A tool parameter dynamic matching method and device based on deep learning

Abstract

The application provides a kind of tool parameter dynamic matching method and device based on deep learning, it is related to data processing field.In the method, laser holographic imaging data and three-dimensional profile scanning data for machine tool cutting area at target timestamp are obtained;According to laser holographic imaging data and three-dimensional profile scanning data, holographic voxel data and trimmed point cloud data are determined;Holographic voxel data and trimmed point cloud data are input into four-dimensional grid containing target timestamp, to obtain four-dimensional snapshot sequence;First tool parameter of machine tool tool at target timestamp is obtained;Four-dimensional snapshot sequence and first tool parameter are input into preset deep learning causal super network, to generate second tool parameter.Implementation of the technical solution provided in the application facilitates dynamic matching of tool parameters during CNC machine tool processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a tool parameter dynamic matching method and device based on deep learning. BACKGROUND

[0002] With the development of manufacturing industry towards intelligence, flexibility and high efficiency, the adjustment of tool parameters of numerical control machine tools has become a key link to improve machining quality and production efficiency.

[0003] At present, the existing technology often uses traditional vibration sensing and two-dimensional vision system to adjust tool parameters, but due to the inherent sampling principle and resolution limit of this way, it is difficult to capture the micro-topography changes of the tool and workpiece contact area. For example, the vibration signal can only reflect the overall cutting force and the change of machine tool stiffness, and it is almost insensitive to micron-level tool wear or surface details; and the two-dimensional camera is affected by the angle of view, lighting and shielding, and can only provide the projection image of the tool side or the workpiece side, and cannot restore the dynamic evolution of the three-dimensional microstructure. Therefore, the above-mentioned method is not conducive to dynamically matching tool parameters during numerical control machine tool machining.

[0004] Therefore, there is an urgent need for a tool parameter dynamic matching method and device based on deep learning. SUMMARY

[0005] The present application provides a tool parameter dynamic matching method and device based on deep learning, which facilitates dynamic matching of tool parameters during numerical control machine tool machining.

[0006] In a first aspect of the present application, a tool parameter dynamic matching method based on deep learning is provided, the method comprising: acquiring laser holographic imaging data and three-dimensional profile scanning data of a machine tool cutting area at a target timestamp; determining holographic voxel data and cropped point cloud data according to the laser holographic imaging data and the three-dimensional profile scanning data; inputting the holographic voxel data and the cropped point cloud data into a four-dimensional grid containing the target timestamp to obtain a four-dimensional snapshot sequence; acquiring a first tool parameter of a machine tool at the target timestamp; inputting the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal super network to generate a second tool parameter.

[0007] By adopting the technical scheme, through the introduction of the multi-modal perception means of laser holographic imaging and three-dimensional profile scanning, combined with the high-resolution voxelization and point cloud clipping technology, a four-dimensional snapshot sequence accurately reflecting the space-time changes in the cutting area is constructed, which not only greatly improves the visualization and quantification ability of the interaction process between the tool and the workpiece, but also realizes the leap of the cutting state information from two-dimensional projection to three-dimensional structure and even four-dimensional dynamics. On this basis, by fusing the four-dimensional snapshot and the current tool parameter into the preset causal super network model, the method breaks through the bottleneck of strong dependence on correlation and poor interpretability of traditional vibration sensing and two-dimensional visual system, so that not only reasonable prediction can be made based on the historical state, but also the deep logical relationship between tool parameter adjustment and actual cutting effect can be excavated from the perspective of causal driving, thereby generating a second tool parameter with more physical meaning and processing reliability. The overall scheme has the advantages of high precision, high adaptability and high interpretability, significantly improves the response speed and intelligent level of tool parameter adjustment, and provides strong technical support for realizing adaptive cutting and intelligent decision-making under complex working conditions. Therefore, it is convenient to dynamically match the tool parameters during numerical control machine tool machining.

[0008] Optionally, the laser holographic imaging data and the three-dimensional profile scanning data of the cutting area of the machine tool at the target timestamp are obtained, specifically including: sending a synchronous acquisition instruction to the laser holographic camera and the three-dimensional profile scanner respectively; reading the timestamp returned by the laser holographic camera and the three-dimensional profile scanner to determine the timestamp as the target timestamp; under the constraint of the target timestamp, receiving the interference fringe sequence sent by the laser holographic camera and receiving the point cloud batch sent by the three-dimensional profile scanner to obtain the laser holographic imaging data and the three-dimensional profile scanning data.

[0009] By adopting the technical scheme, through the realization of accurate synchronous acquisition control between the laser holographic camera and the three-dimensional profile scanner, it is ensured that the two complete data acquisition at the same target timestamp, effectively solving the problems of time misalignment and inconsistent information existing in the traditional multi-modal sensor in the machining monitoring. Specifically, by actively sending a synchronous acquisition instruction and reading a returned timestamp, the strict alignment of the interference fringe sequence and the point cloud batch in time is ensured, so that the two types of data have high timeliness and correspondence. This design not only avoids the complex time calibration process in the later stage, reduces the risk of feature mismatch caused by time drift, but also lays a high-quality data foundation for subsequent construction of a unified four-dimensional snapshot. Further, the laser holographic camera can capture microscopic interference information, reflecting the implicit structural features such as material internal stress and deformation, while the three-dimensional profile scanner provides macroscopic topography and surface profile changes. Through the collaborative fusion of the data from the two sources, dynamic perception from microscopic stress evolution to macroscopic topography change can be realized, greatly improving the comprehensiveness and depth of cutting process modeling.

[0010] Optionally, the determining holographic voxel data and cropped point cloud data according to the laser holographic imaging data and the three-dimensional profile scanning data specifically comprises: performing artifact removal and brightness equalization on an interference fringe pattern contained in the laser holographic imaging data to obtain an initialized image; mapping the initialized image to a three-dimensional voxel according to a predefined spatial grid, each three-dimensional voxel comprising a corresponding local phase and local light intensity; performing statistical filtering on a profile scanning point cloud contained in the three-dimensional profile scanning data to obtain an initialized point cloud; determining a surface normal on the initialized point cloud to obtain an intended region; and converting the three-dimensional voxel and the intended region into a unified coordinate and data structure respectively to obtain the holographic voxel data and the cropped point cloud data.

[0011] By adopting the above technical solution, firstly, the interference fringe pattern in the laser holographic data is subjected to artifact removal and brightness equalization processing, effectively suppressing the interference of environmental noise and uneven illumination and the like, significantly improving the image quality and the accuracy of physical measurement. Subsequently, by mapping the optimized image to a predefined three-dimensional grid structure, a voxel field containing local phase and light intensity information is constructed, not only realizing spatial coding of microstructure changes, but also retaining the continuity and interpretability of material stress field and deformation process. At the same time, in the processing of three-dimensional profile scanning data, statistical filtering means is used to remove abnormal points and outliers, improving the density and effectiveness of the point cloud; further calculating the surface normal to determine the intended machining area, so that the processing process is more focused on the actual tool-workpiece interaction part, enhancing the data utilization efficiency and the pertinence of the downstream model. Finally, the holographic voxel and the cropped point cloud are converted into a unified coordinate and structure, realizing the fusion of heterogeneous data at the spatial and semantic levels, greatly reducing the complexity of multi-modal information in joint modeling.

[0012] Optionally, the inputting the holographic voxel data and the cropped point cloud data into a four-dimensional grid containing the target timestamp to obtain a four-dimensional snapshot sequence specifically comprises: obtaining engineering requirements of the machine tool cutting area; setting a spatial three-dimensional resolution and a time step according to the engineering requirements; filling the holographic voxel data into the four-dimensional grid according to the three-dimensional resolution to obtain a first time slice; filling the cropped point cloud data into the four-dimensional grid according to the time step to obtain a second time slice; and sequentially splicing the first time slice and the second time slice in time order to obtain the four-dimensional snapshot sequence.

[0013] By adopting the technical scheme, the holographic voxel data and the cropped point cloud data are sequentially embedded into a four-dimensional grid structure containing time information, which not only realizes deep fusion of multi-source data in a unified space-time framework, but also significantly improves the continuity and expressiveness of dynamic modeling of the cutting process. First, by obtaining the engineering requirements of the machine tool cutting area, and setting appropriate spatial three-dimensional resolution and time step according to the requirements, it is ensured that the constructed four-dimensional grid can balance data accuracy and real-time performance, meeting the detailed expression of micro-topography evolution and ensuring reasonable allocation of computing resources and model response speed. In the spatial dimension, the holographic voxel data reflects the structural characteristics such as stress distribution and material deformation in the cutting area, and by mapping to the first-time slice in the grid, the initial state information of processing is finely encoded; in the time dimension, the cropped point cloud data dynamically reflects the trajectory of surface topography evolution over time, and by filling in the subsequent time slices, the evolution trend of micro-geometric structure in the cutting process is truly reproduced. Finally, the slices at different time points are spliced in time sequence to form a complete four-dimensional snapshot sequence, successfully realizing the up-conversion expression of static voxel and point cloud data to dynamic and continuous space evolution process.

[0014] Optionally, the first tool parameter of the machine tool at the target timestamp is obtained, specifically including: querying the cutting speed, feed rate and cutting depth of the machine tool at the target timestamp from the CNC controller; synchronously obtaining the tool model and wear grade of the machine tool; determining the first tool parameter according to the cutting speed, feed rate, cutting depth, tool model and wear grade.

[0015] By adopting the technical scheme, first, the cutting speed, feed rate and cutting depth, as the core process parameters determining the machining efficiency, surface quality and tool load, directly reflect the operation behavior of the tool under the current machining condition, and accurate extraction of these indicators can effectively lock the actual machining state, avoiding the prediction deviation caused by the inconsistency between offline setting and online actual. Secondly, the tool model and wear grade, as the key description reflecting the physical performance of the tool, cover the comprehensive influence of tool geometric structure, material composition and service life stage, and can reveal the cutting performance changes caused by the difference in tool health status under the same process parameters. By synchronously obtaining and jointly modeling the five types of parameters, the first tool parameter not only has the operation instruction information in the process dimension, but also includes the state description in the physical level, realizing the accurate association among operation, carrier and efficiency.

[0016] Optionally, the inputting the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal super network to generate a second tool parameter specifically comprises: performing feature normalization and dimensionality reduction mapping on the four-dimensional snapshot sequence and the first tool parameter to obtain a spatiotemporal feature abstract and a tool parameter feature; concatenating the spatiotemporal feature abstract and the tool parameter feature into a unified vector, and appending key node indicators marked in a preset causal graph to obtain a target analysis vector, the key node indicators including a material type and an environmental temperature; using the preset deep learning causal super network constructed by the preset causal graph to infer the target analysis vector in combination with a super network weight to obtain an inference parameter adjustment amount; and superimposing the inference parameter adjustment amount and the first tool parameter to obtain the second tool parameter.

[0017] By using the above technical solution, the causal super network inference framework is introduced on the basis of processing the four-dimensional snapshot sequence and the first tool parameter, breaking through the limitations of poor interpretability and weak generalization ability of traditional data, and significantly improving the scientificity and intelligent level of tool parameter adjustment. First, the four-dimensional snapshot sequence and the tool parameter are processed by using feature normalization and dimensionality reduction mapping, which effectively compresses high-dimensional redundant information while retaining key spatiotemporal change trends and parameter change characteristics, so that the input vector not only maintains the stability of model training, but also enhances the semantic expression ability. Subsequently, by concatenating the extracted spatiotemporal feature abstract and tool parameter feature into a unified vector, and introducing key node indicators such as material type and environmental temperature, a target analysis vector facing physical logic is constructed, which fully reflects the influence path of multi-factor interaction on cutting behavior, expands the perception boundary of the model, and makes the inference process no longer limited to surface statistical correlation, but integrates the causal constraint relationship of the real processing background. The inference result is output in the form of a parameter adjustment amount, which is weighted and superimposed with the first tool parameter to generate a second tool parameter, ensuring that the adjustment result not only has strong responsiveness, but also has dynamic adaptability and actual feasibility.

[0018] Optionally, the method further comprises: receiving a user-specified tool parameter sent by the user for the machine tool cutting area; and controlling the machine tool cutter to dynamically match according to the user-specified tool parameter.

[0019] By adopting the technical scheme, the user-defined tool parameters are introduced and directly applied to the machine tool as dynamic control instructions, which not only breaks through the closed loop link of human-computer interaction, intelligent decision and real-time control, but also significantly enhances the personalized response capability and flexible manufacturing level of the whole system in complex machining environment. First, the user can set tool parameters according to specific process requirements, machining tasks or real-time observation results, including but not limited to cutting speed, feed rate, cutting depth and other key indicators. This hybrid decision-making mechanism combining human intervention and data-driven breaks the passive control mode of traditional automation systems, and makes human experience and intelligent reasoning system form organic cooperation, which is especially suitable for complex scenes such as sudden interference in machining, personalized customization or uncertain material properties.

[0020] In a second aspect of the present application, a tool parameter dynamic matching device based on deep learning is provided. The device comprises an acquisition module and a processing module. The acquisition module is configured to acquire laser holographic imaging data and three-dimensional profile scanning data of a machining area of a machine tool at a target timestamp. The processing module is configured to determine holographic voxel data and cropped point cloud data based on the laser holographic imaging data and the three-dimensional profile scanning data. The processing module is further configured to input the holographic voxel data and the cropped point cloud data into a four-dimensional grid containing the target timestamp to obtain a four-dimensional snapshot sequence. The acquisition module is further configured to acquire a first tool parameter of a machine tool at the target timestamp. The processing module is further configured to input the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal super-network to generate a second tool parameter.

[0021] In a third aspect of the present application, an electronic device is provided. The electronic device comprises a processor, a memory, a user interface and a network interface. The memory is configured to store instructions. The user interface and the network interface are configured to communicate with other devices. The processor is configured to execute the instructions stored in the memory to enable the electronic device to perform the method described above.

[0022] In a fourth aspect of the present application, a computer readable storage medium is provided. The computer readable storage medium stores instructions which, when executed, perform the method described above.

[0023] In summary, one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0024] By introducing the multi-modal perception means of laser holographic imaging and three-dimensional profile scanning, combined with high-resolution voxelization and point cloud clipping technology, a four-dimensional snapshot sequence accurately reflecting the space-time changes in the cutting area is constructed, which not only greatly improves the visualization and quantification of the interaction process between the tool and the workpiece, but also realizes the leap of cutting state information from two-dimensional projection to three-dimensional structure and even four-dimensional dynamics. On this basis, by fusing the four-dimensional snapshot and the current tool parameter into the preset causal super network model, this method breaks through the bottleneck of strong dependence on relevance and poor interpretability of traditional vibration sensing and two-dimensional visual system, enabling not only reasonable prediction based on historical state, but also mining the deep logical relationship between tool parameter adjustment and actual cutting effect from the perspective of causal driving, thereby generating a second tool parameter with more physical meaning and processing reliability. The overall scheme has the advantages of high precision, high adaptability and high interpretability, significantly improving the response speed and intelligent level of tool parameter adjustment, and providing strong technical support for realizing adaptive cutting and intelligent decision-making under complex working conditions. Therefore, it is convenient to dynamically match the tool parameters during numerical control machine tool machining. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 A flowchart of a tool parameter dynamic matching device and method based on deep learning provided by an embodiment of the present application is shown.

[0026] Figure 2 Another flowchart of a tool parameter dynamic matching device and method based on deep learning provided by an embodiment of the present application is shown.

[0027] Figure 3 A module diagram of a tool parameter dynamic matching device and method based on deep learning provided by an embodiment of the present application is shown.

[0028] Figure 4 A structural diagram of an electronic device provided by an embodiment of the present application is shown.

[0029] Explanation of reference numerals: 31, acquisition module; 32, processing module; 41, processor; 42, communication bus; 43, user interface; 44, network interface; 45, memory. DETAILED DESCRIPTION

[0030] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be described clearly and completely in conjunction with the drawings in the embodiments of the specification. Obviously, the described embodiments are only some of the embodiments of the present application, not all.

[0031] In the description of the embodiments of the present application, the words such as "for example" or "for instance" are used to mean that an example, illustration or description is given. Any embodiment or design solution described as "for example" or "for instance" in the embodiments of the present application should not be interpreted as being more preferred or having more advantages than other embodiments or design solutions. Rather, the words such as "for example" or "for instance" are used in the specific manner to present the relevant concept.

[0032] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly indicating the technical features indicated. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. The terms "include", "contain", "have" and their variants mean "including but not limited to", unless otherwise specifically emphasized.

[0033] With the development of the manufacturing industry towards the stage of intelligence, flexibility and high efficiency, the dynamic adjustment of tool parameters in numerical control machine tools has gradually become one of the core technologies to improve the processing quality and production efficiency.

[0034] However, the prior art relies on traditional vibration sensors or two-dimensional vision systems to realize tool state perception and parameter adjustment. Limited by their own perception mechanism and physical resolution capability, it is difficult to meet the real-time mastering demand of micro information in high-precision machining scenarios.

[0035] Specifically, the vibration sensor mainly responds to the macro mechanical characteristics in the overall cutting process, such as cutting force fluctuation and structural rigidity change, and has almost no perception ability for micron-level tool wear, edge defects or contact surface micro-deformation; and the two-dimensional vision system is limited by single-view observation, changes in lighting conditions and occlusion, etc. The image obtained is only a two-dimensional projection of the tool or workpiece surface, and cannot reconstruct the real three-dimensional topography, let alone capture the evolution law in the dynamic machining process. For this reason, the current method often has problems such as parameter adjustment lag, inaccurate response and insufficient adjustment basis when dealing with high complexity and high precision machining tasks, and is difficult to support intelligent and dynamic matching of machine tool parameters.

[0036] To solve the above technical problems, the present application provides a tool parameter dynamic matching method based on deep learning, which refers to Figure 1 , Figure 1 A flowchart of a tool parameter dynamic matching method based on deep learning provided by an embodiment of the present application. The method is applied to a server and includes steps S110 to S150, which are as follows:

[0037] S110, acquiring laser holographic imaging data and three-dimensional profile scanning data of the machine tool cutting area at a target timestamp.

[0038] Specifically, one backend server in the system is responsible for sending data reading requests to two types of high-precision sensors arranged in the cutting area of the numerical control machine tool at a certain precise time (i.e., the target timestamp), and saving the original detection information collected by them for subsequent analysis and processing. Specifically, the laser holographic imaging data is a sequence of fringe images captured by a laser interference camera arranged around the contact area between the tool and the workpiece. These interference fringes contain micron or sub-micron level stress distribution and topography changes in the machining area. The three-dimensional profile scanning data is a point cloud set from a high-speed laser profile or confocal scanner, which is used to reconstruct the geometric profile of the cutting surface and the workpiece surface. The server ensures that the two types of data can be accurately aligned at the same time point through the target timestamp as a time reference, thereby eliminating the errors caused by different device clock drifts or trigger delays.

[0039] For example, in an aviation part processing workshop, when the machine tool processes to the key moment of 15.234 seconds of a certain tooling, the server issues a synchronous collection command to the laser holographic camera to obtain a set of interference fringe images representing this moment; at the same time, it initiates a point cloud sampling request to the three-dimensional profile scanner to obtain tens of thousands to millions of three-dimensional coordinate points at this moment. After receiving the two data, the server stores them in the database according to the unified time label and marks them with "cutting area-15.234s". Subsequently, whether it is based on four-dimensional field reconstruction of digital twin or parameter prediction of causal super network, the two types of high-precision sensor data after alignment can be accurately called to ensure the spatio-temporal consistency and physical reality of the model input.

[0040] In one possible implementation, the laser holographic imaging data and the three-dimensional profile scanning data of the machine tool cutting area at the target timestamp are acquired, specifically including: sending synchronous collection instructions to the laser holographic camera and the three-dimensional profile scanner respectively; reading the timestamps returned by the laser holographic camera and the three-dimensional profile scanner to determine the timestamps as the target timestamps; under the constraint of the target timestamps, receiving the interference fringe sequence sent by the laser holographic camera and the point cloud batch sent by the three-dimensional profile scanner to obtain the laser holographic imaging data and the three-dimensional profile scanning data.

[0041] Specifically, first, the server sends a "start collecting" instruction to the two devices, which carries a pre-set target timestamp, telling the sensor to start data capture at that time. Next, the sensor will complete the data capture internally and send back each frame of interference fringe pattern or each batch of point cloud data with a local timestamp. After receiving the data, the server will compare the timestamp returned by the sensor with the pre-set target timestamp. Only when the two timestamps are consistent or within an acceptable small time error range, the server will formally receive and archive the collected data, otherwise it will retry or alarm. This "pre-trigger, local collection, time verification, confirmation storage" closed-loop mechanism maximizes the risk of asynchronization caused by device response delay, network jitter or internal clock drift, ensuring that the obtained holographic image and three-dimensional point cloud are accurately aligned in time, laying a solid foundation for subsequent four-dimensional reconstruction and causal analysis.

[0042] For example, in an aerospace parts workshop, to monitor the microscopic wear of the tool when machining titanium alloy blades, the server will send a collection command to the laser holographic camera and high-speed three-dimensional profile scanner installed in the cutting area at the critical moment of 12.345 seconds after the start of machining through industrial Ethernet. The holographic camera immediately records a high-resolution image containing thousands of interference fringes and marks it as 12.345 seconds; at the same time, the profile scanner captures about 2 million three-dimensional surface points, also marked with the same timestamp. After comparison, the server confirms that the timestamps of the two data are consistent, and then archives them as "BladeCut_12.345s_holo" and "BladeCut_12.345s_cloud" for subsequent three-dimensional structure restoration and dynamic topography analysis at the same time. This alignment accuracy can reach milliseconds or even microseconds, so that even the slightest change in a very short moment during machining will not be missed, truly realizing high-precision and high-synchronization tool and workpiece state monitoring.

[0043] S120, determining holographic voxel data and cropped point cloud data according to the laser holographic imaging data and the three-dimensional profile scanning data.

[0044] Specifically, the server first reads the laser holographic interference fringe pattern that has been de-noised and corrected in brightness, divides it into a number of small cubic units (voxels) according to a predefined three-dimensional spatial grid. In each voxel, the server will count the phase difference and light intensity distribution of the corresponding pixels of the spatial unit, thereby generating a set of "holographic voxel data". These voxels not only retain the continuous spatial information of the material stress and micro-deformation in the cutting area, but also unify the coordinate systems of different sensing frames, facilitating subsequent spatio-temporal data fusion. For example, in a titanium alloy machining process, the server maps the 512x512 pixel interference pattern to a 200x200x200 cubic voxel grid, and each voxel records the local phase peak value and light intensity average, perfectly restoring the sub-micron stress distribution at the tool-workpiece contact.

[0045] At the same time, the server performs statistical filtering on the original point cloud data provided by the three-dimensional profile scanner, removes isolated noise points and invalid reflection points, and then automatically identifies the "intention area" based on the predicted tool geometry boundary or the real-time calculated surface normal, i.e. the part of the point cloud that is truly contacted by the tool and the workpiece. Next, the server only retains the valid point set in this area to form "clipped point cloud data" and converts it to the same coordinate system as the holographic voxel. For example, when the scanner collects 500000 scattered points, about 450000 valid points remain after three times standard deviation filtering, and about 60000 key points are extracted within a range of 2mm from the tool profile, ensuring that subsequent spatio-temporal modeling focuses on key information and avoids irrelevant background interference. The holographic voxels and clipped point clouds obtained after such processing are efficient and accurate, providing high-quality structured data support for constructing four-dimensional snapshots and causal reasoning.

[0046] In one possible implementation, the holographic voxel data and the clipped point cloud data are determined according to the laser holographic imaging data and the three-dimensional profile scanning data, specifically including: performing de-artifacting and brightness equalization on the interference fringe pattern contained in the laser holographic imaging data to obtain an initialization image; mapping the initialization image to three-dimensional voxels according to a predefined spatial grid, each three-dimensional voxel including a corresponding local phase and local light intensity; performing statistical filtering on the profile scanning point cloud contained in the three-dimensional profile scanning data to obtain an initialization point cloud; determining a surface normal on the initialization point cloud to obtain an intention area; converting the three-dimensional voxels and the intention area into a unified coordinate and data structure to obtain the holographic voxel data and the clipped point cloud data.

[0047] Specifically, first, the server will perform artifact removal and brightness equalization on the interference fringe pattern collected by the laser holographic camera: for example, when mirror reflection or dust and other factors produce conspicuous dark spots or highlights in the image, the artifact removal algorithm can automatically identify and repair these local distortions, and brightness equalization ensures that the light intensity distribution of the entire image is uniform, and details are not lost due to dark edges or bright centers. After processing, the system will convert each two-dimensional interference pattern "slice" into a series of voxels according to the predefined three-dimensional spatial grid (for example, a 200x200x200 cube): each voxel contains phase information (reflecting micro deformation) and light intensity value (reflecting material density or refractive index change) of the sub-region, accurately reproducing the micro stress and topography of the cutting area in three-dimensional space.

[0048] In parallel, the server will also perform statistical filtering and normal estimation on the original point cloud output by the three-dimensional profile scanner to extract the truly valuable "intention area". For example, the scanner may collect nearly a million naked points at a time, including debris and stray reflections of the machine base, fixtures, and tool-workpiece contact area. Statistical filtering will eliminate noise points that are isolated from the main cluster (such as flying chips), and normal calculation will help identify smooth areas where the tool edge actually contacts the workpiece surface. The system only retains these local point clouds, such as 50,000 valid points within 0-2 mm of the tool tip, and maps them to the same coordinate system and data structure as the holographic voxels. The final output of "holographic voxel data" and "clipped point cloud data" is two highly aligned, format-unified three-dimensional descriptions that retain both the hologram's detailed depiction of internal stress fields and the point cloud's geometric restoration of external contours, providing high-quality, directly callable inputs for subsequent four-dimensional grid construction and causal hypernet reasoning.

[0049] S130, input the holographic voxel data and the clipped point cloud data into the four-dimensional grid containing the target timestamp to obtain a four-dimensional snapshot sequence.

[0050] Specifically, the server will first create a four-dimensional array in memory according to the preset spatial resolution (such as subdividing a grid every cubic millimeter) and time step (for example, one frame per millisecond), with the coordinate axes corresponding to the X, Y, Z three spatial dimensions and the T time dimension. Subsequently, it will map the holographic voxel data at the target timestamp to the three-dimensional slice at that time point, recording the phase and light intensity information at the corresponding position in each grid cell; similarly, project the clipped point cloud data at the same timestamp into the same position three-dimensional slice to supplement the voxel information with geometric points and normal attributes. In this way, a single "time point" is no longer just a plane or point set, but a complete three-dimensional body that combines micro internal structure and external geometric contour.

[0051] For example, assume that when processing an aviation aluminum alloy plate, the server sets the spatial resolution to 0.5 mm and the time step to 2 ms. At this frame, t = 10.246 s, it first fills the 200 x 200 x 100 voxels obtained by holographic imaging at this time into the grid slice, reconstructing the internal stress distribution volume graph; then it maps the about 50,000 cutting point clouds collected at the same time onto the same slice, highlighting the surface micro profile and wear shape. Then, the server performs the same operation on t = 10.248 s, 10.250 s, etc. in increments of 2 ms, and finally outputs dozens or even hundreds of frames of continuous four-dimensional snapshot sequences. This sequence not only completely restores the evolution process of the tool and workpiece contact area at the micro level, but also provides high-fidelity, continuous, and consistent input data for subsequent dynamic analysis and parameter optimization based on neural fields or causal super networks.

[0052] In one possible implementation, the holographic voxel data and the cutting point cloud data are input into a four-dimensional grid containing a target timestamp to obtain a four-dimensional snapshot sequence, specifically including: obtaining engineering requirements of a machine tool cutting area; setting a three-dimensional spatial resolution and a time step according to the engineering requirements; filling the holographic voxel data into the four-dimensional grid according to the three-dimensional spatial resolution to obtain a first time slice; filling the cutting point cloud data into the four-dimensional grid according to the time step to obtain a second time slice; and sequentially splicing the first time slice and the second time slice in time order to obtain the four-dimensional snapshot sequence.

[0053] Specifically, first, the system will obtain the processing target from the process engineers, such as the need to monitor the micro deformation within a certain range of millimeters, or the need for a high time resolution to capture rapid vibration or wear process. Then, according to these requirements, the size of the grid in each spatial dimension (for example, per cubic millimeter, per cubic ten microns) and the sampling interval in the time dimension (for example, every 2 ms, every 5 ms) are set. This on-demand resolution and step configuration not only ensures that the grid can restore the subtle differences in internal stress and topography of the cutting area with sufficient accuracy, but also avoids data redundancy and computational bottlenecks caused by excessive resolution, leaving a controllable performance space for subsequent analysis and model reasoning.

[0054] Next, the system will fill the holographic voxel data and the cropped point cloud data obtained above into different "time slices" of the four-dimensional grid in turn according to the above settings: at a certain key moment (such as the moment when the tool begins to cut into the workpiece), the voxelized holographic information is first mapped into the first time slice to fully express the micro phase and light intensity distribution; then, the updated point cloud geometry information is filled into the next slice at a predetermined time step (such as 3 milliseconds later) to reflect the small changes in the surface profile. By connecting these three-dimensional slices in time sequence, the system finally generates a continuous four-dimensional snapshot sequence - just like a series of 3D snapshots are spliced into a "movie" of the micro level evolution of the cutting area over time. For example, in the processing of aerospace parts, engineers may require a spatial resolution of 0.5 mm and a time step of 2 milliseconds, so the system will fill in a frame of holographic voxels at t = 10.000s, a frame of cropped point clouds at t = 10.002s, and the next frame of voxels at t = 10.004s, and so on, expanding to form a complete four-dimensional sequence, providing a highly consistent and continuous spatiotemporal data basis for fine modeling, causal analysis, and online tool parameter optimization.

[0055] S140, acquiring a first tool parameter of the machine tool at a target timestamp.

[0056] Specifically, first, the server will read the cutting speed, feed rate, and cutting depth, etc. three core parameters corresponding to this time point; then, it will also retrieve the model identification of the current installed tool and the real-time wear level or tool life index recorded by the system. In this way, the server can accurately lock the cutting speed, material advancing rate, and cutting depth, etc. basic process parameters of the tool at this moment, while taking into account the physical state of the tool, providing complete and consistent initial basis for subsequent model input and parameter optimization.

[0057] In one possible implementation, acquiring a first tool parameter of the machine tool at a target timestamp specifically includes: querying the cutting speed, feed rate, and depth of cut of the machine tool at the target timestamp from the CNC controller; synchronously acquiring the tool model and wear level of the machine tool; determining the first tool parameter according to the cutting speed, feed rate, depth of cut, tool model, and wear level.

[0058] Specifically, first, the server makes a real-time request to the CNC controller through an industrial communication protocol (such as OPC-UA or MTConnect) to accurately obtain the three core process parameters at that moment, i.e., cutting speed, feed rate, and cutting depth. These three parameters directly reflect the motion state and cutting conditions of the tool under the current machining load and are the basic data for evaluating machining efficiency and surface quality. Next, the server will synchronously query the tool management system or the tool magazine information maintained internally by the CNC to read the model identification of the tool used at the moment (including geometric dimensions, coating type, etc.) and the wear level obtained through online monitoring or the pre-set maintenance system (such as new tool, light wear, moderate wear, etc.). The entire process requires time alignment and data consistency to ensure that each parameter obtained is strictly matched with the target timestamp without delay or frame loss, thereby providing reliable and synchronous input for subsequent modeling and adjustment.

[0059] For example, when machining a batch of aerospace aluminum alloy parts, the server will automatically initiate a data request to the CNC controller at the critical moment of 17.892 seconds in the machining program, receiving a cutting speed of 140 meters / minute, a feed rate of 0.18 millimeters / revolution, and a cutting depth of 2.8 millimeters. At the same time, through the same communication session, it also pulls the corresponding tool model "TiAlN-CoatedØ8mmEndMill" in the tool magazine and the "light wear" (about 25% of the service life) status determined by the online monitoring system. The server packages and organizes these process parameters and tool states to form a complete "first tool parameter" set, which not only covers the motion and cutting conditions of the tool at the moment, but also reflects the physical health level of the tool itself. This set of parameters can be directly fed back to the monitoring interface for engineers to view, or can be passed to intelligent algorithms such as causal super networks to generate more accurate and safer second tool parameter adjustment schemes.

[0060] S150, input the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal super network to generate a second tool parameter.

[0061] Specifically, the server first aggregates the formatted four-dimensional snapshot sequence (reflecting the micro-voxel information and surface point cloud features of the cutting zone at multiple time nodes) with the first tool parameters (including cutting speed, feed rate, cutting depth, tool model, and wear level, etc.) obtained in real time into a multi-dimensional input vector. This vector not only contains rich microstructure information evolving in space and time, but also includes the current process and physical state of the tool. Subsequently, the causal super network uses the node relationships and super network weights learned through causal graph constraints in the pre-training stage to automatically predict a set of optimal parameter adjustment amounts in the latent space along the real influence paths between variables in the causal structure. Finally, the server superimposes these adjustment amounts on the original parameters to form a second set of tool parameters with causal interpretability and physical reasonableness, and prepares them for real-time issuance to the machine tool.

[0062] wherein the preset deep learning causal super network is a deep learning model architecture combining super network mechanism and causal structure prior, which is specifically used to generate parameter adjustment strategies for specific dynamic scenarios. It contains two core components: causal graph module: a directed graph describing the causal relationships between key variables (such as cutting speed, feed rate, cutting depth, tool wear, material type, processing temperature, etc.) is pre-constructed by domain experts according to machine tool processing technology and physical mechanism, or through offline causal discovery algorithms; super network module: a generation network that accepts "target analysis vector" as input and outputs main network parameters (or directly outputs parameter increments). The super network is usually composed of multiple layers of fully connected networks or graph neural networks, and its weights and structure have been pre-trained according to historical multi-condition data and causal graph constraints in the offline stage, that is, the "preset" stage.

[0063] In the offline preset process, first, the historical processing data of multiple batches, multiple materials, and multiple tool combinations are used to verify and fine-tune the causal graph, ensuring that it can reflect the core causal pathways from process input to processing quality output; then, the super network learns a set of parameter sub-network weights that can quickly generate physical and process logic-compliant strategies under the guidance of this causal graph through multi-task or meta-learning methods. In the training target, not only the optimization loss of the final processing quality (such as surface roughness, vibration amplitude, tool life) is included, but also the causal consistency loss is introduced to force the strategy generated by the super network to function within the given causal structure.

[0064] In the online inference stage, when the system collects a new sequence of four-dimensional snapshots and the current tool parameters, they will be normalized, dimensionally reduced, and injected with key causal node indicators, and then sent to the pre-trained deep learning causal super network. The super network uses its internal causal constraints and empirical knowledge to quickly generate a set of adjustment amounts, such as "reduce cutting speed by 3 m / min, increase feed rate by 0.02 mm / rev", etc. This output not only conforms to the causal influence path learned by the model in the preset stage, but also can make reasonable responses to the current dynamic working conditions. The entire process does not require retraining of the main network, ensuring the efficiency and causal explainability of online matching, thereby achieving precise, reliable, and adaptive adjustment of tool parameters in complex and variable machining environments.

[0065] In one possible implementation, the four-dimensional snapshot sequence and the first tool parameter are input into the preset deep learning causal super network to generate the second tool parameter, specifically including: performing feature normalization and dimensionality reduction mapping on the four-dimensional snapshot sequence and the first tool parameter to obtain a spatiotemporal feature abstract and a tool parameter feature; concatenating the spatiotemporal feature abstract and the tool parameter feature into a unified vector, and appending key node indicators marked in the preset causal graph to obtain a target analysis vector, the key node indicators including a material type and an environmental temperature; using a preset deep learning causal super network constructed from the preset causal graph, combining the super network weight to infer the target analysis vector to obtain an inference parameter adjustment amount; and superimposing the inference parameter adjustment amount and the first tool parameter to obtain the second tool parameter.

[0066] Specifically, the server will first convert the four-dimensional snapshot sequence and the first tool parameter obtained above into "feature abstracts" that are easy for the model to process. Specifically, the voxel light intensity, phase distribution, and point cloud geometry information at each time in the four-dimensional snapshot sequence will be normalized to the same numerical range, and a number of key indicators representing spatiotemporal evolution trends will be extracted through dimensionality reduction mapping (such as principal component or autoencoder method), such as vibration amplitude change rate, micro deformation rate, etc.; At the same time, the cutting speed, feed rate, cutting depth, tool type, and wear grade in the first tool parameter are also mapped into a set of ordered numerical features and are also normalized. For example, if it is observed that the vibration gradually increases and the initial appearance of surface micro-cracks during a period of aluminum alloy processing, the server will condense this information into a "spatiotemporal feature abstract", and then map the original 150 m / min cutting speed, 0.12 mm / rev feed rate, 1.8 mm cutting depth, and "TiAlN coated tool - moderate wear" into corresponding parameter features.

[0067] Next, the server will splice the above two types of feature summaries into a complete vector, and attach key node indicators such as "material type = aluminum alloy" and "ambient temperature = 22°C" to form the final target analysis vector. This vector is input into the deep learning causal super network that has been trained according to the preset causal graph in the offline stage. The internal structure and weights of the super network carry causal path information such as "processing speed → thermal deformation → vibration" and "wear → surface quality", and can output a set of optimal parameter adjustment amounts in one forward reasoning. For example, the model may recommend reducing the cutting speed by 10 m / min, increasing the feed rate by 0.02 mm / rev, and keeping the depth of cut unchanged based on the current vibration trend and material properties. The server then adds these adjustment amounts to the original parameters to generate new second tool parameters (i.e., 140 m / min, 0.14 mm / rev, 1.8 mm) and prepares to issue them to the machine tool, achieving efficient and causally driven adaptive matching of tool states.

[0068] In one possible implementation, with reference to Figure 2 , Figure 2 Another flowchart of a deep learning-based tool parameter dynamic matching method provided by an embodiment of the present application. It includes steps S210 to S220, which are as follows: S210, receiving user-defined tool parameters sent for the machine tool cutting area; S220, controlling the machine tool to dynamically match the user-defined tool parameters.

[0069] Specifically, when the user inputs new cutting speed, feed rate, or depth of cut parameters in the monitoring interface or dedicated terminal, these custom instructions are packaged into standardized control messages and sent to the backend server. After receiving the message, the server first checks the parameter range and safety (such as not exceeding the maximum allowable speed of the machine tool, the upper limit of the feed torque, etc.), then converts the legal custom parameters into corresponding numerical control instruction formats (usually updating the server G-code server or issuing immediate parameter adjustment commands through the server OPC-UA server interface), and pushes them to the server CNC server controller in real time. In this way, the system not only realizes seamless connection between user intent and machine tool control, but also performs necessary protection on parameter legality before execution, ensuring that "human-machine collaboration" is both flexible and reliable.

[0070] Taking the machining of aerospace parts as an example, an operator might observe a slight surface scratch during a certain machining process and manually input an adjustment command on the monitoring interface: "Reduce the cutting speed from 150 m / min to 135 m / min, and increase the feed rate from 0.12 mm / rev to 0.14 mm / rev." Upon receiving the command, the server quickly verifies that these parameters are within the machine tool's allowable range and updates the corresponding G-code lines (e.g., G1, F0.14, S135) to the current machining program cache. Subsequently, the server sends the updated command to the machine tool drive system via the real-time communication bus, and the tool immediately executes the cutting action according to the new parameters at the start of the next cutting cycle. The entire process, from user input to machine tool response, can often be completed within a few hundred milliseconds, ensuring rapid improvement in machining quality while preserving the engineer's precise control over the overall process.

[0071] This application also provides a tool parameter dynamic matching device based on deep learning, referring to... Figure 3 , Figure 3 This is a schematic diagram of a deep learning-based tool parameter dynamic matching device provided in an embodiment of this application. The device is a server, which includes an acquisition module 31 and a processing module 32. The acquisition module 31 acquires laser holographic imaging data and three-dimensional contour scanning data of the machine tool cutting area at a target time stamp. The processing module 32 determines holographic voxel data and trimmed point cloud data based on the laser holographic imaging data and the three-dimensional contour scanning data. The processing module 32 inputs the holographic voxel data and trimmed point cloud data into a four-dimensional grid containing the target time stamp to obtain a four-dimensional snapshot sequence. The acquisition module 31 acquires the first tool parameters of the machine tool at the target time stamp. The processing module 32 inputs the four-dimensional snapshot sequence and the first tool parameters into a preset deep learning causal hypernet to generate the second tool parameters.

[0072] In one possible implementation, the acquisition module 31 acquires laser holographic imaging data and three-dimensional contour scanning data of the machine tool cutting area at a target timestamp. Specifically, this includes: the processing module 32 sending synchronous acquisition commands to the laser holographic camera and the three-dimensional contour scanner respectively; the processing module 32 reading the timestamps returned by the laser holographic camera and the three-dimensional contour scanner, and determining the timestamp as the target timestamp; and the processing module 32 receiving the interference fringe sequence sent by the laser holographic camera and the point cloud batch sent by the three-dimensional contour scanner under the constraint of the target timestamp, thereby obtaining the laser holographic imaging data and the three-dimensional contour scanning data.

[0073] In a possible implementation, the processing module 32 determines the holographic voxel data and the cropped point cloud data according to the laser holographic imaging data and the three-dimensional profile scanning data, specifically including: the processing module 32 performs de-artifacting and brightness equalization on the interference fringe pattern contained in the laser holographic imaging data to obtain an initialization image; the processing module 32 maps the initialization image to a three-dimensional voxel according to a predefined spatial grid, each three-dimensional voxel including a corresponding local phase and local light intensity; the processing module 32 performs statistical filtering on the profile scanning point cloud contained in the three-dimensional profile scanning data to obtain an initialization point cloud; the processing module 32 determines a surface normal on the initialization point cloud to obtain an intended region; and the processing module 32 converts the three-dimensional voxel and the intended region into a unified coordinate and data structure to obtain the holographic voxel data and the cropped point cloud data.

[0074] In a possible implementation, the processing module 32 inputs the holographic voxel data and the cropped point cloud data into a four-dimensional grid containing a target timestamp to obtain a four-dimensional snapshot sequence, specifically including: the acquisition module 31 acquires engineering requirements of a machine tool cutting area; the processing module 32 sets a spatial three-dimensional resolution and a time step according to the engineering requirements; the processing module 32 fills the holographic voxel data into the four-dimensional grid according to the three-dimensional resolution to obtain a first time slice; the processing module 32 fills the cropped point cloud data into the four-dimensional grid according to the time step to obtain a second time slice; and the processing module 32 sequentially splices the first time slice and the second time slice in time sequence to obtain the four-dimensional snapshot sequence.

[0075] In a possible implementation, the acquisition module 31 acquires a first tool parameter of a machine tool cutter at a target timestamp, specifically including: the processing module 32 queries the cutting speed, the feed rate, and the cutting depth of the machine tool cutter corresponding to the target timestamp from a CNC controller; the processing module 32 synchronously acquires the tool model and the wear grade of the machine tool cutter; and the processing module 32 determines the first tool parameter according to the cutting speed, the feed rate, the cutting depth, the tool model, and the wear grade.

[0076] In a possible implementation, the processing module 32 inputs the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal super-network to generate a second tool parameter, specifically including: the processing module 32 performs feature normalization and dimensionality reduction mapping on the four-dimensional snapshot sequence and the first tool parameter to obtain a spatiotemporal feature abstract and a tool parameter feature; the processing module 32 splices the spatiotemporal feature abstract and the tool parameter feature into a unified vector, and appends key node indicators marked in a preset causal graph to obtain a target analysis vector, the key node indicators including a material type and an environmental temperature; the processing module 32 uses a preset deep learning causal super-network constructed by the preset causal graph to infer the target analysis vector in combination with super-network weights to obtain an inference parameter adjustment amount; and the processing module 32 superimposes the inference parameter adjustment amount and the first tool parameter to obtain the second tool parameter.

[0077] In a possible implementation, the acquisition module 31 receives the user-defined tool parameters sent by the user for the machine tool cutting area; the processing module 32 controls the machine tool tool to dynamically match according to the user-defined tool parameters according to the user-defined tool parameters.

[0078] It should be noted that: the apparatus provided in the above embodiments, when realizing its functions, is only exemplified by the above division of functional modules, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above-described functions. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be described here.

[0079] The present application also provides an electronic device, referring to Figure 4 , Figure 4 The present application provides an electronic device. The structure of the electronic device is shown in the figure. The electronic device can include: at least one processor 41, at least one network interface 44, a user interface 43, a memory 45, and at least one communication bus 42.

[0080] Among them, the communication bus 42 is used to realize the connection communication between these components.

[0081] Among them, the user interface 43 can include a display screen (Display), a camera (Camera), and optionally the user interface 43 can also include a standard wired interface, a wireless interface.

[0082] Among them, the network interface 44 can optionally include a standard wired interface, a wireless interface (such as a Wi-Fi interface).

[0083] The processor 41 can include one or more processing cores. The processor 41 connects various parts within the server through various interfaces and lines, performs various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 45, and calling data stored in the memory 45. Alternatively, the processor 41 can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 41 can be integrated with a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes operating systems, user interfaces, and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; and the modem is used for processing wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 41, but can be realized by a separate chip.

[0084] The memory 45 can include a random access memory (RAM) and a read-only memory (ROM). Alternatively, the memory 45 includes a non-transitory computer-readable storage medium. The memory 45 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 45 can include a program storage area and a data storage area, wherein the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area can store data involved in the above-mentioned various method embodiments, etc. The memory 45 can also be at least one storage device located away from the aforementioned processor 41. As shown in the figure, the memory 45 as a computer storage medium can include an operating system, a network communication module, a user interface module, and an application program of a deep learning-based tool parameter dynamic matching method. Figure 4 As shown in the figure, the memory 45 as a computer storage medium can include an operating system, a network communication module, a user interface module, and an application program of a deep learning-based tool parameter dynamic matching method.

[0085] In Figure 4In the electronic device shown, the user interface 43 is mainly used to provide an interface for the user to input, and obtain data input by the user; and the processor 41 can be used to call an application program of a kind of tool parameter dynamic matching method based on deep learning stored in the memory 45, which, when executed by one or more processors, causes the electronic device to perform the method of one or more of the above embodiments.

[0086] It should be noted that, for the foregoing method embodiments, in order to simply describe, they are all described as a series of action combinations, but those skilled in the art should know that the present application is not limited to the order of the actions described, because according to the present application, certain steps can be performed in other order or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0087] The present application also provides a computer-readable storage medium, which stores instructions. When executed by one or more processors, the electronic device performs the method of one or more of the above embodiments.

[0088] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0089] In several embodiments provided in the present application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division, and actual implementation can have another division manner. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some service interface, device or unit, and can be electrical or other forms.

[0090] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e. they can be located in one place, or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0091] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0092] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable memory. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned memory includes: a U disk, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0093] The above is only exemplary embodiments of the present disclosure, which cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon considering the specification and practicing the true principles of the present disclosure. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure that follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed by the present disclosure. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for dynamic matching of tool parameters based on deep learning, characterized in that, The method includes: acquiring laser holographic imaging data and three-dimensional contour scanning data of the machine tool cutting area at a target time stamp; determining holographic voxel data and clipping point cloud data based on the laser holographic imaging data and the three-dimensional contour scanning data; inputting the holographic voxel data and the clipping point cloud data into a four-dimensional grid containing the target time stamp to obtain a four-dimensional snapshot sequence; acquiring first tool parameters of the machine tool at the target time stamp; and inputting the four-dimensional snapshot sequence and the first tool parameters into a preset deep learning causal supernet to generate second tool parameters, wherein the four-dimensional snapshot sequence and the first tool parameters are input into the preset deep learning causal supernet. The process of generating second tool parameters specifically includes: performing feature normalization and dimensionality reduction mapping on the four-dimensional snapshot sequence and the first tool parameters to obtain a spatiotemporal feature summary and tool parameter features; concatenating the spatiotemporal feature summary and the tool parameter features into a unified vector, and adding key node indicators marked in a preset causal graph to obtain a target analysis vector, wherein the key node indicators include material type and ambient temperature; using the preset deep learning causal supernet constructed from the preset causal graph, and combining the supernet weights to infer the target analysis vector to obtain an inference parameter adjustment amount; and superimposing the inference parameter adjustment amount and the first tool parameters to obtain the second tool parameters.

2. The deep learning-based dynamic tool parameter matching method according to claim 1, characterized in that, The acquisition of laser holographic imaging data and three-dimensional contour scanning data of the machine tool cutting area at a target timestamp specifically includes: sending synchronous acquisition commands to the laser holographic camera and the three-dimensional contour scanner respectively; reading the timestamps returned by the laser holographic camera and the three-dimensional contour scanner, and determining the timestamps as the target timestamp; under the constraint of the target timestamp, receiving the interference fringe sequence sent by the laser holographic camera, and receiving the point cloud batch sent by the three-dimensional contour scanner, to obtain the laser holographic imaging data and the three-dimensional contour scanning data.

3. The deep learning-based dynamic tool parameter matching method according to claim 1, characterized in that, The step of determining holographic voxel data and cropped point cloud data based on the laser holographic imaging data and the three-dimensional contour scanning data specifically includes: performing artifact removal and brightness equalization on the interference fringe pattern contained in the laser holographic imaging data to obtain an initial image; mapping the initial image into three-dimensional voxels according to a predefined spatial grid, each of the three-dimensional voxels including a corresponding local phase and local intensity; performing statistical filtering on the contour scanning point cloud contained in the three-dimensional contour scanning data to obtain an initial point cloud; determining surface normals on the initial point cloud to obtain an intention region; and converting the three-dimensional voxels and the intention region into unified coordinates and data structures respectively to obtain the holographic voxel data and the cropped point cloud data.

4. The deep learning-based dynamic tool parameter matching method according to claim 1, characterized in that, The step of inputting the holographic voxel data and the clipped point cloud data into a four-dimensional grid containing the target timestamp to obtain a four-dimensional snapshot sequence specifically includes: obtaining the engineering requirements of the machine tool cutting area; setting the spatial three-dimensional resolution and time step according to the engineering requirements; filling the holographic voxel data into the four-dimensional grid according to the three-dimensional resolution to obtain a first time slice; filling the clipped point cloud data into the four-dimensional grid according to the time step to obtain a second time slice; and sequentially stitching the first time slice and the second time slice in chronological order to obtain the four-dimensional snapshot sequence.

5. The deep learning-based dynamic tool parameter matching method according to claim 1, characterized in that, The acquisition of the first tool parameters of the machine tool at the target timestamp specifically includes: querying the cutting speed, feed rate, and depth of cut of the machine tool at the target timestamp from the CNC controller; synchronously acquiring the tool model and wear level of the machine tool; and determining the first tool parameters based on the cutting speed, feed rate, depth of cut, tool model, and wear level.

6. The deep learning-based dynamic tool parameter matching method according to claim 1, characterized in that, The method further includes: receiving custom tool parameters sent by a user for the cutting area of ​​the machine tool; and controlling the machine tool to dynamically match the custom tool parameters according to the custom tool parameters.

7. A tool parameter dynamic matching device based on deep learning, characterized in that, The device includes an acquisition module (31) and a processing module (32), wherein the acquisition module (31) is used to acquire laser holographic imaging data and three-dimensional contour scanning data of the machine tool cutting area at a target time stamp; the processing module (32) is used to determine holographic voxel data and clipping point cloud data based on the laser holographic imaging data and the three-dimensional contour scanning data; the processing module (32) is also used to input the holographic voxel data and the clipping point cloud data into a four-dimensional grid containing the target time stamp to obtain a four-dimensional snapshot sequence; the acquisition module (31) is also used to acquire a first tool parameter of the machine tool at the target time stamp; the processing module (32) is also used to input the four-dimensional snapshot sequence and the first tool parameter into a preset deep learning causal supernet to generate a second tool parameter, so that the device performs the method as described in any one of claims 1 to 6.

8. An electronic device, characterized in that, The electronic device includes a processor (41), a memory (45), a user interface (43), and a network interface (44). The memory (45) is used to store instructions. The user interface (43) and the network interface (44) are both used to communicate with other devices. The processor (41) is used to execute the instructions stored in the memory (45) to cause the electronic device to perform the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 6.

Images