System and method for dynamically adjusting rigidity of grooving machine based on material image recognition

Abstract

The invention discloses a grooving machine rigidity dynamic adjusting system and method based on material image recognition, and relates to the technical field of grooving machine rigidity dynamic adjusting, and the method comprises the following steps: obtaining a material surface and subsurface image sequence of a grooving machine processing area; inputting the image sequence into a pre-trained depth vision analysis model, and outputting a material characteristic parameter set in real time; converting the material characteristic parameter set into a target rigidity value based on a rigidity demand mapping model; within a preset regulation and control window period before the cutter makes contact with the target machining area, the initial rigidity adjustment amount is obtained according to the target rigidity value, and a rigidity adjustment instruction is generated and sent to a rigidity adjustment execution mechanism for dynamic adjustment; in the cutting process, fusion of feedback data and an image analysis result is obtained, and the initial rigidity adjustment amount is corrected; dynamic, accurate and predictive active adjustment of the rigidity of the grooving machine is achieved, and therefore the machining adaptability, intelligence and stability of the grooving machine are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of dynamic stiffness adjustment technology for grooving machines, and more specifically, to a dynamic stiffness adjustment system and method for grooving machines based on material image recognition. Background Technology

[0002] Grooving is a crucial step in mechanical manufacturing and material forming processes, and its processing quality and efficiency directly affect the structural performance and manufacturing cost of products. In practical industrial applications, when grooving machines process materials with different materials or structural properties, the proper matching of the system stiffness is of paramount importance for ensuring processing stability, suppressing tool chatter, and extending equipment life. Existing grooving equipment mostly adopts fixed stiffness structures or manually preset stiffness adjustment schemes, lacking the ability to sense and respond to the material state of the workpiece. This makes it difficult to achieve dynamic stiffness optimization for different material states, especially in complex working conditions involving uneven material hardness, strong structural heterogeneity, or microscopic defects, often leading to the following problems: Traditional systems rely on statically set stiffness parameters, failing to incorporate predictive adjustments based on actual tool entry conditions. This results in a delayed stiffness response at tool contact, easily triggering initial cutting impact and transient instability. Furthermore, existing methods often infer material states based on single sensors or mechanical feedback, failing to achieve coordinated identification of macroscopic texture features and microscopic defects, thus limiting the accuracy and specificity of stiffness adjustment. To address these issues, this invention proposes a solution. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a dynamic stiffness adjustment system and method for grooving machines based on material image recognition, which enables dynamic, precise, and predictive proactive adjustment of the stiffness of the grooving machine, thereby effectively improving the processing adaptability, intelligence, and stability of the grooving machine.

[0004] To achieve the above objectives, the present invention provides the following technical solution: This application provides a method for dynamic stiffness adjustment of a grooving machine based on material image recognition. The method includes: acquiring a sequence of images of the material surface and subsurface in the processing area of ​​the grooving machine; inputting the image sequence into a pre-trained deep vision analysis model and outputting a set of material characteristic parameters in real time; converting the set of material characteristic parameters into a target stiffness value based on a stiffness requirement mapping model; within a preset control window period before the tool contacts the target processing area, obtaining an initial stiffness adjustment amount based on the target stiffness value, and generating a stiffness adjustment command to send to the stiffness adjustment actuator for dynamic adjustment; during the cutting process, obtaining the fusion of feedback data and image analysis results to correct the initial stiffness adjustment amount.

[0005] In one embodiment, the image sequence is input into a pre-trained deep visual analysis model to output a set of material characteristic parameters in real time. Specifically, the image sequence is preprocessed to obtain a standardized input image; the input image is input into the first branch module of the deep visual analysis model to extract a macroscopic texture feature map; the input image is input into the second branch module of the deep visual analysis model to extract a microscopic defect feature map; the macroscopic texture feature map is input into an offset prediction module to generate an offset; the microscopic defect feature map is spatially aligned and calibrated according to the offset to obtain a calibrated microscopic feature map that is spatially aligned with the macroscopic texture feature map; and based on the fusion layer module, the calibrated microscopic feature map and the macroscopic texture feature map are concatenated along the channel dimension to obtain a fused feature map.

[0006] In one embodiment, the material characteristic parameter set is output in real time, specifically by: performing channel compression and feature reshaping on the fused feature map to obtain a material visual characteristic representation map; inputting the material visual characteristic representation map into a preset prediction module, performing corresponding task inference according to the target task type, and outputting the material characteristic parameter set.

[0007] In one embodiment, the deep visual analysis model is a multi-scale feature fusion network, including a first branch module, a second branch module, and a fusion layer module: the first branch module includes multiple dilated convolutional layers with different dilation rates stacked together; the second branch module includes a residual network structure composed of multiple residual units, and an attention mechanism module is embedded in each residual unit; the fusion layer module performs channel splicing of macroscopic texture features and microscopic defect features to output the material characteristic parameter set.

[0008] In one embodiment, based on the stiffness demand mapping model, the material characteristic parameter set is converted into a target stiffness value. Specifically, the material characteristic parameter set is cascaded to construct a joint feature vector; a multilayer perceptron network with residual connections is used to perform a nonlinear transformation on the joint feature vector to extract the final residual expression vector; an output regression layer is constructed based on the final residual expression vector, which is mapped to the target stiffness value, and the model is trained to obtain the stiffness demand mapping model; based on the stiffness demand mapping model, the input material characteristic parameter set is used for joint, residual perception, and regression calculations, and the predicted target stiffness value is output in real time.

[0009] In one embodiment, within a preset control window period before the tool contacts the target machining area, specifically: acquiring path planning data; obtaining the minimum distance between the tool and the target machining area based on the path planning data; and activating the preset stiffness adjustment control window period when the minimum distance is less than a set activation threshold.

[0010] In one embodiment, an initial stiffness adjustment amount is obtained based on the target stiffness value, and a stiffness adjustment command is generated and sent to the stiffness adjustment actuator for dynamic adjustment. Specifically, within the control window period, the system stiffness value, the target stiffness value, and the difference between them are acquired to obtain the initial stiffness adjustment amount, and a multi-level stiffness adjustment command sequence is generated. Within the control window period, the actuator gradually implements stiffness adjustment actions according to the stiffness adjustment command sequence, so that the system stiffness value dynamically transitions to the target stiffness value. The last time slice before the tool contacts the target machining area is acquired. Within the last time slice, the current stiffness value is collected and compared with the target stiffness value. If the error is within a preset tolerance range, the stiffness adjustment is confirmed to be complete, and the tool is allowed to contact the target machining area.

[0011] In one embodiment, during the cutting process, the initial stiffness adjustment amount is corrected by fusing feedback data and image analysis results. Specifically, this involves: acquiring key feature parameters characterizing the process state; normalizing the key feature parameters and constructing a unified cutting state perception vector using a multimodal feature fusion method based on principal component analysis and mutual information weight allocation; fusing the cutting state perception vector with the initial stiffness adjustment amount to obtain a stiffness offset correction amount under the current cutting state; superimposing the stiffness offset correction amount onto the initial stiffness adjustment amount to obtain a dynamically corrected second stiffness adjustment amount; constructing a stability leaf lobe diagram based on system operating parameters, substituting the second stiffness adjustment amount into the stability leaf lobe diagram for feasibility verification, obtaining the critical cutting depth under the current corrected stiffness, and comparing it with the current actual cutting depth; if the critical cutting depth is greater than the actual cutting depth, the corrected second stiffness adjustment amount is deemed valid; otherwise, the current second stiffness adjustment amount is deemed unsatisfactory; under the condition that the correction is verified to be feasible, the second stiffness adjustment amount is used as the final stiffness adjustment amount, a second stiffness adjustment command is generated and sent to the stiffness adjustment actuator.

[0012] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. By integrating multimodal high-resolution imaging and a deep visual analysis network, multi-scale visual features of materials are extracted from surface and subsurface image sequences. A dual-branch structure based on dilated convolution and attention enhancement is used to achieve collaborative modeling of macroscopic textures and microscopic defects. Deformable convolution is combined to calibrate feature space alignment, significantly improving the extraction accuracy and interpretability of material property parameters. Furthermore, by introducing a residual-aware multilayer perceptron to construct a stiffness demand mapping model, a high-precision, nonlinear mapping between material microscopic properties and processing stiffness is achieved. This effectively supports adaptive stiffness scheduling and dynamic control of processing equipment such as grooving machines, and has high real-time performance, scalability, and industrial practical value.

[0013] 2. By introducing a "preset adjustment window" mechanism, dynamic pre-adjustment of stiffness is initiated before the tool contacts the target machining area. Path planning enables advance prediction and response, ensuring that the stiffness is adjusted to the target value before critical machining, significantly improving machining stability and accuracy. Simultaneously, multi-source sensing (force, vibration, image) and deep learning models (LSTM) are integrated to achieve dynamic correction of stiffness deviation. The adaptability of stiffness adjustment under working conditions is verified by combining stability leaflet diagrams. A stiffness adjustment system with forward-looking and closed-loop adaptive characteristics is constructed, improving the system's real-time response capability to chatter and machining quality anomalies and the level of intelligence in the machining process. Attached Figure Description

[0014] Figure 1 This is a schematic flowchart of a method for dynamically adjusting the stiffness of a grooving machine based on material image recognition, provided in an embodiment of this application.

[0015] Figure 2 This is a schematic diagram of the dynamic stiffness adjustment system for a grooving machine based on material image recognition, provided in 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] In one exemplary embodiment, Figure 1 A flowchart illustrating the dynamic stiffness adjustment method for a grooving machine based on material image recognition, provided in this application embodiment, includes the following steps: S1, acquire the image sequence of the material surface and subsurface in the grooving machine processing area.

[0018] The material surface image sequence refers to a collection of images of the exposed surface of a material, continuously captured by high-resolution imaging equipment (such as industrial cameras) before or during material processing. This includes surface textures (rolling marks, machining marks, granular structure), surface defects (cracks, oxidation, spalling, corrosion), and surface structural heterogeneity (different grain regions, heat treatment traces). The material subsurface image sequence refers to a sequence of structural images within a depth of micrometers to millimeters below the material surface, indirectly obtained through imaging methods (including but not limited to structured light, polarization imaging, short-wave infrared, super-resolution reconstruction, etc.). This includes locally hardened zones, annealed residual layers, microcrack initiation zones, or crack tip propagation directions. Acquiring material surface and subsurface image sequences provides a high-dimensional, interpretable source of input features for stiffness control, processing parameter optimization, and intelligent prediction during processing.

[0019] S2, input the image sequence into the pre-trained deep visual analysis model, and output the material characteristic parameter set in real time.

[0020] Optionally, the deep visual analysis model can achieve end-to-end mapping from image pixels to material mechanical properties by fusing macroscopic texture and microscopic defect features.

[0021] The deep visual analysis model is a multi-scale feature fusion network, including a first branch module, a second branch module, and a fusion layer module: The first branch module includes multiple stacked dilated convolutional layers with different void ratios; Optionally, dilated convolutional layers with different dilation rates can be stacked. This can be understood as stacking multiple dilated convolutional layers layer by layer, with each layer having a different dilation rate ri (e.g., r1=1, r2=2, r3=4), in order to achieve texture feature extraction at different scales, enabling the model to perceive both local details and large-scale structures simultaneously. Here, r1, r2, and r3 represent the dilation rates of different dilated convolutional layers, respectively. The second branch module includes a residual network structure composed of multiple residual units, and an attention mechanism module is embedded in each residual unit to enhance the expressive ability of key microscopic defect regions in the image. The attention mechanism module is a joint mechanism of channel attention and spatial attention. The fusion layer module performs channel splicing of macroscopic texture features and microscopic defect features to output the material characteristic parameter set.

[0022] The image sequence is input into a pre-trained deep visual analysis model, which outputs a set of material characteristic parameters in real time, specifically: The image sequence is preprocessed to obtain a standardized input image; The preprocessing operations include size normalization, pixel channel normalization, and image enhancement processing, which are used to improve the model's adaptability and robustness to different material image sequences. The input image is fed into the first branch module of the deep visual analysis model to extract macroscopic texture feature maps; Among them, the macro-texture feature map refers to the feature tensor generated by extracting the large-scale texture distribution, structural periodicity and contextual organization information of the input image through stacking multiple dilated convolutional layers (with different dilation rates) in the first branch. It is mainly used to characterize the texture direction, arrangement pattern or organizational structure of materials at the macro scale. The input image is fed into the second branch module of the deep visual analysis model to extract microscopic defect feature maps; Among them, the micro-defect feature map refers to the feature tensor generated after extracting micro-targets such as local small defect regions, low-contrast foreign objects, and edge damage in the image through the residual network structure with an embedded attention mechanism (CBAM or SEblock) in the second branch. It emphasizes the ability to represent fine-grained regions and improves the sensitivity and accuracy of defect detection. The macroscopic texture feature map is input into the offset prediction module to generate an offset. The offset prediction module is composed of multiple 3×3 convolutions and nonlinear activation units, so that the offset can adaptively learn the geometric deviation between macroscopic texture and microscopic defects at the structural boundary. Based on the offset, the micro-defect feature map is spatially aligned and calibrated to obtain a calibrated micro-feature map that is spatially aligned with the macro-texture feature map. Specifically, the spatial alignment and calibration involves: The offset is used as a sampling position correction parameter and input to the deformable convolution module to perform convolution sampling on the micro-defect feature map, so that it is aligned with the macro-texture feature map in terms of feature geometry. Based on the fusion layer module, the calibrated microscopic feature map and the macroscopic texture feature map are spliced ​​along the channel dimension to obtain a fused feature map, which is used to characterize the comprehensive feature expression of the input image at both macroscopic and microscopic scales.

[0023] It should be noted that the nonlinear activation unit refers to a nonlinear function module introduced after each convolutional (or fully connected) operation. Its core function is to endow the neural network with the ability to fit nonlinear mappings, thereby enhancing the model's ability to express complex patterns (such as boundary changes, texture mutations, etc.). In the offset prediction module, the nonlinear activation unit typically consists of the following structure: after performing a linear convolution transformation on the input macroscopic texture feature map, it is input to the nonlinear activation function. In the middle, the output is obtained: Where W is the convolution kernel parameter, b is the bias term, Y is the offset, and the non-linear activation function refers to the LeakyReLU activation function.

[0024] Real-time output of material characteristic parameter set, specifically: The fused feature map is subjected to channel compression and feature reshaping using a group of convolution operators containing at least one 1×1 or 3×3 convolution kernel to achieve cross-scale information fusion and suppression of redundant features, resulting in a material visual characteristic representation map. Among them, the material visual characteristic representation map refers to the high-dimensional feature map with discriminative ability formed after multi-scale feature extraction, fusion and reshaping of the input image or image sequence through a deep visual analysis model. It is used to characterize the visual attributes of the material to be detected in terms of macroscopic texture structure, microscopic defect distribution, edge morphology and other aspects.

[0025] The visual characteristic representation of the material is input into a preset prediction module. Based on the target task type, the module performs corresponding task inference and outputs a set of material characteristic parameters. The task inference includes parameter estimation, classification and recognition, or defect detection operations. The set of material characteristic parameters includes the local hardness distribution of the material, defect categories, and material structural parameters. The prediction module includes at least one fully connected layer or convolutional prediction head. It constructs a network structure for performing task-related inference on the input feature map and outputs the corresponding set of material characteristic parameters based on the target task type.

[0026] The prediction module refers to a network submodule that, based on the fused feature map output by the proposed deep visual analysis model, performs numerical regression, classification, or spatial localization of specific physical / semantic attributes through a structured neural network computation path. Its structural design depends on the task attributes and can be a classifier, regressor, detector head, or semantic segmentation head, etc.

[0027] S3, based on the stiffness requirement mapping model, converts the set of material characteristic parameters into target stiffness values.

[0028] The stiffness requirement mapping model is a functional model that uses a set of material property parameters as input variables and predicts the required stiffness value through a mapping function or learning model. In grooving, different material regions have varying hardness and defect distributions. If a fixed stiffness parameter is used uniformly, problems such as tool runout and uneven machining can easily occur. The stiffness requirement mapping model can output the optimal clamping stiffness or structural stiffness required at a given point based on real-time or pre-detected material properties, providing feedback control to the actuator for adjustment, thus achieving a dynamic, precise, and low-loss machining strategy.

[0029] Based on the stiffness requirement mapping model, the set of material characteristic parameters is converted into target stiffness values, specifically as follows: The set of material characteristic parameters is cascaded to construct a joint feature vector; A multilayer perceptron network based on residual connections is used to perform a nonlinear transformation on the joint feature vector to extract the final residual representation vector, specifically: Configure an L-layer MLP, and define the output of each layer. Defined in residual form as: in, =0,1,...,L-1, For activation function, This is the weight matrix. Here, L represents the bias term of the MLP, and L is the total number of layers in the multilayer perceptron. For the first The input joint feature vector of the layer, For the first The output feature vector of the layer; The initial input is a joint feature vector, and the final residual representation vector after feature compression is obtained. ; An output regression layer is constructed based on the final residual representation vector, which is then mapped to the target stiffness value. The model is then trained to obtain the stiffness requirement mapping model. Based on the stiffness requirement mapping model, the input material characteristic parameter set is used for joint, residual perception and regression calculation, and the predicted target stiffness value is output in real time.

[0030] The target stiffness value is used to guide the optimal stiffness value for equipment processing or structural response. This stiffness value reflects the stiffness response capability that the grooving machine should have under the current material condition, so as to ensure the stability, accuracy and safety of the processing.

[0031] For example, suppose a workpiece surface has uneven hardness and local microcracks. If the equipment stiffness is configured too high or too low, it may cause tool runout, tearing of the machined surface, or local collapse. In this case, by inputting this microscopic information into the model, the predicted target stiffness value is 180 N / mm. The equipment can then adjust the stiffness of the clamping mechanism or support based on this value to achieve an adaptive response in the machining area and ensure forming quality.

[0032] The specific calculation formula for the stiffness requirement mapping model is as follows: in, For the target stiffness value, For the activation function of the regression layer, For the weights of the regression layer, For the regression layer bias, This is the final residual representation vector.

[0033] It should be noted that by introducing a deep nonlinear mapping network with a residual mechanism, the problems of gradient vanishing and feature degradation in traditional deep MLPs are effectively alleviated, ensuring the stable transmission and full extraction of feature representations in the deep network. Finally, the target stiffness value is output through a regression layer with Softplus activation, achieving a high-precision mapping between the microscopic properties of materials and macroscopic stiffness requirements. By applying the residual connection mechanism to the material stiffness modeling task, combined with nonlinear activation and structural regularization, the training stability and generalization performance of the model are effectively improved. The benefit is that it significantly improves the accuracy and robustness of stiffness prediction, providing reliable stiffness prediction support for the adaptive parameter scheduling of intelligent processing equipment such as slotting machines.

[0034] S4. Within a preset control window period before the tool contacts the target machining area, the initial stiffness adjustment amount is obtained based on the target stiffness value, and a stiffness adjustment command is generated and sent to the stiffness adjustment actuator for dynamic adjustment.

[0035] This method involves introducing a preset control window before the tool contacts the target machining area, predicting the tool position in real time based on path planning data, and initiating a stiffness adjustment process before approaching the target area to achieve dynamic pre-adjustment of the machining system stiffness. This ensures that the system stiffness is adjusted to the preset target value before the tool enters the critical machining area, thereby effectively suppressing structural vibrations during the cutting process, improving machining stability and accuracy, and ensuring the continuity and quality of the machining process.

[0036] Within the preset control window period before the tool contacts the target machining area, specifically: Obtain path planning data; The minimum distance between the tool and the target machining area is obtained based on the path planning data; When the minimum distance is less than the set activation threshold, the preset stiffness adjustment control window period is activated.

[0037] The path planning data defines the tool's motion trajectory, precisely describing the path the tool takes from its current state to the target machining area. This includes the three-dimensional coordinates the tool passes through sequentially during machining, the curve form of the cutting path (such as a spline curve or segmented line segments), feed rate, acceleration, and dynamic motion parameters such as the tool's current position. Based on this data, the system can calculate the minimum spatial distance between the tool's current position and the boundary of the target machining area.

[0038] The initial stiffness adjustment amount is obtained based on the target stiffness value, and a stiffness adjustment command is generated and sent to the stiffness adjustment actuator for dynamic adjustment, specifically: During the control window period, the system stiffness value and target stiffness value, as well as the difference between them, are obtained to obtain the initial stiffness adjustment amount. A multi-level stiffness adjustment instruction sequence is generated according to the preset adjustment strategy, wherein each level adjustment instruction includes the target stiffness setting value, the action time segment, and the execution priority identifier. Optionally, a multi-level stiffness adjustment command sequence is generated according to a preset adjustment strategy. This can be understood as discretizing the entire stiffness adjustment process into multiple control stages based on a predefined adjustment strategy (derived from process rules, model predictions, optimization control, or historical experience, etc.). Each stage forms a stiffness setting command, constituting a control sequence with time sequence and execution logic, and is issued to the stiffness adjustment actuator to gradually complete the stiffness transition. Among them, the initial stiffness adjustment amount refers to the numerical difference between the current actual stiffness value of the system and the preset target stiffness value. It is used to quantify the range of stiffness changes that the system needs to complete within the control window period. It is the direct driving force and control basis for stiffness adjustment behavior. The stiffness adjustment command sequence is sent to the stiffness adjustment actuator, which parses the commands one by one according to their timing and priority to prepare for the stiffness adjustment action. Among them, the stiffness adjustment actuator (such as magnetorheological damper, intelligent clamping device) completes the equivalent stiffness dynamic adjustment of at least one of the spindle system, guide rail system or tool clamping system before the tool cutting begins. After receiving the stiffness adjustment command sequence, it analyzes each command in sequence according to the timing and priority, and prepares to execute the corresponding physical structure adjustment action (such as preload adjustment, structural support stiffness change, etc.). The actuator can respond to and analyze priority commands in real time to realize multi-level dynamic adjustment of stiffness. During the control window period, the actuator gradually implements stiffness adjustment, causing the system stiffness value to dynamically transition towards the target stiffness value. This process has the following control characteristics: The response time control and smoothness optimization are as follows: the response time control is to complete each adjustment step within the constraint response time to avoid excitation overshoot; the smoothness optimization is to make stiffness changes continuous and smooth through interpolation functions or S-curve planning to avoid system vibration excitation. The system acquires the last time slice before the tool is about to contact the target machining area. During this last time slice, the system collects the current stiffness value and compares it with the target stiffness value. If the error is within the preset tolerance range, the system confirms that the stiffness adjustment is complete and allows the tool to contact the target area.

[0039] If the error is within the preset tolerance range, it can be understood that the difference between the two is greater than the preset minimum tolerance value and less than the preset maximum tolerance value.

[0040] S5, during the cutting process, integrates the feedback data with the image analysis results to correct the initial stiffness adjustment amount.

[0041] Among them, by correcting the initial stiffness adjustment amount, high-precision dynamic closed-loop control of stiffness adjustment during the cutting process is realized. It can sense changes in the processing state in real time and actively correct the stiffness adjustment strategy, which significantly improves the system's response to chatter and surface quality abnormalities.

[0042] During the cutting process, the feedback data and image analysis results are fused to correct the initial stiffness adjustment amount, specifically as follows: A multi-source sensor data acquisition model for a cutting machining system is established. This model acquires key characteristic parameters representing the process state. The model includes: a three-dimensional force sensor for real-time acquisition of cutting forces; a three-axis vibration sensor for capturing spindle and workpiece vibration characteristics; and an industrial image acquisition system for monitoring the surface morphology of the cutting area. These three types of data are acquired synchronously and time-aligned to construct a unified multi-source feature input sequence in the time domain. The key characteristic parameters include the mean cutting force, pulsation amplitude, and frequency extracted from the force sensor data; the dominant frequency mode, amplitude envelope, and order response extracted from the vibration sensor data; and workpiece surface roughness texture features (which can indirectly reflect the influence of machining parameters on surface quality and identify the presence of scratches, burrs, etc.) and boundary contour integrity indicators (used to identify machining contour anomalies caused by vibration, tool offset, or workpiece stiffness instability) extracted from the image data using a deep convolutional neural network. The key feature parameters are normalized, and a unified cutting state perception vector is constructed by using a multimodal feature fusion method based on principal component analysis and mutual information weight allocation. The cutting state perception vector is fused with the initial stiffness adjustment amount and input into the preset stiffness offset prediction model to obtain the stiffness offset correction amount under the current cutting state. The stiffness offset prediction model is constructed based on LSTM (Long Short-Term Memory Neural Network). The specific formula for calculating the stiffness offset correction is as follows: in, This is the stiffness offset correction amount. This is the cutting state sensing vector. This is the initial stiffness adjustment amount. It is a long short-term memory network module. As the output layer of the multilayer perceptron, the LSTM output is nonlinearly mapped to generate the final stiffness correction.

[0043] The stiffness offset correction is added to the initial stiffness adjustment to obtain the second stiffness adjustment after dynamic correction. Optionally, the stiffness offset correction amount can be added to the initial stiffness adjustment amount, which can be understood as the sum of the stiffness offset correction amount and the initial stiffness adjustment amount. A stability lobe diagram is constructed based on system operating parameters, including tool speed, axial depth of cut, and material damping ratio. The stability lobe diagram is used to characterize the system chatter stability boundary under different stiffnesses and generates a mapping function: SLD:(K,ω)→a, where K is the second stiffness adjustment amount, ω is the main spindle speed, and a is the critical depth of cut. The feasibility of the second stiffness adjustment is verified by substituting it into the stability lobe diagram. Based on the step mapping function, the critical cutting depth under the current modified stiffness is obtained and compared with the current actual cutting depth. If the critical cutting depth is greater than the actual cutting depth, the stiffness adjustment is deemed effective. Otherwise, the current second stiffness adjustment is determined to not meet the dynamic stability requirements; Under the condition that the correction is feasible, the second stiffness adjustment amount is taken as the final stiffness adjustment amount, a second stiffness adjustment command is generated and sent to the stiffness adjustment actuator to complete the dynamic closed-loop control of stiffness during the cutting process.

[0044] Example 2, Figure 2 This is a schematic diagram of the dynamic stiffness adjustment system for a grooving machine based on material image recognition, provided in an embodiment of this application. It includes an image acquisition module, an image parsing module, a target stiffness value generation module, a stiffness adjustment module, and a stiffness adjustment correction module, with connections between the modules. The image acquisition module is used to acquire image sequences of the material surface and subsurface in the grooving machine processing area; The image parsing module is used to input the image sequence into a pre-trained deep visual parsing model and output a set of material characteristic parameters in real time. The target stiffness value generation module is used to convert the set of material characteristic parameters into target stiffness values ​​based on the stiffness requirement mapping model. The stiffness adjustment module is used to obtain the initial stiffness adjustment amount based on the target stiffness value within a preset control window period before the tool contacts the target machining area, and generate a stiffness adjustment command to send to the stiffness adjustment actuator for dynamic adjustment. The stiffness adjustment and correction module is used to fuse feedback data and image analysis results during the cutting process to correct the initial stiffness adjustment amount.

[0045] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0046] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0047] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0048] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0049] 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 method for dynamic adjustment of the stiffness of a grooving machine based on material image recognition, characterized in that, Includes the following steps: Acquire image sequences of the material surface and subsurface in the grooving machine processing area; The image sequence is input into a pre-trained deep visual analysis model, which outputs a set of material characteristic parameters in real time. Based on the stiffness requirement mapping model, the set of material characteristic parameters is converted into target stiffness values; Within a preset control window period before the cutting tool contacts the target machining area, the initial stiffness adjustment amount is obtained based on the target stiffness value, and a stiffness adjustment command is generated and sent to the stiffness adjustment actuator for dynamic adjustment. During the cutting process, the feedback data and image analysis results are fused together to correct the initial stiffness adjustment.

2. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 1, characterized in that, The image sequence is input into a pre-trained deep visual analysis model, which outputs a set of material characteristic parameters in real time, specifically: The image sequence is preprocessed to obtain a standardized input image; The input image is fed into the first branch module of the deep visual analysis model to extract macroscopic texture feature maps; The input image is fed into the second branch module of the deep visual analysis model to extract microscopic defect feature maps; The macroscopic texture feature map is input into the offset prediction module to generate the offset; Based on the offset, the micro-defect feature map is spatially aligned and calibrated to obtain a calibrated micro-feature map that is spatially aligned with the macro-texture feature map. Based on the fusion layer module, the calibrated microscopic feature map and the macroscopic texture feature map are spliced ​​together along the channel dimension to obtain the fused feature map.

3. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 2, characterized in that, The real-time output set of material characteristic parameters is specifically as follows: The fused feature map is subjected to channel compression and feature reshaping to obtain a visual characteristic representation map of the material. The material visual characteristic representation map is input into a preset prediction module, and the corresponding task reasoning is performed according to the target task type to output a set of material characteristic parameters.

4. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 3, characterized in that, The deep visual analysis model is a multi-scale feature fusion network, including a first branch module, a second branch module, and a fusion layer module: The first branch module includes multiple stacked dilated convolutional layers with different void ratios; The second branch module includes a residual network structure composed of multiple residual units, and an attention mechanism module is embedded in each residual unit; The fusion layer module performs channel splicing of macroscopic texture features and microscopic defect features to output the material characteristic parameter set.

5. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 1, characterized in that, The stiffness requirement mapping model converts the set of material characteristic parameters into target stiffness values, specifically as follows: The material property parameter set is cascaded to construct a joint feature vector; A multilayer perceptron network with residual connections is used to perform a nonlinear transformation on the joint feature vector to extract the final residual representation vector. An output regression layer is constructed based on the final residual expression vector, which is then mapped to the target stiffness value. The model is then trained to obtain the stiffness requirement mapping model. Based on the stiffness requirement mapping model, the input material characteristic parameter set is used for joint, residual perception and regression calculation, and the predicted target stiffness value is output in real time.

6. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 1, characterized in that, The preset control window period before the tool contacts the target machining area specifically includes: Obtain path planning data; The minimum distance between the tool and the target machining area is obtained based on the path planning data; When the minimum distance is less than the set activation threshold, the preset stiffness adjustment control window period is activated.

7. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 6, characterized in that, The step of obtaining an initial stiffness adjustment amount based on the target stiffness value and generating a stiffness adjustment command to be sent to the stiffness adjustment actuator for dynamic adjustment specifically involves: During the control window period, the system stiffness value and target stiffness value, as well as the difference between them, are obtained to obtain the initial stiffness adjustment amount and generate a multi-level stiffness adjustment command sequence. During the control window period, the actuator gradually implements stiffness adjustment actions according to the stiffness adjustment command sequence, so that the system stiffness value dynamically transitions to the target stiffness value; The last time slice before the tool is about to contact the target machining area is obtained. During the last time slice, the current stiffness value is collected and compared with the target stiffness value. If the error is within the preset tolerance range, the stiffness adjustment is confirmed to be complete, and the tool is allowed to contact the target machining area.

8. The method for dynamic adjustment of grooving machine stiffness based on material image recognition according to claim 7, characterized in that, The process of fusing feedback data and image analysis results during cutting to correct the initial stiffness adjustment amount specifically involves: Obtain key characteristic parameters that characterize the process state; The key feature parameters are normalized, and a unified cutting state perception vector is constructed by using a multimodal feature fusion method based on principal component analysis and mutual information weight allocation. The cutting state perception vector is fused with the initial stiffness adjustment amount to obtain the stiffness offset correction amount under the current cutting state; The stiffness offset correction is added to the initial stiffness adjustment to obtain the second stiffness adjustment after dynamic correction. A stability lobe diagram is constructed based on system operating parameters. The second stiffness adjustment is substituted into the stability lobe diagram to verify its feasibility. The critical cutting depth under the current modified stiffness is obtained and compared with the current actual cutting depth. If the critical cutting depth is greater than the actual cutting depth, then the second stiffness adjustment amount obtained by the correction is deemed valid; Otherwise, the current second stiffness adjustment amount is determined to be insufficient; If the correction is verified to be feasible, the second stiffness adjustment amount is taken as the final stiffness adjustment amount, a second stiffness adjustment command is generated and sent to the stiffness adjustment actuator.

9. A system using the material image recognition-based dynamic stiffness adjustment method for a grooving machine as described in any one of claims 1-8, characterized in that, It includes a graphics acquisition module, a graphics parsing module, a target stiffness value generation module, a stiffness adjustment module, and a stiffness adjustment correction module. These modules are interconnected. The image acquisition module is used to acquire image sequences of the material surface and subsurface in the grooving machine processing area; The image parsing module is used to input the image sequence into a pre-trained deep visual parsing model and output a set of material characteristic parameters in real time. The target stiffness value generation module is used to convert the set of material characteristic parameters into target stiffness values ​​based on the stiffness requirement mapping model. The stiffness adjustment module is used to obtain the initial stiffness adjustment amount based on the target stiffness value within a preset control window period before the tool contacts the target machining area, and generate a stiffness adjustment command to send to the stiffness adjustment actuator for dynamic adjustment. The stiffness adjustment and correction module is used to fuse feedback data and image analysis results during the cutting process to correct the initial stiffness adjustment amount.

Images