value is increased. When the state control factor change increases, and the crack length change and crack width change in the current acquisition cycle are not greater than the corresponding changes in the previous acquisition cycle, the proportion of the state disturbance term in the risk assessment value is decreased to reduce the impact of equipment impact disturbance on crack risk judgment. A corrected risk level is generated based on the risk assessment value, and the parameter weights of vibration intensity, impact frequency, and processing load in the state control factor are corrected. The corrected state control factor is then used in the imaging parameter adjustment, signal sampling segment determination, and crack identification model calculation for the next acquisition cycle.

9. A crack detection system for diamond sheet mining process, used to perform the crack detection method for diamond sheet mining process according to any one of claims 1-8, characterized in that, It includes an imaging device, a light source adjustment device, a vibration sensor, an acoustic emission sensor, a processor, and a memory; Vibration sensors are used to collect vibration signals, acoustic emission sensors are used to collect acoustic emission signals, imaging devices are used to collect images of the diamond sheet surface, and light source adjustment devices are used to change the illumination angle. The processor is used to generate state control factors based on vibration intensity, impact frequency, and processing load, and to control exposure time, acquisition frame rate, number of consecutive image frames, light source illumination angle, and imaging magnification based on the component changes of the state control factors. The processor is also used to determine synchronous sampling segments, extract image, vibration, and acoustic emission features, form a first structural difference quantity and a second structural difference quantity, generate a fused feature vector, and enable the fused feature vector and state control factors to jointly participate in the crack identification model calculation, outputting crack location, crack size, and initial risk level. The memory is used to store baseline samples, historical baseline intervals, model parameters, and risk assessment records.

10. A crack detection system for diamond sheet mining process according to claim 9, characterized in that, The memory is used to store benchmark samples, historical benchmark intervals, crack identification model parameters, and risk assessment records of continuous acquisition cycles collected under low, medium, and high load conditions when the diamond sheet has no cracks. The processor is also used to normalize and weight the vibration intensity, impact frequency, and processing load according to the value range of the samples collected when the diamond sheet has no cracks, generating a state control factor including impact intensity components, impact frequency components, and load coupling components. It generates confidence weights based on the first and second structural difference quantities, and fuses image features, vibration features, and acoustic emission features according to the confidence weights to form a fused feature vector. It generates state adjustment weights for the image feature channel, vibration feature channel, and acoustic emission feature channel in the crack identification model based on the state control factors. The state adjustment weights are used to adjust the input contribution of the corresponding feature channels in the crack identification model. The crack size change includes crack length change and crack width change. When the risk level increases and at least one of the changes in crack length and crack width increases, the adjustment range of imaging parameters, the coverage of synchronous sampling segments, and the adjustment range of feature channel weights in the next acquisition cycle are increased by 10%-30% based on the current range.

Images