A real-time monitoring method for aerospace and aviation part processing line

Abstract

The present application belongs to the technical field of image communication, and particularly relates to a real-time monitoring method for aerospace and aviation part processing lines, which comprises the following steps: collecting processing line images and performing gray scale processing; dividing each frame of gray scale image into non-overlapping blocks, determining the feature importance of the blocks according to the time sequence gray scale change; determining the distribution discrete degree of the blocks based on the pixel gray scale distribution; combining the feature importance and the distribution discrete degree to construct a quantitative priority index; introducing a dynamic threshold self-adaptive mechanism, dynamically adjusting the quantitative priority division threshold according to the processing state index and the system resource state index, and determining the quantitative number of each block; and finally, differentiating and compressing the blocks according to the quantitative number. The present application realizes the balance between high fidelity of key areas and overall high efficiency, and provides reliable real-time monitoring support for high-precision manufacturing process of aerospace and aviation parts.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image communication. More particularly, the present application relates to a real-time monitoring method for aerospace parts processing line. BACKGROUND

[0002] As the core product of high-end manufacturing industry, the machining precision and quality of aerospace parts are directly related to the safety and reliability of aircraft. In the machining process of key parts such as aircraft engine blades and turbine discs, real-time monitoring system plays a crucial role in timely detecting machining abnormalities such as tool wear, material delamination, and surface cracks. With the development of intelligent manufacturing technology, image monitoring methods based on vision have gradually become an important means of quality control in high-end manufacturing process due to their non-contact, rich information, and rapid response. However, the amount of raw image data is huge, and if the raw image is directly transmitted to the monitoring server or cloud for analysis, it will bring huge bandwidth pressure and storage burden, so efficient image compression technology is the key to realizing real-time monitoring of aerospace parts processing.

[0003] At present, the absolute moment double threshold compression (AMBTC) algorithm is widely used in industrial visual monitoring field due to its simple calculation and high compression efficiency. AMBTC algorithm compresses image blocks by calculating the high and low mean values within the blocks as two quantization values and assigning a binary label to each pixel. However, AMBTC algorithm uses the same two quantization values for all blocks, which cannot adapt to the information density difference of different regions. For complex texture processing areas (such as surface scratches, micro-cracks, and tool cutting marks), the lack of quantization levels leads to blurred or distorted key features in the reconstructed image, and serious loss of detail information. For smooth areas (such as uniform metal surfaces), two quantization values result in low compression efficiency and a large amount of redundant information. This "one-size-fits-all" compression strategy makes it difficult for AMBTC algorithm to meet the dual requirements of high fidelity in key areas and high efficiency overall, affecting the accuracy of real-time monitoring of aerospace parts processing. SUMMARY

[0004] To solve the technical problem that the AMBTC algorithm cannot meet the dual requirements of high fidelity in key areas and high efficiency overall, affecting the accuracy of real-time monitoring of aerospace parts processing, the present application provides a real-time monitoring method for aerospace parts processing line, comprising:

[0005] Images from an aerospace parts manufacturing line are acquired, and each frame is converted to grayscale. Each grayscale image is divided into several non-overlapping blocks, and the feature importance of each block is determined based on the grayscale changes of the same block in the grayscale images at different times. The distribution dispersion of each block is determined based on the grayscale distribution of pixels in each block. Based on the distribution dispersion and feature importance of each block, the quantization priority of each block is determined. Based on a dynamic threshold adaptive mechanism of system status and manufacturing process, the quantization quantity of each block is determined. The blocks are compressed according to the quantization quantity of each block to obtain compressed data. Real-time monitoring and anomaly warning are performed based on the compressed data. The dynamic threshold adaptive mechanism includes: dynamically adjusting the division threshold of quantization priority according to the current manufacturing status and the resource status of edge computing devices. When the manufacturing process is in a high-risk stage, the quantization quantity of the blocks is increased to ensure image quality, and when the edge computing device resources are scarce, the quantization quantity of the blocks is decreased to ensure system real-time performance.

[0006] Preferably, the feature importance satisfies the expression: ;in, Indicates the sequence number of the current frame. Indicates the first The first frame of the grayscale image The feature importance of each block Indicates the first The first frame of the grayscale image In the block, the th The grayscale value of each pixel; Indicates the first The first frame of the grayscale image In the block, the th The grayscale value of each pixel; Indicates the size of the block; Indicates the number of historical frames used for reference; For hyperparameters; Represents the hyperbolic tangent function; Represents the absolute value symbol.

[0007] Preferably, the degree of dispersion of the distribution satisfies the expression: ;in, Indicates the sequence number of the current frame. Indicates the first The first frame of the grayscale image The degree of dispersion of the distribution of each block; Indicates the first The first frame of the grayscale image In the block, the th The grayscale value of each pixel; Indicates the first The first frame of the grayscale image The minimum gray value in each block Indicates the first The first frame of the grayscale image The maximum grayscale value in each block; Indicates the size of the block; This represents the minimum value function.

[0008] Preferably, the quantization priority satisfies the expression: ;in, Indicates the first The first frame of the grayscale image Quantization priority of each block; Indicates the first The first frame of the image The degree of dispersion of the distribution of each block; Indicates the first The first frame of the grayscale image The feature importance of each block; Represents the natural exponential function; This is a hyperparameter used to adjust the sensitivity of quantization priority to feature importance.

[0009] Preferably, hyperparameters The method for obtaining it is: according to the first In a frame of grayscale image, the feature importance of all blocks is used to cluster all blocks into two classes. The mean of the feature importance of all blocks within a class is taken as the representative feature importance of that class, and the mean of the representative feature importance of the two classes is taken as... The value of .

[0010] Preferably, the step of dynamically adjusting the quantization priority division threshold based on the current processing status and the resource status of the edge computing device includes: determining the processing status index based on the change in the feature importance of each block in the current frame grayscale image relative to the feature importance of the corresponding block in the historical grayscale image; determining the system resource status index by combining the CPU utilization, memory occupancy, and bandwidth utilization of the edge computing device; and dynamically adjusting the quantization priority division threshold based on the processing status index and the system resource status index.

[0011] Preferably, the processing status index satisfies the expression: ;in, Indicates the sequence number of the current frame. Indicates the first Processing status indicators of grayscale images; Indicates the total number of blocks; Indicates the first The first frame of the grayscale image The feature importance of each block; and respectively represent the mean and standard deviation of the feature importance of the i-th patch in the past frames of the grayscale image, represents a preset reference number; represents a hyperbolic tangent function; represents an absolute value symbol.

[0012] Preferably, the method for obtaining the system resource state indicator comprises: obtaining the maximum value of the CPU usage rate, the memory occupancy rate and the bandwidth utilization rate of the edge computing device at the current time; subtracting the maximum value from 1 to obtain a result as the system resource state indicator.

[0013] Preferably, the method for dynamically adjusting the partition threshold of the quantization priority according to the processing state indicator and the system resource state indicator comprises: correcting the initial partition threshold according to the processing state indicator and the system resource state indicator to obtain an instant partition threshold of the current frame of the grayscale image; and performing weighted average on the instant partition threshold and a historical dynamic partition threshold to obtain a dynamic partition threshold of the current frame of the grayscale image; the instant partition threshold satisfies the expression: ; wherein, represents the serial number of the current frame, represents the i-th instant partition threshold of the i-th frame of the grayscale image, represents the i-th instant partition threshold of the i-th frame of the grayscale image, represents the i-th initial partition threshold, represents the i-th initial partition threshold, represents the i-th initial partition threshold, represents the processing state indicator corresponding to the i-th frame of the grayscale image, represents the system resource state indicator at the current time, represents a processing process normal state threshold, represents a system resource adequacy threshold, represents a processing state adjustment coefficient, represents a system resource adjustment coefficient.

[0014] ​​Preferably, the compression of the sub-blocks according to the quantization number of each sub-block to obtain compressed data comprises: when the quantization number of the sub-block is C=1, calculating the average gray value of all pixels in the sub-block as the only quantization value, and not generating a bitmap; when the quantization number of the sub-block is C=2, calculating the high and low average values as quantization values, and generating a 1-bit / pixel bitmap; when the quantization number of the sub-block is C=3 or C=4, using a K-means clustering algorithm to divide the pixels in the sub-block into C clusters, and calculating the average gray value of all pixels in each cluster as the quantization value, while generating a 2-bit / pixel bitmap; when the quantization number of the sub-block is C=5, using a K-means clustering algorithm to divide the pixels in the sub-block into 5 clusters, and calculating the average gray value of all pixels in each cluster as the quantization value, while generating a 3-bit / pixel bitmap; and the quantization values of all sub-blocks and the bitmaps are combined to form the compressed data.

[0015] The beneficial effects of the present application are that the present application realizes the intelligent strategy of "prior fidelity of key regions and efficient compression of non-key regions" through the dual evaluation mechanism of feature importance and distribution dispersion degree, and realizes high-precision and high-reliability real-time monitoring of aerospace part machining processes.

[0016] The present application can accurately capture key abnormal features such as tool cutting into the material interface region, surface cracks, and material delamination by analyzing the time sequence change of feature importance to identify machining state abnormalities.

[0017] The dynamic threshold adaptive mechanism of the present application can adjust the compression strategy in real time according to the CPU usage, memory occupancy and bandwidth utilization of the edge computing device. When the system resources are tight, the quantization number of the non-key region is automatically reduced to ensure stable operation of the monitoring system; when the resources are abundant, the quantization number of the key region is appropriately increased to enhance the monitoring accuracy.

[0018] The present application accurately identifies different stages of the machining process by analyzing the standardized deviation of feature importance relative to historical data, and dynamically adjusts the compression strategy. In the abnormal machining stage, the system automatically improves the image quality of the key region; in the stable machining stage, the system automatically improves the compression efficiency, realizes the intelligent balance of the image quality of the key region and the overall compression efficiency, and ensures the accuracy of the real-time monitoring of the aerospace part machining line. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart schematically showing a real-time monitoring method for an aerospace part machining line in the present application. DETAILED DESCRIPTION

[0020] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.

[0021] The specific implementation of the present application will be described in detail below with reference to the drawings.

[0022] The embodiments of the present application disclose a real-time monitoring method for aerospace part processing line, referring to Figure 1 , comprising steps S1 to S7:

[0023] S1, collecting images of the aerospace part processing line, and performing gray scale processing on each frame of the collected images.

[0024] Specifically, a high-frame-rate industrial camera is used to continuously shoot the cutting area of the key parts such as the aero-engine blade and turbine disc being processed in the numerical control machine tool, and a dynamic image stream containing the contact process of the tool and the workpiece is obtained.

[0025] It should be noted that the image acquisition frequency in the present embodiment is set to 120 frames / s, so as to sufficiently capture the transient phenomena such as slight vibration and surface deformation generated in the high-speed milling process. In other embodiments, the implementer can set the acquisition frequency according to the actual implementation.

[0026] Further, the subsequent calculation overhead is reduced, and the present application performs gray scale processing on each frame of the collected images.

[0027] It should be noted that the gray scale processing can not only reduce the data dimension and reduce the subsequent calculation overhead, but also enhance the robustness of the image to light fluctuation, which is conducive to highlighting the structure and texture changes in the images of the aerospace part processing line.

[0028] S2, dividing each frame of the gray scale image into a plurality of mutually non-overlapping blocks, and determining the feature importance of the blocks according to the gray scale changes of the same block in different time gray scale images.

[0029] Specifically, each frame of the gray scale image is divided into a plurality of mutually non-overlapping blocks. The size of the block is set to , , i.e., each block is pixels in size. In other embodiments, the implementer can set the block size according to the actual implementation.

[0030] It should be noted that, in aerospace part machining, the tool path is fixed, the background area is basically static, and the cutting area, the feed point and the like change greatly in gray scale over time, so the key monitoring area can be identified by pixel difference.

[0031] Specifically, the feature importance satisfies the expression:

[0032]

[0033] wherein, represents the serial number of the current frame, represents the serial number of the frame, the feature importance of the i-th block of the frame gray scale image, the gray scale value of the j-th pixel point in the i-th block of the frame gray scale image, the gray scale value of the j-th pixel point in the i-th block of the frame gray scale image, the size of the block, the gray scale value of the j-th pixel point in the i-th block of the frame gray scale image, the gray scale value of the j-th pixel point in the i-th block of the frame gray scale image, the size of the block, the number of historical frames for reference, in the embodiment, , in other embodiments, the implementer can set the size of according to the actual situation; is a hyperparameter, used to prevent the feature importance from being constantly close to 1 due to a larger , in the embodiment, is taken as , represents the average gray scale difference of the corresponding pixel points between the i-th frame gray scale image and the previous frame gray scale image. In other embodiments, the implementer can set the hyperparameter according to the actual implementation situation . represents the hyperbolic tangent function; represents the absolute value symbol. When the gray scale value of the pixel point in the i-th block of the i-th frame gray scale image is greater than the gray scale value of the pixel point in the corresponding block of the previous frame gray scale image, it means that the i-th block is more likely to contain important features that can reflect the change of the machining state, such as tool cutting, chip generation and surface morphology mutation.

[0034] When the gray scale value of the pixel point in the i-th block of the i-th frame gray scale image is greater than the gray scale value of the pixel point in the corresponding block of the previous frame gray scale image, it means that the i-th block is more likely to contain important features that can reflect the change of the machining state, such as tool cutting, chip generation and surface morphology mutation. ​​​​​​​​​​The greater the feature importance of a segment, the more attention will be paid to preserving the detailed information of that region during subsequent compression to ensure accurate monitoring of critical processing states. Conversely, when the feature importance of a segment is lower, the compression process will focus more on preserving its detailed information to ensure accurate monitoring of critical processing states. The first frame of the grayscale image The grayscale value of the pixel in each block is compared with the previous one. The smaller the difference in grayscale values ​​of pixels in corresponding blocks within a frame of a grayscale image, the more abbreviated the grayscale value. The first segment is more likely to contain static information such as the background area and the fixing fixture. The lower the feature importance of each block, the lower the compression quality of that region can be appropriately reduced in subsequent compression processes to improve overall compression efficiency.

[0035] It should be noted that this invention uses the hyperbolic tangent function. Limit feature importance to Within this range, the consistency and comparability of feature importance assessment are ensured under different lighting conditions and image contrasts, effectively adapting to the complex and varied lighting environment of aerospace parts processing sites. In other embodiments, implementers can choose other suitable nonlinear mapping functions to replace the hyperbolic tangent function according to the actual implementation situation. .

[0036] S3. Determine the degree of dispersion of each block based on the grayscale distribution of pixels in each block.

[0037] It should be noted that the AMBTC compression algorithm compresses image blocks by calculating the high and low mean values ​​within each block as two quantization values ​​and assigning a binary label to each pixel. However, using only two quantization values ​​for all blocks leads to significant loss of detail in areas with complex textures (such as finely machined textures on aerospace parts or tool marks) due to insufficient quantization, resulting in blurred or distorted key features in the reconstructed image. For smooth areas (such as uniform metal surfaces), the compression efficiency is low, and redundant information exists. Therefore, this invention analyzes the grayscale distribution characteristics of pixels within each block to determine the degree of dispersion in the block distribution. This allows for differentiated quantization of different blocks based on the degree of dispersion, achieving an adaptive strategy of "refined representation of complex areas and efficient compression of simple areas."

[0038] It should be further explained that in images of aerospace parts manufacturing lines, the larger and more uniformly distributed the area (such as a processing area with rich surface texture), the more quantization values ​​are needed to reduce information loss and ensure accurate capture of minute surface defects or processing anomalies. On the other hand, the smaller the area and the closer the distribution is to the ends of the grayscale range (such as a uniform surface area), the fewer quantization values ​​can be used to quantize it, thereby improving compression efficiency and reducing data transmission and storage burden.

[0039] Specifically, the degree of dispersion of the distribution satisfies the expression:

[0040]

[0041] in, Indicates the sequence number of the current frame. Indicates the first The first frame of the grayscale image The degree of dispersion of the distribution of each block; Indicates the first The first frame of the grayscale image In the block, the th The grayscale value of each pixel; Indicates the first The first frame of the grayscale image The minimum gray value in each block Indicates the first The first frame of the grayscale image The maximum grayscale value in each block; Indicates the size of the block; This represents the minimum value function.

[0042] When the The first frame of the grayscale image When the grayscale values ​​of pixels in a block are distributed over a large and relatively uniform range within the grayscale interval, it indicates that the block contains rich texture information. The degree of dispersion of the distribution of each block If the value is relatively large, more quantization is needed in subsequent compression processes to retain detailed information. Conversely, when the value is relatively small... The first frame of the grayscale image If the gray values ​​of pixels in a block are concentrated at both ends of the gray value range or are generally similar, it indicates that the information in that block is relatively simple. In this case, the first... The degree of dispersion of the distribution of each block The smaller the quantization, the less quantization can be used in subsequent compression processes to improve compression efficiency.

[0043] It should be noted that the distribution dispersion index used in this invention can effectively reflect the "inward distribution" characteristics of pixel values ​​within the grayscale range. Compared with traditional variance or entropy indices, it can more accurately guide the quantization strategy of aerospace parts processing line images. For example, a certain block may contain a large number of pixels close to 0 and 255 (such as bright spots and dark areas coexisting). , Then the majority of pixels within the block One term is close to 0, resulting in a low degree of dispersion in the block distribution, close to 0; while the other block's gray level is concentrated between 100 and 170, and is uniformly distributed within this range. , , so that the distribution of the block is more discrete. Then the previous block is more suitable for quantization with two representative values (high and low mean value), and the latter block needs more quantization numbers to retain details.

[0044] S4, determine the quantization priority of each block based on the distribution discrete degree and feature importance of each block.

[0045] It should be noted that the key area reflects the contact state of the tool and the workpiece, the surface quality change and the potential defect and other core processing information, so for the key area, even if the gray distribution is simple, more quantization values are needed to ensure accurate monitoring of key processing state and timely capture of abnormal conditions. The non-key area (such as background) has low monitoring value for processing quality and is not the monitoring focus, so for the non-key area, even if the gray distribution is complex (such as the texture in the background), fewer quantization values can be used to improve the compression efficiency. Therefore, the present application combines the feature importance with the distribution discrete degree to construct a quantization priority index to realize the intelligent strategy of "key area priority fidelity, non-key area efficient compression".

[0046] Specifically, the quantization priority satisfies the expression:

[0047]

[0048] wherein, represents the quantization priority of the i-th block of the j-th frame gray image; represents the distribution discrete degree of the i-th block of the j-th frame image; represents the feature importance of the i-th block of the j-th frame gray image; represents the natural exponential function; is a hyperparameter, used to adjust the sensitivity of the quantization priority to the feature importance. In this embodiment, the acquisition method of the hyperparameter is: according to the feature importance of all blocks in the j-th frame gray image, all blocks are clustered into two categories, the mean value of the feature importance of all blocks contained in the category is taken as the representative feature importance of the category, and the mean value of the representative feature importance of the two categories is taken as

[0049] ​​​​​​​The value of . It should be noted that the clustering algorithm in this embodiment is K-means clustering. In other embodiments, implementers can choose the clustering algorithm according to the actual implementation situation. Since the tool path is fixed and the background area is basically static in the machining of aerospace parts, while the cutting area, the entry point and other positions will undergo drastic grayscale changes over time, in the two categories obtained by clustering, the category representing the background, fixed fixture and other static areas with low feature importance is the background, fixed fixture and other static areas, and the category representing the tool contact area, cutting area and other dynamic areas with significant changes is the category representing the tool contact area, cutting area and other dynamic areas with high feature importance.

[0050] It should be noted that when the first The first frame of the grayscale image Feature importance of each block Greater than hyperparameters And the larger the number, the more important it is to increase the number of... Each segment is more likely to be a critical monitoring area such as the tool contact area. Greater than 0, making For a decimal between (0,1), according to the properties of the power function, even if the first... The degree of dispersion of the distribution of each block Smaller, quantization priority The value will still be inflated, allowing more quantization values ​​to be allocated to the blocks corresponding to key monitoring areas; conversely, when the feature importance is low... Less than hyperparameters And the lower the value, the more important it is to improve the quality of the first... Individual blocks are more likely to be non-critical areas such as the background or fixed fixtures. Less than 0, making If the value is greater than 1, according to the properties of the power function, even if the distribution is highly discrete... Larger, quantification priority It will also be moderately suppressed, allowing for the allocation of fewer quantization values ​​to blocks corresponding to non-critical regions to improve compression efficiency. When the... The first frame of the grayscale image Feature importance of each block equal to hyperparameters At that time, the first Each segment is more likely to be a transitional area between the key region and the background, at which point... Equal to 0, making The value is 1, and the priority is quantized based on the properties of the power function. Depends on the first The degree of dispersion of the distribution of each block Size.

[0051] S5. Based on the dynamic threshold adaptive mechanism of system state and processing, determine the quantization quantity of each block.

[0052] It should be noted that, for The original data amount is bits. When quantization values are adopted, the compressed data amount is bits, wherein is the number of bits required by the quantization values, and is the number of bits required by the bitmap. When the compressed data amount is less than the original data amount, the compression is effective. When the block size is 8x8 pixel size, it is calculated that the compression is effective when . However, considering the characteristics of aerospace and aviation part machining images and the real-time requirements of edge computing devices, the present application limits to the range of {1, 2, 3, 4, 5}, that is, a maximum of 5 quantization values are used. Therefore, 4 threshold values need to be set to divide the quantization priority into 5 intervals, corresponding to 5 quantization numbers.

[0053] In one embodiment, fixed division threshold values , , , are adopted to determine the quantization number of the block according to the quantization priority of the block. Specifically:

[0054] When the quantization priority of the block is less than , the quantization number of the block is 1; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 2; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 3; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 4; and when the quantization priority of the block is greater than or equal to , the quantization number of the block is 5.

[0055] Wherein, the empirical values of the division threshold values , , , are 0.15, 0.35, 0.60, and 0.80, respectively. In other embodiments, the implementer can adjust the division threshold value settings according to the actual machining material and process characteristics, for example, for the finishing stage with higher surface quality requirements, the threshold values can be appropriately reduced to increase the overall quantization number.

[0056] It should be noted that the abnormal state (such as tool wear, material layering, etc.) in the aerospace part processing process often shows the mutation characteristics of a specific area in the image, and these characteristics are early signs of quality deterioration. The traditional fixed threshold compression method cannot distinguish between abnormal characteristics and normal characteristics, and is prone to over-compression of key abnormal information and loss. At the same time, the resource state (such as CPU usage, memory occupancy, network bandwidth) of the edge computing device will change dynamically over time. The traditional fixed threshold mapping method cannot adapt to this dynamic change, often leading to insufficient key stage information or system crashes when resources are tight. Therefore, in another embodiment, the present application introduces a dynamic threshold adaptive mechanism, which dynamically adjusts the division threshold of the quantization priority by monitoring the processing state and system resource state in real time, and realizes the intelligent compression strategy of "high-fidelity in abnormal stage, running in resource shortage".

[0057] In another embodiment, an initial division threshold is set; a system resource state indicator of the edge computing device at the current time is determined according to the CPU usage, memory occupancy, and bandwidth utilization of the edge computing device at the current time; a processing state indicator of the current frame grayscale image is determined according to the change of the feature importance of each block in the current frame grayscale image relative to the feature importance of the corresponding block in the historical grayscale image; the initial division threshold is corrected according to the processing state indicator and the system resource state indicator at the current time to obtain the instant division threshold of the current frame grayscale image; the instant division threshold and the historical dynamic division threshold are weighted and averaged to obtain the dynamic division threshold of the current frame grayscale image; and the dynamic division threshold is used to determine the quantization quantity of each block in the current frame grayscale image according to the quantization priority of each block in the current frame grayscale image.

[0058] Specifically, the initial division threshold is set, including:

[0059] In this embodiment, the initial division threshold , , , The empirical values are 0.15, 0.35, 0.60, and 0.80, respectively. In other embodiments, the implementer can set them according to the actual implementation.

[0060] Further, the system resource state indicator satisfies the expression:

[0061]

[0062] Wherein, represents the system resource state indicator at the current time; represents the CPU usage of the edge computing device at the current time; represents the memory occupancy of the edge computing device at the current time; This indicates the bandwidth utilization rate of the edge computing device at the current moment. This represents the maximum value function. Since system resource status is mainly constrained by the most strained resources, this embodiment takes the maximum value of CPU, memory, and bandwidth utilization, and subtracts this maximum value from 1 to obtain the resource remaining amount. When any resource is close to saturation (e.g., CPU utilization reaches 90%), the overall system resource status becomes strained, potentially leading to data processing delays or loss. In this case, the smaller the system resource status index, the better. Conversely, when all resources are abundant (e.g., CPU utilization is only 20%, memory utilization is only 30%, and bandwidth utilization is only 25%), the overall system resource status is more relaxed, capable of handling more computing tasks, and the larger the system resource status index, the better. This embodiment, by constructing a system resource status index, can effectively reflect the overall resource status of edge devices, facilitating rapid decision-making.

[0063] Furthermore, the processing status index satisfies the expression:

[0064]

[0065] in, Indicates the sequence number of the current frame. Indicates the first Processing status indicators of grayscale images; Indicates the total number of blocks; Indicates the first The first frame of the grayscale image The feature importance of each block; and They represent the first In the past, each block The mean and standard deviation of feature importance in a frame of grayscale image. This indicates a preset reference quantity. In this embodiment, In other embodiments, implementers may set the appropriate parameters according to the actual implementation situation. ; This represents the hyperbolic tangent function, used to limit the processing status index to the range [0,1]. Represents the absolute value symbol.

[0066] It should be noted that this embodiment identifies abnormal changes in machining status by analyzing the standardized deviation of feature importance relative to historical data. The machining process of aerospace parts typically exhibits cyclical characteristics (such as turning of rotating workpieces and milling with periodic feeds), and feature importance should show regular changes. When feature importance deviates from historical patterns, it is more likely to indicate machining anomalies or critical stages. For example, in the milling of aero-engine blades, when the tool cuts into the interface between titanium alloy and composite materials, the feature importance of the cutting zone will be significantly higher than the historical average, leading to… The value increases, and the system automatically identifies that it is in a high-risk stage.

[0067] Further, the instant division threshold satisfies the expression:

[0068]

[0069] wherein, represents the serial number of the current frame, represents the frame gray image, the instant division threshold, ; represents the initial division threshold; represents the machining state index corresponding to the frame gray image, represents the system resource state index at the current time, represents the normal state threshold of the machining process, represents the system resource sufficiency threshold, represents the machining state adjustment coefficient, represents the system resource adjustment coefficient.

[0070] In the embodiment, the empirical value of the normal state threshold of the machining process is 0.3, the empirical value of the system resource sufficiency threshold is 0.5, the empirical value of the machining state adjustment coefficient is 0.4, and the empirical value of the system resource adjustment coefficient is 0.2. In other embodiments, the implementer can set them according to the actual implementation.

[0071] When the aerospace part machining process is in a high-risk stage, the machining process state index increases, and the instant division threshold is reduced, so that more blocks obtain a higher quantization number (such as from C=3 to C=4), thereby improving the image quality of the region where the machining abnormality may exist; when the edge computing device is in a resource shortage state (such as high CPU usage, high memory occupancy, and high bandwidth utilization), the system resource state index decreases, and the instant division threshold is increased, at which time the quantization number of the non-critical region can be appropriately reduced (such as from C=2 to C=1), so as to ensure the real-time performance of the system; conversely, when the aerospace part machining process is in a low-risk stage, the machining process state index decreases, and the instant division threshold Increasing the number of quantizations reduces the number of quantizations (e.g., from C=2 to C=1), thereby improving overall compression efficiency; when edge computing devices have sufficient resources (e.g., low CPU utilization, low memory usage, low bandwidth utilization), system resource status indicators... Increase, and At that time, the threshold is divided in real time. If the value is reduced, the quantization level of the key region can be appropriately increased (e.g., from C=3 to C=4) to further improve the image quality of the key region.

[0072] Furthermore, the dynamic partitioning threshold satisfies the expression:

[0073]

[0074] in, Indicates the sequence number of the current frame. Indicates the first The first frame of the grayscale image A dynamic partitioning threshold, ; Indicates the first The first frame of the grayscale image One dynamic partitioning threshold; Indicates the first The first frame of the grayscale image One instantaneous segmentation threshold; This represents the smoothing coefficient, used to control the smoothness of threshold changes. In this embodiment... In other embodiments, the implementer can set the threshold according to the actual implementation situation. It should be noted that the instantaneous threshold... It reflects the immediate state requirements of the current frame, but the threshold may jump due to sudden changes in the state indicators. Therefore, this embodiment introduces a threshold smooth transition mechanism to ensure gradual changes in compression quality and avoid sudden changes in image quality.

[0075] It should be noted that, in order to prevent excessive threshold adjustment and maintain the basic hierarchical logic, this embodiment applies upper and lower limit constraints to the dynamic division threshold to ensure that the dynamic division threshold is within a reasonable range.

[0076] Specifically, in response to ,Keep Unchanged; Responding to ,Will The value is adjusted to ; in response ,Will The value is adjusted to ,in, Indicates the first An initial partitioning threshold is set. It should be noted that, since the quantization priority ranges from [0,1], when... greater than , indicates that the quantization number of the block is at most , since the quantization priority cannot be greater than 1, the quantization priority of all blocks is less than the threshold , the corresponding quantization number will not exceed . For example, when is adjusted to , the quantization number of all blocks is at most 4, that is, the case of C=5 will not occur.

[0077] Further, the dynamic division threshold is adopted to determine the quantization number of each block in the current frame gray image according to the quantization priority of each block in the current frame gray image, comprising:

[0078] When the quantization priority of the block is less than , the quantization number of the block is 1; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 2; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 3; when the quantization priority of the block is greater than or equal to and less than , the quantization number of the block is 4; when the quantization priority of the block is greater than or equal to , the quantization number of the block is 5. Wherein, indicates the serial number of the current frame, , , , indicates the dynamic division threshold of the current frame gray image.

[0079] It should be noted that the dynamic threshold adaptive mechanism of the present application can more accurately capture abnormal states by analyzing the time sequence change of the feature importance, and can improve the adaptability of the system under complex working conditions, thereby providing a more intelligent and reliable image compression solution for real-time monitoring of aerospace part processing.

[0080] S6, compressing the blocks according to the quantization number of each block.

[0081] Specifically, for any one block, the quantization process is:

[0082] When the quantization number of the block is C=1, calculate the gray mean value of all pixels of the block, take the gray mean value as the only quantization value, do not generate a bitmap, and the data amount after compression is 8 bits;

[0083] When the number of quantization of the sub-block is C=2, the high and low mean values q1 and q2 are calculated as the quantization values, and a 1-bit / pixel bitmap is generated, and the data amount after compression is bits.

[0084] When the number of quantization of the sub-block is C=3 or C=4, the K-means clustering algorithm is used to divide the pixels in the sub-block into C clusters, the mean value of the gray scale of all the pixels in each cluster is calculated as the quantization value, and a 2-bit / pixel bitmap is generated, and the data amount after compression is bits.

[0085] When the number of quantization of the sub-block is C=5, the K-means clustering algorithm is used to divide the pixels in the sub-block into 5 clusters, the mean value of the gray scale of all the pixels in each cluster is calculated as the quantization value, and a 3-bit / pixel bitmap is generated, and the data amount after compression is bits.

[0086] The quantization values and the bitmap of all the sub-blocks are combined to form compressed data.

[0087] It should be noted that in other embodiments, the implementer can adjust the clustering algorithm or bitmap encoding method according to the actual implementation situation to adapt to specific hardware platforms or application requirements. For example, for edge devices with extremely limited computing resources, a fast threshold segmentation can be used instead of K-means clustering to reduce the computational complexity; for scenarios with extremely high requirements for surface defect detection, the maximum value of C can be appropriately increased (such as C=6 or C=7) to further improve the image quality of the key area.

[0088] S7, real-time monitoring and abnormality warning based on the compressed data.

[0089] Specifically, the edge computing device transmits the compressed data to the cloud or the local monitoring server, the cloud or the local monitoring server decodes the compressed data, reconstructs the image according to the quantization values and the bitmap, analyzes the feature changes of the contact area between the tool and the workpiece in the image, and generates an abnormality alarm when detecting surface topography abnormalities, tool wear or material delamination, etc., and sends a warning to the operator.

[0090] It should be noted that the convolutional neural network is used to identify surface topography abnormalities, tool wear or material delamination, etc. in this embodiment, and in other embodiments, the implementer can determine the abnormality detection algorithm according to the actual implementation situation.

Claims

1. A real-time monitoring method for aerospace parts processing line, characterized in that, The application relates to an image acquisition method and device for a spaceflight and aviation part processing line. The method comprises the following steps: acquiring images of the spaceflight and aviation part processing line, and performing gray scale processing on each acquired image frame; each gray scale image frame is divided into a plurality of non-overlapping blocks, the feature importance of each block is determined according to the gray scale change of the same block in different time gray scale images, the distribution dispersion degree of each block is determined according to the gray scale distribution of the pixel points in each block, and the quantization priority of each block is determined based on the distribution dispersion degree and the feature importance of each block; the quantization number of each block is determined based on a dynamic threshold self-adaptive mechanism of the system state and the processing process, and the blocks are compressed according to the quantization number of each block to obtain compressed data; real-time monitoring and abnormal early warning are performed based on the compressed data; the dynamic threshold self-adaptive mechanism comprises the following steps: dynamically adjusting the division threshold of the quantization priority according to the processing state at the current moment and the resource state of the edge computing device, determining the quantization number of each block based on the comparison result of the obtained dynamic division threshold and the quantization priority of each block, increasing the quantization number of each block to ensure the image quality when the processing process is in a high-risk stage, and reducing the quantization number of each block to ensure the real-time performance of the system when the resource of the edge computing device is tight.

2. A real-time monitoring method for aerospace parts processing line as claimed in claim 1 wherein, The feature importance satisfies the expression: ； wherein, denotes the sequence number of the current frame, denotes the feature importance of the block of the grayscale image of the frame, denotes the pixel point in the block of the grayscale image of the frame; denotes the pixel point in the block of the grayscale image of the frame; denotes the size of the block; denotes the number of historical frames for reference; denotes the hyperbolic tangent function; denotes the absolute value symbol.

3. A real time monitoring method for aerospace parts machining line as claimed in claim 1 wherein, The distribution dispersion degree satisfies the expression: ； in, Indicates the sequence number of the current frame. Indicates the first The first frame of the grayscale image The degree of dispersion of the distribution of each block; Indicates the first The first frame of the grayscale image In the block, the th The grayscale value of each pixel; Indicates the first The first frame of the grayscale image The minimum gray value in each block Indicates the first The first frame of the grayscale image The maximum grayscale value in each block; Indicates the size of the block; This represents the minimum value function.

4. A real-time monitoring method for aerospace parts processing line as claimed in claim 1 wherein, The quantization priority satisfies the expression: ； wherein, represents the quantization priority of the i-th patch of the j-th frame image; represents the quantization priority of the i-th patch of the j-th frame grayscale image; represents the quantization priority of the i-th patch of the j-th frame grayscale image; represents the distribution dispersion degree of the i-th patch of the j-th frame image; represents the distribution dispersion degree of the i-th patch of the j-th frame image; represents the distribution dispersion degree of the i-th patch of the j-th frame image; represents the feature importance degree of the i-th patch of the j-th frame grayscale image; represents the feature importance degree of the i-th patch of the j-th frame grayscale image; represents the feature importance degree of the i-th patch of the j-th frame grayscale image; represents the natural exponential function; is a hyperparameter for adjusting the sensitivity of the quantization priority to the feature importance degree.

5. A real-time monitoring method for aerospace parts processing line as claimed in claim 4 wherein, Hyperparameters The acquisition method is: According to the first The feature importance of all the patches in the frame grayscale image clusters all the patches into two classes, takes the mean of the feature importance of all the patches contained in the class as the representative feature importance of the class, and takes the mean of the representative feature importance of the two classes as the representative feature importance of the frame grayscale image. The value of the representative feature importance of the frame grayscale image is the value of the representative feature importance of the frame grayscale image.

6. A real time monitoring method for aerospace parts machining line as claimed in claim 1 wherein, The method for dynamically adjusting the division threshold of the quantization priority according to the processing state at the current moment and the resource state of the edge computing device comprises the following steps: a processing state index is determined according to the change of the feature importance of each block in the current gray scale image frame relative to the feature importance of the corresponding block in the historical gray scale image, a system resource state index is determined by combining the CPU usage, the memory occupancy and the bandwidth utilization of the edge computing device, and the division threshold of the quantization priority is dynamically adjusted according to the processing state index and the system resource state index.

7. A real-time monitoring method for aerospace parts machining line as claimed in claim 6 wherein, The processing state index satisfies the expression: ； wherein, denotes a sequence number of a current frame, denotes a processing state indicator of a frame grayscale image; denotes a total number of blocks; denotes a feature importance of a block of a frame grayscale image; and denote a mean and a standard deviation, respectively, of a feature importance of a block in past frame grayscale images, denotes a preset reference number; denotes a hyperbolic tangent function; denotes an absolute value sign.

8. A real-time monitoring method for aerospace parts processing line as claimed in claim 6 wherein, The method for obtaining the system resource state index comprises the following steps: the maximum value of the CPU usage, the memory occupancy and the bandwidth utilization of the edge computing device at the current moment is obtained, 1 is subtracted from the maximum value, and the obtained result is used as the system resource state index.

9. A real-time monitoring method for aerospace parts processing line as claimed in claim 6 wherein, The method for dynamically adjusting the division threshold of the quantization priority according to the processing state index and the system resource state index comprises the following steps: an initial division threshold is corrected according to the processing state index and the system resource state index to obtain an instant division threshold of the current gray scale image frame, the instant division threshold and the historical dynamic division threshold are weighted and averaged to obtain a dynamic division threshold of the current gray scale image frame; The instant division threshold satisfies the expression: ； wherein, represents the serial number of the current frame, represents the serial number of the first frame, represents the serial number of the first frame gray image, represents the serial number of the first instant division threshold value of the frame gray image; represents the serial number of the first initial division threshold value, represents the serial number of the first initial division threshold value, represents the serial number of the first machining state index corresponding to the frame gray image, represents the serial number of the first machining state index corresponding to the frame gray image, represents the serial number of the current system resource state index, represents the machining process normal state threshold value, represents the system resource sufficiency threshold value, represents the machining state adjustment coefficient, represents the system resource adjustment coefficient.

10. A real-time monitoring method for aerospace parts processing line as claimed in claim 1 wherein, The method for compressing the blocks according to the quantization number of each block to obtain compressed data comprises the following steps: When the quantization number of the block is C=1, the gray mean value of all pixels in the block is calculated as the only quantization value, and no bitmap is generated; when the quantization number of the block is C=2, the high and low mean values are calculated as the quantization values, and a bitmap of 1 bit / pixel is generated; when the quantization number of the block is C=3 or C=4, the K-means clustering algorithm is used to divide the pixels in the block into C clusters, the gray mean value of all pixels in each cluster is calculated as the quantization value, and a bitmap of 2 bits / pixel is generated; when the quantization number of the block is C=5, the K-means clustering algorithm is used to divide the pixels in the block into 5 clusters, the gray mean value of all pixels in each cluster is calculated as the quantization value, and a bitmap of 3 bits / pixel is generated; and the quantization values of all blocks and the bitmaps are combined to form compressed data.

Images