A visual monitoring system for high-density compression molding of composite energetic materials

Abstract

The application discloses a kind of composite energetic material high-density press forming visual monitoring system, specifically related to energetic material monitoring field, including multimodal control data acquisition module, image acquisition module, multiscale image fusion module, physical inversion module, process knowledge graph construction module and decision-making module.A kind of composite energetic material high-density press forming visual monitoring system is realized through multimodal control data acquisition module and image acquisition module that energetic material pressing process is continuously carried out without interruption, while guaranteeing the time sequence accurate alignment of mechanical parameter and observation data, provides the basis for capturing the continuous dynamic data of particle motion and crack propagation;Through multiscale image fusion module, the integration normalization of multi-source data under a unified coordinate system is realized, and the information fragmentation problem caused by multidimensional data fragmentation is alleviated;Through physical inversion module, the accurate quantification of particle motion and crack propagation under multi-field coupling is realized, and critical mutation characteristics are captured.

Description

Technical Field

[0001] This invention relates to the field of energetic materials monitoring technology, and more specifically, to a visual monitoring system for high-density compression molding of composite energetic materials. Background Technology

[0002] With the increasing demand for highly reliable energetic materials, the precise control of their pressing and molding process is particularly important. The traditional technical architecture involves applying unidirectional or bidirectional pressure to energetic material powder or molding powder in a closed mold using a press. Based on classical pressing theory, the particle system undergoes stages such as rearrangement and elastoplastic deformation to achieve densification. However, actual measurements show that due to the lack of observation capabilities, there is a lack of direct evidence for the causes of problems such as uneven density, stress concentration, and crack defects inside the pressed parts. Process optimization still largely relies on trial and error based on experience, which restricts the improvement of product performance and consistency.

[0003] To overcome the observation barriers of traditional compression, existing technology involves pressing the particles under a certain initial pressure on a press, removing them, and then performing a micro-CT scan. After the scan, the particles are placed back on the press and subjected to a high-pressure re-pressurization, followed by another micro-CT scan. This process is repeated to achieve phased testing of the particles during the compression process, shifting the focus from relying on macroscopic indirect signals to seeking direct, localized visual information.

[0004] However, in practical use, it still has some drawbacks. For example, the intermittent interruption of CT scans disrupts the continuity of the pressing process, making it impossible to obtain dynamic and continuous images of particle movement and crack propagation, thus making it difficult to capture transient evolution and critical abrupt changes. The structural limitations of simple transparent molds or single-window observations mean that only two-dimensional information from a single perspective can be obtained, making it impossible to simultaneously monitor the multi-dimensional evolution of radial, axial, and internal cross-sections, resulting in fragmented and one-sided observation information. These factors combined make it difficult for existing technologies to comprehensively, continuously, and quantitatively reveal the micromechanical mechanisms under multi-field coupling during the pressing process, thus failing to provide clear causal direct evidence for process optimization and mold design. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, the present invention provides a visual monitoring system for high-density compression molding of composite energetic materials, which solves the problems mentioned in the background art through the following solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A visualization monitoring system for high-density compression molding of composite energetic materials includes: Multimodal control data acquisition module: It is used to synchronously send trigger signals to the press controller and image acquisition module based on a preset timing reference, control the press to continuously press the energetic material molding powder placed in the transparent mold, and synchronously acquire the first process sequence parameters during the pressing process; Image acquisition module: In response to the trigger signal, it synchronously acquires the second process sequence parameters of the energetic material molding powder on the radial surface, axial end face and internal cross-section during the pressing process through the acquisition device deployed at the exit of the multi-channel integrated optical path; Multi-scale image fusion module: used to bind the first process sequence parameters and the second process sequence parameters, and through preprocessing operations, map them to a preset mold physical coordinate system to obtain the first process features; Physical inversion module: Calculates the full-field displacement field, velocity field and strain field of the particle system in the first process feature, and quantitatively extracts the geometric features and evolution sequence of the crack through the crack dynamic identification algorithm, and outputs the quantitative second process feature; Process knowledge graph construction module: used to receive the second process features and combine them with the first process sequence parameters in the pressing process to construct a process knowledge graph; Decision-making module: Based on the process knowledge graph and in response to preset target quality indicators, it outputs process parameter optimization recommendations and feeds them back to the multimodal control data acquisition module.

[0007] Preferably, the multimodal control data acquisition module acquires the first process sequence parameters, specifically including: At the start of the pressing process, a first trigger signal is generated and synchronously sent to the press controller to start continuous pressing according to the timing reference, and a second trigger signal is sent to the image acquisition module to start image acquisition, so that mechanical loading and optical observation start from the same absolute time starting point; During the continuous pressing of molding powder under the control of the press controller, the axial load signal and punch displacement signal of the press are collected in real time by the sensor, and a first timestamp from the time reference is added to each set of collected load and displacement data. The axial load signal, punch displacement signal, and corresponding first timestamp are aligned in real time to form the first process sequence parameters.

[0008] Preferably, the image acquisition module acquires the second process sequence parameters, specifically including: In response to the second trigger signal, multiple high-speed cameras arranged at the observation positions of the radial surface, axial end face and internal profile are simultaneously triggered to start image acquisition of energetic material molding powder during the pressing process. Throughout the pressing process, multiple high-speed cameras continuously acquired images, obtaining radial particle dispersion sequences, axial end-face deformation sequences, and internal cross-sectional structure evolution sequences, respectively. Each captured image frame is stamped with a second timestamp originating from the same temporal reference; Images from different observation locations that share the same second timestamp are used to generate a spatiotemporally unified multi-dimensional image data set, which is then used as the parameters of the second process sequence.

[0009] Preferably, the multi-scale image fusion module, after binding the first process sequence parameters and the second process sequence parameters, maps them to a preset mold physical coordinate system, specifically including: Establish a physical coordinate system for the mold corresponding to the transparent mold. ,in The shaft is parallel to the pressing axis; Based on the pre-calibrated parameters of each camera, the pixel coordinates of the images acquired by each acquisition device are obtained. The mapping relationship to the physical coordinate system of the mold; Using the mapping relationship, the multi-channel image data in the second process sequence parameters are uniformly converted to the mold physical coordinate system and stored together with the corresponding axial load signal and punch displacement signal to form a feature dataset arranged in time sequence, which serves as the first process feature.

[0010] Preferably, the physical inversion module obtains the second process feature, specifically including: Based on the continuous image frames arranged in time sequence in the first process feature, analysis nodes are set up in a preset grid matrix in each observation plane under the physical coordinate system of the mold. For each pair of image frames at adjacent time points, at each analysis node, the displacement vector of that node between adjacent time points is obtained by calculating the normalized cross-correlation coefficient and performing subpixel fitting. The displacement vectors of all nodes are integrated across the entire field to generate a continuous displacement field for the corresponding time interval; The instantaneous velocity field of the particle motion is calculated by performing time difference on the displacement field. Spatial gradient calculation is performed on the displacement field to obtain the normal strain and shear strain components in the observation plane, and then the principal strain field is calculated. Based on the spatial distribution of the principal strain field, strain localization regions are identified.

[0011] Preferably, the physical inversion module performs the following steps in the image sub-region corresponding to each analysis node: Within, by maximizing the image grayscale matrix at the reference time selected in the physical coordinate system of the mold. With the image grayscale matrix at time t to be analyzed Normalized cross-correlation coefficients between Specifically, it is expressed as: in, and These are respectively represented in the image pixel coordinate system, for the current frame Relative to reference frame Trial values ​​of integer pixel displacement in the row and column directions and These represent the traversal of image sub-regions. At that time, the row index and column index of the pixel within the sub-region, Represented as a reference frame At pixel position grayscale value at that location and Each is represented as a sub-region Inner reference frame With the current frame The average gray value in the image. Represented as the current frame The displacement estimate was offset. Then, the corresponding pixel position The grayscale value at that location; For the obtained displacement estimates Subpixel fitting is performed on the neighborhood of the node to obtain the displacement used to represent the subpixel accuracy of the analysis node in the image pixel coordinate system. Based on the mapping relationship of the multi-scale image fusion module, the image pixels are shifted. Converted to the physical coordinate system of the mold Displacement vector The components of its vector in the x and z directions are denoted as follows: and ; Based on the displacement field The instantaneous velocity field is calculated by differentiating time t. Specifically, it is expressed as: in, Represented as the first Each image acquisition moment, This represents the time interval between adjacent image frames.

[0012] Preferably, the physical inversion module defines strain components based on the displacement field, specifically as follows: in, and These represent the normal strain components in the x and z directions within the observation plane, respectively. Represented as shear strain components in the xz plane; And calculate the principal strain, specifically expressed as: in, include and , and These are represented as the first principal strain and the second principal strain, respectively.

[0013] Preferably, the physical inversion module, in acquiring the second process feature, further includes: The continuous image sequence in the first process feature is processed frame by frame to segment the potential crack area from the image background; Morphological skeletonization is performed on the segmented binary crack region to extract the crack center line with a single pixel width. The physical length of the crack centerline in the physical coordinate system of the mold is calculated, its orientation angle is obtained based on the main direction fitting, and its instantaneous propagation rate is calculated based on the extension change of the same crack between adjacent frames. The same crack identified at different times is correlated and tracked across frames, and the evolution sequence of crack length, orientation angle and propagation rate in time order is output.

[0014] The technical effects and advantages of this invention are as follows: 1. This invention enables the uninterrupted and continuous pressing process of energetic materials through a multimodal control data acquisition module and an image acquisition module, while ensuring precise temporal alignment between mechanical parameters and observation data. It breaks through the structural limitations of single-window observation, completely solves the problem of fragmentation and one-sidedness of observation caused by two-dimensional information from a single perspective, and realizes the synchronous capture and integration of multi-view, full-dimensional evolution information. It avoids the loss of transient evolution information caused by process interruption, and provides a foundation for capturing continuous dynamic data of particle motion and crack propagation. 2. This invention achieves the integration and normalization of multi-source data in a unified coordinate system through a multi-scale image fusion module, which alleviates the information fragmentation problem caused by the separation of multi-dimensional data and provides unified data support for the comprehensive analysis of subsequent micromechanical mechanisms. 3. This invention solves the problems of difficulty in capturing transient evolution and critical mutation phenomena and inability to quantitatively reveal micromechanical mechanisms through a physical inversion module, and achieves accurate quantification of particle motion and crack propagation under multi-field coupling, capturing critical mutation characteristics. Attached Figure Description

[0015] Figure 1This is a block diagram of a visualization monitoring system for high-density compression molding of composite energetic materials, provided according to an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0018] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0019] As attached Figure 1 The visualization monitoring system for high-density compression molding of composite energetic materials shown includes a multimodal control data acquisition module, an image acquisition module, a multi-scale image fusion module, a physical inversion module, a process knowledge graph construction module, and a decision-making module.

[0020] The multimodal control data acquisition module is used to synchronously send trigger signals to the press controller and image acquisition module based on a preset timing reference, control the press to continuously press the energetic material molding powder placed in the transparent mold, and synchronously acquire the first process sequence parameters during the pressing process.

[0021] Specifically, a unified timing reference is generated based on a high-precision clock source. At the start of the pressing process, a first trigger signal is generated and synchronously sent to the press controller to start continuous pressing, and a second trigger signal is sent to the image acquisition module to start image acquisition, so that mechanical loading and optical observation begin at the same absolute time starting point. Under the control of the press controller, during the uninterrupted continuous pressing of the molding powder, the axial load signal and punch displacement signal of the press are synchronously acquired in real time by sensors, and a first timestamp from the timing reference is added to each set of acquired load and displacement data. The axial load signal, punch displacement signal and the corresponding first timestamp are bound and aligned in real time to form the first process sequence parameters.

[0022] It should be noted that the control trigger signal starts from the same absolute time starting point through a multi-channel synchronous data acquisition card. The acquisition card provides a synchronous sampling clock for each sensor channel, and performs synchronous analog-to-digital conversion at a sampling rate of not less than 10kHz to ensure that the time deviation between each channel is less than one sampling interval. An internal counter generates a timestamp at the sampling moment, and its resolution matches the sampling interval. The converted raw data is then converted into physical quantities after real-time digital filtering. The internal clock of the press controller is aligned and calibrated with the system clock of the image acquisition computer to ensure that the synchronization error between its time reference and the core clock source is less than 100μs. The first timestamp is directly assigned by the high-precision clock source, or generated by the FPGA counter synchronized by it at the moment of sensor signal acquisition. The timestamp resolution is not less than 1 microsecond, thereby ensuring that transient events in a fast dynamic process can be distinguished. The first process sequence parameter is a continuous data sequence with equal time intervals or synchronized with physical events, which completely characterizes the macroscopic mechanical state of the entire pressing process.

[0023] In one embodiment, the trigger signal is a 5V TTL level square wave pulse, which is output to the digital input port of the press controller and the external trigger port of the high-speed camera, respectively; the first trigger signal contains a standardized instruction data packet, which is sent to the programmable logic controller of the press via the industrial Ethernet protocol, and its instruction content includes at least: target pressing force curve, pressing speed, holding time and number of cycles.

[0024] In one embodiment, the sensor includes at least a high-linearity pressure sensor and a grating ruler displacement sensor mounted on the press. The analog signals of the sensors are synchronously sampled via a 24-bit high-precision analog-to-digital converter. The pressure sensor's range must cover the maximum pressing force, and its nonlinearity must be less than 0.1%FS. The grating ruler has a resolution of not less than 1 micrometer.

[0025] The image acquisition module is used to respond to the trigger signal and simultaneously acquire the second process sequence parameters of the energetic material molding powder on the radial surface, axial end face and internal cross-section during the pressing process through the acquisition device deployed at the exit of the multi-channel integrated optical path.

[0026] Specifically, in response to the second trigger signal, multiple high-speed cameras positioned at the radial surface, axial end face, and internal profile observation positions are simultaneously triggered to begin image acquisition of energetic material molding powder during the pressing process. Throughout the pressing process, the multiple high-speed cameras continuously acquire images, obtaining radial particle dispersion sequences, axial end face deformation sequences, and internal profile structure evolution sequences, respectively. Each acquired image frame is stamped with a second timestamp originating from the same time series reference. Images from different observation positions that share the same second timestamp are used to generate a spatiotemporally unified multi-dimensional image data set, which serves as the second process sequence parameter.

[0027] It should be noted that the second timestamp originates from a high-precision clock source with the same origin as the first timestamp. After each frame of image is acquired by the camera, the precise timestamp at that time is embedded in the data packet header. To facilitate high-speed acquisition, high-brightness, low-ripple LED planar light sources are set for the radial, axial, and cross-sectional observation fields of view, respectively. The driving power of each light source is synchronously modulated by the trigger signal, reaching the rated brightness only during camera exposure, thereby ensuring image clarity while avoiding sample thermal effects. For radial observation, the camera lens is directly aligned with the outer surface of the transparent mold through the C / CS interface bracket, with the lens axis perpendicular to the observation surface and no other optical elements in between. For axial observation, the camera is mechanically docked with the optical exit of the light guide column or reflector group at the center of the upper / lower punch through the F interface lens, ensuring that the field of view is full and the image plane is clear. For cross-sectional observation, after the openable mold is opened, a camera equipped with a macro lens is used and precisely positioned in front of the cross-section for shooting through a three-dimensional adjustment frame.

[0028] In this embodiment, the high-speed camera is preferably a global shutter CMOS camera with a frame rate of no less than 500 frames per second at a full resolution of 1280×1024 pixels to ensure that it can distinguish millisecond-level particle movement and crack propagation transients; the single pixel size is no larger than 5.5μm, and with an optical lens of appropriate magnification, the spatial resolution is better than 15μm, which is sufficient to identify the individual movement and morphological changes of modeling powder particles with a particle size of about 100-500μm; at the same time, the signals are distributed in parallel to the external trigger port of each high-speed camera through coaxial cables of equal length, and all cameras are forced to start exposure at the same physical electrical signal edge in a hardware direct connection manner, eliminating random delays in software commands and controlling the time deviation between multiple images within the microsecond level.

[0029] In this embodiment, by synchronous triggering and uninterrupted acquisition, a continuous data stream with millisecond or even microsecond time resolution from the start to the end of the pressing is obtained. This is equivalent to recording a dynamic process in the entire time domain. It can not only record the entire process of crack initiation, expansion and penetration, but also analyze its instantaneous expansion rate and dynamic evolution path through high frame rate images, thus solving the problem of being unable to capture transient evolution.

[0030] The multi-scale image fusion module is used to bind the first process sequence parameters and the second process sequence parameters, and through preprocessing operations, map them to a preset mold physical coordinate system to obtain the first process features.

[0031] In one possible implementation, binding the first process sequence parameters with the second process sequence parameters and mapping them to a preset mold physical coordinate system includes: establishing a mold physical coordinate system corresponding to the transparent mold. ,in The axis is parallel to the pressing axis, and x is the horizontal direction of the observation plane; based on the pre-calibrated parameters of each camera, the pixel coordinates of the image acquired by each acquisition device are obtained. The mapping relationship to the physical coordinate system of the mold is established; using the mapping relationship, the multi-channel image data in the second process sequence parameters are uniformly converted to the physical coordinate system of the mold, and stored together with the corresponding axial load signal and punch displacement signal to form a feature dataset arranged in time sequence, which serves as the first process feature.

[0032] It should be noted that this is based on the first timestamp in the first process sequence parameters. The second timestamp of each frame of the image in the second process sequence parameters Frame-level matching is performed, and the matching execution steps are as follows: For each image frame Its timestamp is , under load With displacement A sequence of data points In the middle, find the absolute value of the time deviation. The smallest data point, if this minimum deviation is less than the preset synchronization tolerance threshold. If the match is successful, then the synchronization tolerance threshold is considered to be... Set it to less than 50% of the exposure time of a single frame image to achieve sub-millisecond alignment; set the corresponding load With displacement With image frame Bind; for points between two successful matches and The image frames between which the corresponding load and displacement values ​​can be obtained by... and and and Linear interpolation is performed between the data points to obtain a spatiotemporally strictly aligned fused data sequence. ,in The unified time index after fusion ensures the continuity of data and fundamentally solves the problem of evolutionary sequence breakage caused by staged interruptions in existing CT scanning methods.

[0033] In this embodiment, to achieve the mapping from multi-view 2D images to a unified physical space, the calibration process includes: before pressing, accurately placing a calibration plate with known size characteristics into the mold cavity; driving all cameras to simultaneously capture images of the calibration plate at multiple positions and orientations to ensure clear calibration features within the field of view of each camera; for each camera, using the Zhang Zhengyou calibration method, calculating its internal parameters and external parameters relative to the calibration plate coordinate system using multiple acquired images, wherein the internal parameters include, but are not limited to, focal length, principal point, distortion coefficient, etc., and the external parameters include at least the rotation matrix and translation vector; before coordinate mapping, correcting the pixel coordinates of the original image using the obtained radial distortion coefficient and tangential distortion coefficient to obtain distortion-free ideal pixel coordinates, and then substituting them into the homography transformation for subsequent calculations; for observing planar scenes such as the inner wall of the mold, directly calculating the homography matrix H to establish a direct mapping between the image pixel coordinates (u,v) and the physical plane coordinates (x,z) of the mold: Where s is the scale factor, which is an arbitrary non-zero scaling factor. Multiple image data streams are synchronized with the corresponding axial load signal and punch displacement signal, enabling the first... Frame corresponding time , This is expressed as frame rate.

[0034] It should be noted that the homography matrix H is obtained by extracting at least four sets of non-collinear feature point pairs from the calibration board image, establishing a system of equations, and solving the nine parameters of the H matrix using direct linear transformation. That is, H is solved by constructing and solving the overdetermined linear equation system Ah=0, where A consists of point pair coordinates and h is the element vector of the H matrix. In this embodiment, h33=1 is fixed, that is, eight degrees of freedom are actually solved. It mainly defines the projection relationship from the pixel plane to the physical plane by associating rotation and scaling through its upper left 2x2 submatrix, associating translation through the first two elements of the third column, and associating perspective transformation through the last row. The Zhang Zhengyou calibration method is a widely used classic camera calibration algorithm based on a planar calibration board.

[0035] In this embodiment, through a multi-path optical path integrated design, combined with a synchronous camera array arranged radially, axially, and in cross-section, synchronous monitoring of radial surface particle dispersion, axial end face overall deformation, and internal cross-sectional structure evolution during the pressing process is achieved. More importantly, through the precise calibration and coordinate unification algorithm of the multi-scale image fusion module, these two-dimensional image information from different perspectives are accurately mapped to a unified global physical coordinate system of the mold, thereby constructing a data set that can reflect multi-dimensional and spatially correlated evolution, completely overcoming the information limitations of a single perspective.

[0036] The physical inversion module calculates the full-field displacement field, velocity field, and strain field of the particle system in the first process feature, and quantitatively extracts the geometric features and evolution sequence of the crack through a crack dynamic identification algorithm, outputting a quantitative second process feature.

[0037] Furthermore, based on the continuous image frames arranged in a time sequence in the first process feature, analysis nodes are arranged in a preset grid array in each observation plane under the physical coordinate system of the mold; for each pair of image frames at adjacent time points, at each analysis node, the displacement vector of the node between adjacent time points is obtained by calculating the normalized cross-correlation coefficient and performing sub-pixel fitting; the displacement vectors of all nodes are integrated across the entire field to generate a continuous displacement field corresponding to the time interval; the instantaneous velocity field of particle motion is calculated by performing time difference on the displacement field; the spatial gradient of the displacement field is calculated to obtain the normal strain and shear strain components in the observation plane, and then the principal strain field is calculated, and the strain localization region is identified based on the spatial distribution of the principal strain field.

[0038] In this embodiment, the identification criteria for the strain localization region include: principal strain values. More than twice the average strain value of the material; principal strain gradient Exceeding a preset threshold; in this embodiment, the preset threshold is obtained by analyzing data from historical defect-free pressing and molding experiments; the area of ​​the connected region that meets the above conditions is greater than the area of ​​the minimum sensitive region; in this embodiment, the area of ​​the minimum sensitive region is 10 times the particle area; when the above conditions are met simultaneously, the region is automatically marked as a strain localization region.

[0039] It should be noted that in the image sub-region corresponding to each analysis node Within, by maximizing the image grayscale matrix at the reference time selected in the physical coordinate system of the mold. With the image grayscale matrix at time t to be analyzed Normalized cross-correlation coefficients between Specifically, it is expressed as: in, and These are respectively represented in the image pixel coordinate system, for the current frame Relative to reference frame Trial values ​​of integer pixel displacement in the row and column directions and These represent the traversal of image sub-regions. At that time, the row index and column index of the pixel within the sub-region, Represented as a reference frame At pixel position grayscale value at that location and Each is represented as a sub-region Inner reference frame With the current frame The average gray value in the image is calculated by summing the gray values ​​of all pixels within the defined sub-region Ω in real time and then dividing by the total number of pixels. This value is used to eliminate the influence of local brightness uniformity variations on matching accuracy. Represented as the current frame The displacement estimate was offset. Then, the corresponding pixel position The grayscale value at the location; the obtained displacement estimate. Subpixel fitting is performed on the neighborhood of the node to obtain the displacement used to represent the subpixel accuracy of the analysis node in the image pixel coordinate system. Based on the mapping relationship of the multi-scale image fusion module, the image pixels are shifted. Converted to the physical coordinate system of the mold Displacement vector The components of its vector in the x and z directions are denoted as follows: and Based on the displacement field The instantaneous velocity field is calculated by differentiating time t. Specifically, it is expressed as: in, Represented as the first Each image acquisition moment, This is represented as the time interval between adjacent image frames; in this embodiment, the obtained... The instantaneous velocity field has physical meaning, with units of millimeters per second; the strain components are defined based on the displacement field, specifically expressed as follows: in, and These represent the normal strain components in the x and z directions within the observation plane, respectively. The strain is expressed as the shear strain components in the xz plane; and the principal strain is calculated, specifically as follows: in, include and , and These are represented as the first principal strain and the second principal strain, respectively.

[0040] It should be noted that for the analysis of the full compression process, the incremental correlation method is preferred, using the previous frame of each frame as the reference frame for displacement calculation to accumulate the total displacement field and avoid mismatches caused by excessive deformation; image sub-regions The size should contain sufficient texture features and be set to 3 to 5 times the average diameter of the feature particles in the observation field of view. In this embodiment, if the particle image diameter is about 10 pixels, the sub-region can be set to 31×31 or 41×41 pixels. The grid step size, i.e. the spacing between analysis nodes, is set to 50% to 70% of the sub-region size to ensure the spatial continuity of the displacement field and avoid over-computation.

[0041] In this embodiment, when obtaining integer pixel displacement Then, in order to obtain sub-pixel precision displacement The preferred method is quadratic surface fitting, specifically: extracting... The normalized cross-correlation coefficient C in the 3×3 neighborhood centered at the center is fitted using the least squares method as follows: The quadratic surface, where, , , , , and These are the undetermined fitting coefficients, which have no independent physical meaning. Their values ​​are obtained by substituting the known (p,q,C) data points within a 3×3 neighborhood centered at the integer pixel displacement point (p,q) into the above model and solving it once using the least squares regression algorithm. The coordinates of the extreme points of the surface are then used as the subpixel displacements. To ensure accuracy while maintaining high computational efficiency, it is suitable for real-time or near-real-time processing.

[0042] Furthermore, the continuous image sequence in the first process feature is processed frame by frame to segment potential crack regions from the image background; the segmented binary crack regions are subjected to morphological skeletonization processing to extract the crack centerline with a single pixel width; the physical length of the crack centerline in the mold physical coordinate system is calculated, its direction angle is obtained based on the main direction fitting, and its instantaneous propagation rate is calculated based on the extension change of the same crack between adjacent frames; the same crack identified at different times is correlated and tracked across frames, and the evolution sequence of crack length, direction angle and propagation rate in time order is output.

[0043] In this embodiment, the segmentation threshold for potential crack regions from the image background is automatically calculated using the maximum inter-class variance method. The optimal threshold is determined by maximizing the inter-class variance between background and crack pixels, without manual intervention, and adaptable to different lighting and contrast conditions. To highlight the linear crack features, eigenvalue filtering based on the Hessian matrix is ​​used. For each pixel in the image, its second Gaussian derivative with scale σ is calculated to form a Hessian matrix. By analyzing the eigenvalues ​​of this matrix, the linear structure at a specific scale can be enhanced, i.e., one eigenvalue is much larger than the other. The filter kernel size, i.e., the Gaussian scale σ, is related to the typical width of the crack to be detected and is experimentally set to a pixel value slightly larger than the width of the crack image.

[0044] It should be noted that for the skeletonized crack, its physical length is obtained by accumulating the center distance of adjacent pixels after 8-neighbor chain code tracking of all pixels of the skeleton line, and multiplying it by the calibration scale. The minimum crack length is used as a criterion to filter out small pseudo-cracks caused by noise. Its value is determined according to the image resolution and actual research needs, and is usually the physical length corresponding to 2 to 5 times the pixel size. The cross-frame association of cracks adopts the Hungarian algorithm based on multi-feature matching. Taking the crack object of the previous frame as the reference, a matching cost is calculated for each crack object in the current frame. The cost function considers: the Euclidean distance of the endpoint position, the change of crack direction angle, and the continuity of crack length. By solving the matching pair with the minimum total cost, robust association can be achieved even if cracks branch or merge.

[0045] In this embodiment, during a typical 2-second pressing process, the present invention can acquire more than 2,000 consecutive images and corresponding load-displacement curves, while CT scans can usually only provide 2 to 3 static three-dimensional images before and after pressurization. This makes it possible to capture the critical abrupt change phenomenon of crack initiation and rapid propagation within tens of milliseconds, which is something that CT methods cannot achieve. The present invention directly outputs full-field, continuous, and quantified physical field data and defect evolution sequences. By analyzing the strain field cloud map, it can directly and quantitatively locate the region where the maximum principal strain exceeds 1% of the critical value, and accurately read the instantaneous pressure value when the strain concentration region appears. This provides a precise basis for process optimization and mold design that is unattainable by traditional methods.

[0046] The process knowledge graph construction module is used to receive the second process feature and, in conjunction with the first process sequence parameters in the pressing process, construct a process knowledge graph.

[0047] Specifically, constructing the process knowledge graph includes: receiving and storing quantitative second process features from the physical inversion module and first process sequence parameters from the multimodal control data acquisition module, and designing a relational database; mining frequent itemsets and strong association rules in historical data; in this embodiment, the minimum support is set to 0.1, indicating that the rule appears in more than 10% of the experiments, and the minimum confidence is set to 0.8, then the rule mined is: {pressing speed > 5 mm / s, high initial porosity} is {crack initiation in the mold corner region}, where the minimum support and minimum confidence are empirical parameters used to control the universality and reliability of the rule, and can be adjusted by the user according to the amount of data; using a random forest regression model to predict the final quality index; in this embodiment, the process parameters and the maximum strain of the whole process response are used as input features, and the final density uniformity is used as the prediction target; a PC algorithm is introduced for causal structure learning, with process parameters and process response variables as nodes, irrelevant edges are gradually eliminated based on conditional independence tests, and finally a directed acyclic graph is output.

[0048] In one embodiment, the construction and training of the random forest regression model includes: extracting data from all completed experiments from a historical experimental database, treating each experimental record as a sample; the input features of each sample include: key process parameters: maximum pressing force, average pressing speed, and process response features output from the physical inversion module during the pressing process: maximum principal strain value, average strain rate, and particle velocity variance for that stage; the output label of each sample is the final quality index of the corresponding pressed body: density uniformity, i.e., the standard deviation of density at multiple measurement points; the dataset is randomly divided into a training set and a test set in a 7:3 ratio; on the training set, grid search and cross-validation are used to optimize the model hyperparameters, including: the number of decision trees is 100, the maximum tree depth is 10, and the minimum number of samples required for leaf nodes is set to 5; the prediction performance of the final model on the test set must meet the preset requirements, i.e., the coefficient of determination calculated on the independent test set is used as the evaluation index, and it can be put into application only when it is greater than 0.85.

[0049] In one embodiment, the PC algorithm is implemented as follows: each pressing experiment is considered as an independent data sample, and the node variables of the causal network are defined, including: process parameter nodes, process response nodes, and quality index nodes; starting from a completely undirected graph with edges between all nodes, based on the sample data, conditional independence tests are performed on each pair of nodes under the condition of a given set of other possible nodes. In this embodiment, a partial correlation test under the Gaussian distribution assumption is used, and the significance level α is set to 0.05; if the test shows that a pair of nodes is independent under the given set of possible nodes, the edge between them is removed; after multiple iterations, a skeleton graph is obtained, and the direction of some edges is determined based on the directionality rule, and a causal graph that may contain some directed edges is output. The causal graph obtained in this embodiment shows that pressing speed and die roughness are common causes affecting die angle strain, and die angle strain directly affects crack initiation.

[0050] It should be noted that the entities and relationships mined above are stored in a database and displayed using a front-end visualization library; in the knowledge graph, the node size represents the frequency of occurrence of the corresponding factor, and the edge thickness represents the strength of the relationship.

[0051] In this embodiment, the core data tables of the database include at least the following: an experimental record table: storing metadata such as experimental ID, material batch, and mold number; a process parameter table: using the experimental ID and timestamp as a joint primary key, storing pressure, displacement, pressing speed, etc. at each sampling moment; a process response table: also associated with the experimental ID and timestamp, storing the corresponding physical field statistical characteristics, including at least indicators such as maximum principal strain, average strain rate, number of identified cracks, total crack length, and intensity of particle rearrangement; a defect event table: recording key events, including but not limited to the initiation timestamp, location coordinates, pressure at initiation, final length, and maximum propagation rate of each crack; and a final quality table: recording the overall density, density distribution variance, and ultrasonic testing results of the pressed body corresponding to the experimental ID. All tables are associated through the experimental ID and high-precision timestamp to ensure that the process conditions, material response, and final quality at any given moment are traceable.

[0052] In this embodiment, by constructing a process knowledge graph, massive amounts of continuous, multi-dimensional, and quantitative data are transformed into structured knowledge. In a certain process knowledge graph, it is revealed that "when the pressing speed V>3mm / s, the probability of strain concentration at 10mm on the die corner (x=5mm, z=-10mm) exceeds 80%, which leads to radial cracks with a depth greater than 2mm." A strong correlation and even causal inference chain from process parameters to local mechanical response and then to specific defects is established, which greatly improves the accuracy and efficiency of process optimization.

[0053] The decision-making module is used to output process parameter optimization recommendations based on the process knowledge graph and in response to preset target quality indicators, and feed them back to the multimodal control data acquisition module.

[0054] Specifically, the strongly correlated rules extracted from the process knowledge graph each contain preconditions and conclusion actions. In real-time monitoring, the current process parameters and process response variables are matched with the preconditions in the rule base. When the process parameters enter their condition range, an early warning or adjustment suggestion is triggered. The key indicator curves of the pressing process are displayed in real time through the human-computer interaction interface, and the real-time judgment of the decision module is highlighted on the corresponding time point and image field of view. The causal chain evidence behind any optimization recommendation is displayed. In this embodiment, it is displayed in the form of a graph path: "Increase the initial pressing speed → (leads to) → insufficient particle rearrangement → (causes) → local porosity increase → (further) → crack initiation under intermediate pressure".

[0055] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A visual monitoring system for high-density compression molding of composite energetic materials, characterized in that, include: Multimodal control data acquisition module: It is used to synchronously send trigger signals to the press controller and image acquisition module based on a preset timing reference, control the press to continuously press the energetic material molding powder placed in the transparent mold, and synchronously acquire the first process sequence parameters during the pressing process; Image acquisition module: In response to the trigger signal, it synchronously acquires the second process sequence parameters of the energetic material molding powder on the radial surface, axial end face and internal cross-section during the pressing process through the acquisition device deployed at the exit of the multi-channel integrated optical path; Multi-scale image fusion module: used to bind the first process sequence parameters and the second process sequence parameters, and through preprocessing operations, map them to a preset mold physical coordinate system to obtain the first process features; Physical inversion module: Calculates the full-field displacement field, velocity field and strain field of the particle system in the first process feature, and quantitatively extracts the geometric features and evolution sequence of the crack through the crack dynamic identification algorithm, and outputs the quantitative second process feature; Process knowledge graph construction module: used to receive the second process features and combine them with the first process sequence parameters in the pressing process to construct a process knowledge graph; Decision-making module: Based on the process knowledge graph and in response to preset target quality indicators, it outputs process parameter optimization recommendations and feeds them back to the multimodal control data acquisition module.

2. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 1, characterized in that: The multimodal control data acquisition module acquires the parameters of the first process sequence, specifically including: At the start of the pressing process, a first trigger signal is generated and synchronously sent to the press controller to start continuous pressing according to the timing reference, and a second trigger signal is sent to the image acquisition module to start image acquisition, so that mechanical loading and optical observation start from the same absolute time starting point; During the continuous pressing of molding powder under the control of the press controller, the axial load signal and punch displacement signal of the press are collected in real time by the sensor, and a first timestamp from the time reference is added to each set of collected load and displacement data. The axial load signal, punch displacement signal, and corresponding first timestamp are aligned in real time to form the first process sequence parameters.

3. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 1, characterized in that: The image acquisition module acquires the second process sequence parameters, specifically including: In response to the second trigger signal, multiple high-speed cameras arranged at the observation positions of the radial surface, axial end face and internal profile are simultaneously triggered to start image acquisition of energetic material molding powder during the pressing process. Throughout the pressing process, multiple high-speed cameras continuously acquired images, obtaining radial particle dispersion sequences, axial end-face deformation sequences, and internal cross-sectional structure evolution sequences, respectively. Each captured image frame is stamped with a second timestamp originating from the same temporal reference; Images from different observation locations that share the same second timestamp are used to generate a spatiotemporally unified multi-dimensional image data set, which is then used as the parameters of the second process sequence.

4. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 1, characterized in that: The multi-scale image fusion module binds the first process sequence parameters and the second process sequence parameters, and then maps them to a preset mold physical coordinate system, specifically including: Establish a physical coordinate system for the mold corresponding to the transparent mold. ,in The shaft is parallel to the pressing axis; Based on the pre-calibrated parameters of each camera, the pixel coordinates of the images acquired by each acquisition device are obtained. The mapping relationship to the physical coordinate system of the mold; Using the mapping relationship, the multi-channel image data in the second process sequence parameters are uniformly converted to the mold physical coordinate system and stored together with the corresponding axial load signal and punch displacement signal to form a feature dataset arranged in time sequence, which serves as the first process feature.

5. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 1, characterized in that: The physical inversion module obtains the second process feature, specifically including: Based on the continuous image frames arranged in time sequence in the first process feature, analysis nodes are set up in a preset grid matrix in each observation plane under the physical coordinate system of the mold. For each pair of image frames at adjacent time points, at each analysis node, the displacement vector of that node between adjacent time points is obtained by calculating the normalized cross-correlation coefficient and performing subpixel fitting. The displacement vectors of all nodes are integrated across the entire field to generate a continuous displacement field for the corresponding time interval; The instantaneous velocity field of the particle motion is calculated by performing time difference on the displacement field. Spatial gradient calculation is performed on the displacement field to obtain the normal strain and shear strain components in the observation plane, and then the principal strain field is calculated. Based on the spatial distribution of the principal strain field, strain localization regions are identified.

6. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 5, characterized in that: The physical inversion module, in the image sub-region corresponding to each analysis node Within, by maximizing the image grayscale matrix at the reference time selected in the physical coordinate system of the mold. With the image grayscale matrix at time t to be analyzed Normalized cross-correlation coefficients between Specifically, it is expressed as: in, and These are respectively represented in the image pixel coordinate system, for the current frame Relative to reference frame Trial values ​​of integer pixel displacement in the row and column directions and These represent the traversal of image sub-regions. At that time, the row index and column index of the pixel within the sub-region, Represented as a reference frame At pixel position grayscale value at that location and Each is represented as a sub-region Inner reference frame With the current frame The average gray value in the image. Represented as the current frame The displacement estimate was offset. Then, the corresponding pixel position The grayscale value at that location; For the obtained displacement estimates Subpixel fitting is performed on the neighborhood of the node to obtain the displacement used to represent the subpixel accuracy of the analysis node in the image pixel coordinate system. Based on the mapping relationship of the multi-scale image fusion module, the image pixels are shifted. Converted to the physical coordinate system of the mold Displacement vector The components of its vector in the x and z directions are denoted as follows: and ; Based on the displacement field The instantaneous velocity field is calculated by differentiating time t. Specifically, it is expressed as: in, Represented as the first Each image acquisition moment, This represents the time interval between adjacent image frames.

7. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 6, characterized in that: The physical inversion module defines strain components based on the displacement field, specifically as follows: in, and These represent the normal strain components in the x and z directions within the observation plane, respectively. Represented as shear strain components in the xz plane; And calculate the principal strain, specifically expressed as: in, include and , and These are represented as the first principal strain and the second principal strain, respectively.

8. The visualization monitoring system for high-density compression molding of composite energetic materials according to claim 1, characterized in that: The physical inversion module, which obtains the second process feature, specifically includes: The continuous image sequence in the first process feature is processed frame by frame to segment the potential crack area from the image background; Morphological skeletonization is performed on the segmented binary crack region to extract the crack center line with a single pixel width. The physical length of the crack centerline in the physical coordinate system of the mold is calculated, its orientation angle is obtained based on the main direction fitting, and its instantaneous propagation rate is calculated based on the extension change of the same crack between adjacent frames. The same crack identified at different times is correlated and tracked across frames, and the evolution sequence of crack length, orientation angle and propagation rate in time order is output.

Images