Self-adaptive control system for grinding force of carbon fiber plate

Abstract

The invention discloses a carbon fiber plate grinding force self-adaptive control system which comprises the following components: S1, a grinding execution module which is used for grinding a carbon fiber plate; s2, a multi-modal data acquisition module is used for acquiring multi-modal data in the grinding process in real time through a sensor; s3, a multi-modal data fusion module processes the collected data to generate a high-dimensional processing state representation vector; s4, based on a dynamic modeling module of self-supervised learning, a processing state feature mapping model is constructed by adopting a comparative learning technology, the processing characteristics of the carbon fiber plate are captured, and optimal grinding parameters are dynamically predicted; s5, the real-time self-adaptive control module adjusts machining parameters of the grinding execution module in real time according to the optimal grinding parameters output by the dynamic modeling module; and S6, the closed-loop feedback module collects machining state data changing in real time and feeds back the machining state data to the data fusion module and the dynamic modeling module. The method has the advantages of being high in dynamic adaptability, high in machining precision and fast in real-time response.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon fiber sheet processing, and particularly to an adaptive control system for the grinding force of carbon fiber sheets. Background Art

[0002] With the wide application of high-performance materials in fields such as aerospace and automotive manufacturing, carbon fiber sheets have become important structural materials due to their lightweight and high-strength characteristics. However, during the processing of carbon fiber sheets, due to the anisotropic and high-hardness characteristics of the materials, it is difficult to accurately control the grinding force, which seriously affects the processing quality and efficiency. As a key control parameter during the processing, the instability of the grinding force may lead to damage to the processed surface, increased tool wear, and even equipment failures.

[0003] In the prior art, traditional methods for controlling the grinding force of carbon fiber sheets mainly rely on static control strategies based on fixed rules and empirical control methods adjusted manually. These methods have the following defects when facing complex and dynamically changing processing environments: 1. Lack of dynamic adaptability: Traditional control methods cannot dynamically adjust the grinding parameters according to the real-time changes in the state during the processing of carbon fiber sheets, resulting in difficulty in stabilizing the grinding force within the target range.

[0004] 2. Inefficient data utilization: Existing methods fail to fully fuse and analyze multi-modal data (such as grinding force, vibration signals, temperature, and acoustic emission signals) during the processing, and cannot extract the deep feature relationships of the processing state from complex data.

[0005] 3. Limitations of the model: Existing grinding force control methods are mostly based on simple linear models and cannot effectively capture the non-linear characteristics and multi-variable dynamic relationships exhibited by carbon fiber sheets during processing.

[0006] 4. Lack of real-time performance: Traditional control methods have a slow response speed to changes in the processing state. Especially when the processing conditions change rapidly, it is difficult to achieve rapid adjustment of the grinding parameters.

[0007] Therefore, how to provide an adaptive control system for the grinding force of carbon fiber sheets is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0008] An object of the present invention is to propose an adaptive control system for the grinding force of carbon fiber sheets based on self-supervised learning, which makes full use of multi-modal data fusion technology, contrastive learning algorithms, and real-time dynamic modeling methods, and details the technical solution for accurately controlling the grinding force of carbon fiber sheets by dynamically optimizing the grinding parameters, and has the advantages of strong dynamic adaptability, high processing accuracy, and fast real-time response.

[0009] A self - adaptive control system for grinding force of carbon fiber plates according to an embodiment of the present invention includes: S1. A grinding execution module for grinding the carbon fiber plates. The grinding execution module includes a variable - speed grinding wheel, a motor drive unit, a machining depth control unit, and a force - sensing unit connected to the grinding device; S2. A multi - modal data acquisition module that obtains multi - modal data during the machining process in real - time through sensors. The sensors include a vibration sensor, a temperature sensor, and an acoustic emission sensor; S3. A multi - modal data fusion module that dynamically fuses the multi - modal data to generate a high - dimensional machining state characterization vector; S4. A dynamic modeling module based on self - supervised learning that uses contrastive learning technology to construct a machining state feature mapping model through the high - dimensional machining state characterization vector, captures the anisotropy and high - hardness machining characteristics of the carbon fiber plates, and dynamically predicts the optimal grinding parameters; S5. A real - time self - adaptive control module that adjusts the machining parameters of the grinding execution module in real - time according to the optimal grinding parameters output by the dynamic modeling module, so that the grinding force is always within the target range; S6. A closed - loop feedback module that collects the multi - modal data that changes in real - time during the machining process and feeds it back to the multi - modal data fusion module and the dynamic modeling module to form a dynamic data cycle.

[0010] Optionally, the S1 specifically includes: S11. A variable - speed grinding wheel for machining the carbon fiber plates by dynamically adjusting the linear velocity of the grinding wheel during grinding, where the adjustment of the linear velocity is based on the real - time force signal collected by the force - sensing unit; S12. The motor drive unit is connected to the variable - speed grinding wheel and is used to provide the power for driving the grinding operation. The motor drive unit supports a speed - regulation function based on variable - frequency control, and its speed range is , where represents the instantaneous speed provided by the motor drive unit, represents the lowest speed allowed by the motor drive unit, represents the highest speed allowed by the motor drive unit; S13. The machining depth control unit is linked with the motor drive unit and is used to control the machining depth between the variable - speed grinding wheel and the surface of the carbon fiber plates. The machining depth control unit includes a mechanical feed mechanism and a position feedback device, and makes the machining depth dynamically adjusted within the set range by collecting the contact displacement signal in real - time, where and respectively represent the minimum and maximum values of the machining depth; S14, a force sensing unit, is installed between the grinding device and the workpiece to be machined, and is used to monitor the grinding force generated during the grinding process in real time. Among them, the grinding force includes the normal force and the tangential force .

[0011] Optionally, the S3 specifically includes: S31, a data receiving unit, receives the multimodal data collected by the multimodal data acquisition module, including the normal force collected by the force sensing unit and the tangential force , the vibration signal collected by the vibration sensor , the temperature signal collected by the temperature sensor and the acoustic emission signal collected by the acoustic emission sensor , and performs preliminary dynamic fusion on the multimodal data to generate a preliminary fusion result ; S32, a data alignment unit, performs alignment processing on the data in the preliminary fusion result in the time dimension, so that the timestamps of the data items in the preliminary fusion result are consistent, and generates a time-aligned data set ; S33, a data denoising unit, uses an adaptive filtering algorithm to perform denoising processing on the time-aligned data set and performs residual error minimization optimization: ; Among them, represents the value of the time-aligned data set at time , is the weight parameter of the filter, is the sampling length, is the lag order, and by iteratively optimizing the value of the weight parameter of the filter, the noise signal is eliminated, and finally an effective data set after denoising is generated ; S34, a data normalization unit, performs normalization processing on the effective data set after denoising to generate a normalized data set ; S35, a state characterization generation unit, constructs a high-dimensional machining state characterization vector based on the normalized data set , and the high-dimensional machining state characterization vector is defined as: ; Among them, Represents a feature mapping function extracted from the corresponding multimodal data.

[0012] Optionally, S4 specifically includes: S41. A feature data preprocessing unit for screening the high-dimensional processed state characterization vector generated by the multimodal data fusion module to extract key feature components and construct a feature input matrix : ; Wherein, represents the th eigenvalue in the th processing state sample, is the number of samples, is the feature dimension; S42. A comparison sample generation unit that generates training samples for contrastive learning from the feature input matrix ; S43. A feature mapping model construction unit that constructs a processing state feature mapping model based on the contrastive learning algorithm and optimizes the contrastive loss function: ; Wherein, represents the contrastive loss function, and are positive sample pairs in the training samples, is a negative sample in the training samples, represents the similarity between samples, is the temperature coefficient, M is the total number of samples, and a feature mapping model capable of capturing the characteristics of the processing state is generated; S44. A dynamic modeling unit that combines the feature mapping model and the high-dimensional processed state characterization vector to dynamically model the processing state of carbon fiber sheets, and generate a processing state dynamic prediction model by identifying non-linear characteristics and multivariable dynamic relationships; S45. An optimal parameter calculation unit that uses the processing state dynamic prediction model to calculate the target grinding parameter set , where is the wheel linear speed, is the feed speed, is the depth of cut.

[0013] Optionally, S42 specifically includes: S421. A dynamic processing parameter sampling unit that dynamically adjusts the target grinding parameter set according to the real-time processing state, and calculates the time series change rate of the high-dimensional processed state characterization vector : ; wherein, is the sampling time interval; S422. An intelligent sample generation unit constructs positive and negative sample pairs from the dynamically adjusted target grinding parameter set ; The positive sample pairs are generated based on the similarity of the machining state feature vectors: ; wherein, is the similarity threshold of the positive sample pairs, represents the similarity function between feature vectors, represents the machining parameter combination corresponding to the feature vector ; represents the machining parameter combination corresponding to the feature vector ; The negative sample pairs are generated based on the difference of the machining state feature vectors: ; wherein, is the similarity threshold of the negative sample pairs, represents the machining parameter combination corresponding to the feature vector ; S423. A multi-dimensional feature weight adjustment unit dynamically assigns weights based on the feature distribution of the sample pairs: ; wherein, is the weight of the positive sample or the negative sample pair , is the adjustment factor of the weight distribution, is the square of the Euclidean distance between samples; S424. By optimizing the positive and negative sample ratio and combining the sample pair distribution after weight adjustment, the final training samples are generated.

[0014] Optionally, the S44 specifically includes: S441. A dynamic feature relationship analysis unit mines the dynamic relationship between machining state features through non-linear regression analysis based on the feature mapping model and the high-dimensional machining state representation vector : ; wherein, is the dynamic characteristic mapping function of the machining state, is the weight matrix, is the bias vector, represents the feature activation function; S442. Multivariable characteristic recognition unit, combined with the dynamic characteristic mapping function and the target grinding parameter set , captures the anisotropic and high-hardness characteristics of carbon fiber sheets through multivariable relationship modeling: ; Among them, is the current machining state feature vector, and are the state transition matrix and the parameter influence matrix respectively, is the random disturbance term, represents time; S443. Machining state prediction unit, combined with the real-time input machining state characterization vector and the target grinding parameter set , generates a dynamic prediction model for estimating the machining state at future moments: ; Among them, is the predicted machining state vector.

[0015] Optionally, the S45 specifically includes: S451. Parameter range generation unit, according to the predicted machining state vector and the set target machining state, generates the target parameter range through dynamic analysis; Determine the target machining state based on feature similarity: ; Among them, is the target machining state feature vector, is the similarity threshold; Dynamically adjust the target parameter range based on the temporal variation of the machining state : ; Among them, is the initial parameter range, is the parameter increment, represents the sensitivity of the machining state to the parameters; S452. Construct the objective function based on the multi-objective optimization principle of machining efficiency and machining stability : ; Among them, is the objective function, represents the normalized error between the machining state prediction and the target state, represents the machining time, and are the weight factors for machining stability and efficiency respectively, and the target grinding parameter set is the variable to be optimized; S453. The adaptive optimization unit, based on the objective function and the target parameter range , adopts a hybrid optimization algorithm, including a combination of local gradient search and global heuristic search, and iteratively calculates the target grinding parameter set , achieving fast convergence through local gradient search and avoiding being trapped in local optima through global heuristic search: .

[0016] Optionally, the S5 specifically includes: S51. The machining parameter receiving unit receives the target grinding parameter set output by the dynamic modeling module ; S52. The real-time force monitoring unit real-time monitors the normal force and the tangential force generated during the grinding process, and transmits the collected grinding force to the grinding force adjustment unit; S53. The grinding force adjustment unit dynamically calculates the grinding force deviation based on the preset target grinding force range and the real-time monitored grinding force : ; Adjust the grinding parameters according to the grinding force deviation: ; wherein, is the grinding force adjustment gain matrix, represents the adjusted machining parameter set; S54. The machining parameter execution unit, according to the adjusted machining parameter set , real-time controls the grinding execution module, including dynamically adjusting the wheel linear speed , the feed speed and the machining depth , so that the machining process conforms to the target range.

[0017] The beneficial effects of the present invention are: (1) By combining multi-modal data fusion, self-supervised learning, and dynamic modeling techniques, the present invention achieves a deep understanding and dynamic adaptation of key state parameters during the processing of carbon fiber sheets. By fusing multi-modal data such as grinding force, vibration signal, temperature, and acoustic emission signal, and using a contrastive learning algorithm to construct a processing state feature mapping model, it can accurately capture the anisotropy and high hardness characteristics of carbon fiber sheets, thereby dynamically predicting the optimal grinding parameters and effectively coping with complex processing environments.

[0018] (2) Through the real-time adaptive control module combined with a closed-loop feedback mechanism, the present invention realizes the real-time optimization and adjustment of the processing parameters of the grinding execution module, ensuring that the grinding force is always stable within the target range. Compared with traditional rule-based static control methods, the present invention can quickly respond to changes in the processing state, significantly improving the adaptability of the system to dynamic environments and processing efficiency.

[0019] (3) By introducing a hybrid optimization algorithm and a multi-objective optimization strategy, the present invention effectively balances processing efficiency and stability in dynamic modeling and parameter optimization, reducing processing errors and improving grinding quality. At the same time, through the time-series analysis and dynamic modeling of the processing state, the dependence on manual intervention is significantly reduced, enhancing the intelligence and automation level of the processing process. Brief Description of the Drawings

[0020] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1 is an overall framework diagram of an adaptive control system for the grinding force of a carbon fiber sheet proposed by the present invention. Detailed Embodiments

[0021] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0022] Refer to Figure 1 , an adaptive control system for the grinding force of a carbon fiber sheet, comprising: S1. A grinding execution module for performing grinding operations on carbon fiber sheets. The grinding execution module includes a variable-speed grinding wheel, a motor drive unit, a processing depth control unit, and a force sensing unit connected to the grinding device; In this embodiment, S1 specifically includes: S11. A variable-speed grinding wheel for processing carbon fiber sheets by dynamically adjusting the linear speed of the grinding wheel during grinding , where the adjustment of the linear speed is based on the real-time force signal collected by the force sensing unit; S12. The motor drive unit is connected to the adjustable-speed grinding wheel and is used to provide the power for driving the grinding operation. The motor drive unit supports a speed regulation function based on variable-frequency control, and its speed range is , where represents the instantaneous speed provided by the motor drive unit, represents the minimum speed allowed by the motor drive unit, represents the maximum speed allowed by the motor drive unit; S13. The machining depth control unit is linked with the motor drive unit and is used to control the machining depth between the adjustable-speed grinding wheel and the surface of the carbon fiber sheet , and the machining depth control unit includes a mechanical feed mechanism and a position feedback device. By collecting the contact displacement signal in real time, the machining depth is dynamically adjusted within the set range , where and respectively represent the minimum value and the maximum value of the machining depth; S14. The force sensing unit is installed between the grinding device and the workpiece to be machined and is used to monitor the grinding force generated during the grinding process in real time. Among them, the grinding force includes the normal force and the tangential force .

[0023] In this embodiment, through the combined design of the grinding execution module, the adjustable-speed grinding wheel, the motor drive unit, the machining depth control unit and the force sensing unit are organically combined to achieve precise control of the machining of carbon fiber sheets. The dynamic adjustment of the linear speed is realized through the adjustable-speed grinding wheel, the stable power is provided by combining the variable-frequency control of the motor drive unit, and at the same time, the contact depth between the grinding wheel and the workpiece is dynamically adjusted by using the machining depth control unit, and the changes of the normal force and the tangential force of the grinding force are monitored in real time in cooperation with the force sensing unit.

[0024] S2. The multi-modal data acquisition module obtains multi-modal data during the machining process in real time through sensors. The sensors include vibration sensors, temperature sensors, and acoustic emission sensors; In this embodiment, S2 specifically includes: The multi-modal data acquisition module collects multi-modal data during the machining process in real time through various sensors installed on the grinding device, including the normal force and tangential force data collected by the force sensing unit, the vibration signal collected by the vibration sensor, the machining area temperature change data collected by the temperature sensor, and the acoustic emission signal collected by the acoustic emission sensor.

[0025] This embodiment provides comprehensive monitoring of the carbon fiber sheet processing process by real-time collecting multi-modal data such as grinding force, vibration signal, temperature, and acoustic emission signal, covering not only the current processing state characteristics but also capturing the dynamic change relationships among different modal data. Through the fusion and analysis of multi-modal data, abnormal states and key characteristics in the processing process can be identified more accurately, providing a high-quality data basis for subsequent dynamic modeling and parameter optimization, and significantly improving the system's perception ability of complex processing environments and data utilization efficiency.

[0026] S3. A multi-modal data fusion module that dynamically fuses multi-modal data to generate a high-dimensional processing state characterization vector; In this embodiment, S3 specifically includes: S31. A data receiving unit that receives the multi-modal data collected by the multi-modal data acquisition module, including the normal force collected by the force sensing unit and the tangential force , the vibration signal collected by the vibration sensor , the temperature signal collected by the temperature sensor and the acoustic emission signal collected by the acoustic emission sensor , and performs preliminary dynamic fusion on the multi-modal data to generate a preliminary fusion result ; S32. A data alignment unit that performs alignment processing on the data in the preliminary fusion result in the time dimension, so that the timestamps of the data items in the preliminary fusion result are consistent, generating a time-aligned data set ; S33. A data denoising unit that uses an adaptive filtering algorithm to denoise the time-aligned data set , and performs residual error minimization optimization: ; where represents the value of the time-aligned data set at time , is the weight parameter of the filter, is the sampling length, is the lag order, and by iteratively optimizing the value of the filter weight parameter , the noise signal is eliminated, and finally an effective data set after denoising is generated; S34. A data standardization unit that standardizes the effective data set after denoising to generate a standardized data set ; S35. The state representation generation unit constructs a high-dimensional machining state representation vector based on the standardized data set. The high-dimensional machining state representation vector is defined as: ; ; where represents the feature mapping function extracted from the corresponding multimodal data.

[0027] In this embodiment, through the multimodal data fusion module, various data sources such as grinding force, vibration signal, temperature signal, and acoustic emission signal are dynamically fused, aligned, denoised, and standardized to generate a high-dimensional machining state representation vector. This not only comprehensively integrates multi-dimensional machining state information but also captures complex non-linear features and dynamic correlations between multimodals during the machining process, significantly improving the accuracy and usability of machining state data, providing high-quality data support for subsequent dynamic modeling and parameter optimization, and enhancing the system's understanding and adaptation ability to complex machining states.

[0028] S4. The dynamic modeling module based on self-supervised learning uses contrastive learning technology to construct a machining state feature mapping model through the high-dimensional machining state representation vector, capturing the anisotropy and high-hardness machining characteristics of carbon fiber sheets and dynamically predicting the optimal grinding parameters. In this embodiment, S4 specifically includes: S41. The feature data preprocessing unit is used to screen the high-dimensional machining state representation vector generated by the multimodal data fusion module, extract key feature components, and construct a feature input matrix: ; where represents the ; th eigenvalue in the th machining state sample, is the sample number, and is the feature dimension; is the feature dimension; S42. The contrastive sample generation unit generates training samples for contrastive learning from the feature input matrix. ; S43. The feature mapping model construction unit constructs a machining state feature mapping model based on the contrastive learning algorithm and optimizes the contrastive loss function: ; where represents the contrastive loss function, and are positive sample pairs in the training samples, and is the negative sample in the training samples. represents the similarity between samples is the temperature coefficient, M is the total number of samples, and a feature mapping model capable of capturing the characteristics of the machining state is generated; S44, the dynamic modeling unit, combines the feature mapping model and the high-dimensional machining state characterization vector , dynamically models the machining state of carbon fiber sheets, and generates a dynamic prediction model of the machining state by identifying non-linear characteristics and multi-variable dynamic relationships; S45, the optimal parameter calculation unit, calculates the target grinding parameter set using the dynamic prediction model of the machining state , where is the wheel linear speed, is the feed rate, is the depth of cut.

[0029] S421, the dynamic machining parameter sampling unit, dynamically adjusts the target grinding parameter set according to the real-time machining state , calculates the time sequence change rate of the high-dimensional machining state characterization vector : : ; where is the sampling time interval; S422, the intelligent sample generation unit, constructs positive sample pairs and negative sample pairs from the dynamically adjusted target grinding parameter set ; The positive sample pairs are generated based on the similarity of the machining state feature vectors: ; where is the similarity threshold of the positive sample pairs, represents the similarity function between feature vectors, represents the feature vector corresponding machining parameter combination, represents the feature vector corresponding machining parameter combination; The negative sample pairs are generated based on the difference of the machining state feature vectors: ; where is the similarity threshold of the negative sample pairs, represents the feature vector corresponding machining parameter combination; S423, the multi-dimensional feature weight adjustment unit, dynamically assigns weights based on the feature distribution of the sample pairs: ; where Is a positive sample Or a negative sample pair Of the weight, Is the adjustment factor of the weight distribution, Is the square of the Euclidean distance between samples; S424. By optimizing the positive and negative sample ratio and combining the sample pair distribution after weight adjustment, generate the final training samples.

[0030] S441. The dynamic feature relationship analysis unit, based on the feature mapping model and the high-dimensional processing state representation vector , discovers the dynamic relationships between processing state features through non-linear regression analysis: ; Among them, Is the dynamic characteristic mapping function of the processing state, Is the weight matrix, Is the bias vector, Represents the feature activation function; S442. The multi-variable characteristic recognition unit, combining the dynamic characteristic mapping function And the target grinding parameter set , captures the anisotropic and high-hardness characteristics of carbon fiber sheets through multi-variable relationship modeling: ; Among them, Is the current processing state feature vector, And Are the state transition matrix and the parameter influence matrix respectively, Is the random perturbation term, Represents time; S443. The processing state prediction unit, combining the real-time input processing state representation vector And the target grinding parameter set , generates a dynamic prediction model for estimating the processing state at a future time: ; Among them, Is the predicted processing state vector.

[0031] S451. The parameter range generation unit, based on the predicted processing state vector And the set target processing state, generates the target parameter range through dynamic analysis ; Determine the target processing state based on feature similarity: ; Among them, Is the target processing state feature vector, is the similarity threshold; Dynamically adjust the target parameter range based on the timing changes of the processing status : ; in, is the initial parameter range, is the parameter increment, Indicates the sensitivity of the processing state to the parameters; S452, construct objective function based on multi-objective optimization principle of processing efficiency and processing stability : ; in, is the objective function, represents the normalized error between the machining state prediction and the target state, Indicates the processing time, and are the weight factors of machining stability and efficiency, and the target grinding parameter set is the variable to be optimized; S453, adaptive optimization unit, based on the objective function and target parameter range , a hybrid optimization algorithm, including a combination of local gradient search and global heuristic search, is used to iteratively calculate the target grinding parameter set , achieve fast convergence through local gradient search, and avoid falling into local optimum through global heuristic search: .

[0032] This implementation method constructs a processing state feature mapping model by adopting a contrastive learning algorithm, combining the high-dimensional processing state representation vector with the nonlinear feature relationship. It can not only capture the dynamic changes of anisotropy and high hardness characteristics during the processing of carbon fiber plates, but also dynamically predict the optimal grinding parameters through deep correlation mining between features.

[0033] S5, a real-time adaptive control module, which adjusts the processing parameters of the grinding execution module in real time according to the optimal grinding parameters output by the dynamic modeling module, so that the grinding force is always within the target range; In this implementation, S5 specifically includes: S51, a processing parameter receiving unit receives a target grinding parameter set output by a dynamic modeling module ; S52, real-time force monitoring unit, real-time monitoring of the normal force generated during the grinding process and tangential force , and the collected grinding force Transmitted to the grinding force adjustment unit; S53. A grinding force adjustment unit, based on a preset target grinding force range and the real-time monitored grinding force , dynamically calculates the grinding force deviation: ; Adjusts the grinding parameters according to the grinding force deviation: ; Wherein, is the grinding force adjustment gain matrix, represents the adjusted set of machining parameters; S54. A machining parameter execution unit, according to the adjusted set of machining parameters , real-time controls the grinding execution module, including dynamically adjusting the grinding wheel linear velocity , feed rate and machining depth , to make the machining process conform to the target range.

[0034] In this embodiment, through the real-time adaptive control module combined with the output of the dynamic modeling module, the machining parameters of the grinding execution module are adjusted in real time, realizing the precise control of the grinding force of the carbon fiber sheet. This module not only combines the target grinding parameters and the characteristics of the real-time monitored grinding force, but also through the dynamic calculation and feedback adjustment of the grinding force deviation, ensures that the grinding force during the machining process is always stable within the target range, thus effectively improving the machining quality and stability.

[0035] S6. A closed-loop feedback module, collects multi-modal data that changes in real time during the machining process, and feeds it back to the multi-modal data fusion module and the dynamic modeling module to form a data dynamic cycle.

[0036] In this embodiment, S6 specifically includes: The closed-loop feedback module is used to collect multi-modal data that changes in real time during the machining process, including grinding force, vibration signal, temperature signal and acoustic emission signal, and dynamically feeds the collected data back to the multi-modal data fusion module and the dynamic modeling module based on self-supervised learning, and continuously updates the machining state model and optimizes the grinding parameters through closed-loop interaction.

[0037] In this embodiment, by dynamically interacting the machining state monitored by the multi-modal data acquisition module with the closed-loop feedback module, a comprehensive analysis of the carbon fiber sheet machining process is provided, which can not only capture the current machining state characteristics, but also continuously optimize the grinding parameters based on the time series changes of the machining state.

[0038] Example: To verify the feasibility of the present invention, the adaptive control system for the grinding force of the carbon fiber sheet of the present invention is applied to the precision machining workshop of an aviation manufacturing enterprise, which is mainly used for the production of high-precision carbon fiber parts for aerospace. Due to its lightweight and high-strength characteristics, carbon fiber sheets have important applications in the aerospace field. However, during the processing, due to the anisotropy and high hardness of the material, problems such as large fluctuations in grinding force, unstable surface quality, and increased grinding wheel wear often occur, affecting the machining accuracy and production efficiency of the parts. To improve the processing quality, the enterprise decides to introduce the control system proposed in the present invention to replace the traditional processing control method based on fixed rules.

[0039] In the actual machining scenario, the system first deploys a multi-modal data acquisition module to collect various state signals generated during the grinding process in real time, including grinding force, vibration signal, temperature signal, acoustic emission signal, etc. These data are processed by the multi-modal data fusion module. The system denoises, aligns, and normalizes the data, and generates a high-dimensional machining state characterization vector to represent the multi-modal characteristics of the current machining state. The dynamic modeling module based on self-supervised learning dynamically models these state characterization vectors, uses contrastive learning technology to construct a machining state feature mapping model, captures the complex non-linear characteristics during the carbon fiber sheet processing, and dynamically predicts the optimal grinding parameter set in combination with historical machining data, including the grinding wheel linear speed, feed speed, and machining depth.

[0040] To test the actual performance of the system, three typical carbon fiber sheet machining tasks with different sizes are selected for the experiment, namely: small part machining, medium-sized structural part machining, and large composite plate machining. The experimental data are shown in Table 1, covering key indicators such as grinding force fluctuation, surface roughness, grinding wheel life, and machining efficiency during the processing. During the experiment, the system adjusts the machining parameters in real time in each scenario, and continuously optimizes the machining state through the closed-loop feedback module to ensure that the grinding force is stable within the target range.

[0041] Table 1 Comparison of machining performance of carbon fiber sheets It can be seen from the experimental results that due to the lack of real-time adjustment ability for the dynamic machining state, the traditional control method leads to a large fluctuation range of grinding force, a high surface roughness, a low machining efficiency, and a machining defect rate as high as 7.8%. In contrast, the system of the present invention significantly reduces the grinding force fluctuation range to ±2.8% through multi-modal data fusion and dynamic modeling technology, reduces the surface roughness to 0.75 µm, doubles the grinding wheel life to 36 hours, increases the average machining efficiency by about 16.7%, and reduces the machining defect rate to 1.3%. In addition, the system realizes the full-automatic adjustment of machining parameters without manual intervention, further improving the machining stability and production efficiency.

[0042] In a specific experiment, the system was applied to the processing tasks of medium-sized structural parts in Group B. During the middle stage of processing, the real-time force monitoring unit detected an instantaneous fluctuation in the grinding force, and the vibration signal exceeded the set target range. The system updated the processing state prediction model in real time through the dynamic modeling module and optimized the feed rate and processing depth within a short period of time. The adjusted grinding force was stabilized within the target range, and finally, the surface of the processed part had no scratches, and the overall quality was better than that of the traditional method.

[0043] In addition, in order to evaluate the long-term operation performance of the system, the processing tasks of large composite plates in Group C were selected for testing. The test showed that during the long-term continuous processing, the system was able to continuously optimize the grinding wheel linear speed and feed rate, significantly reducing the abnormal wear of the grinding wheel and avoiding the processing defects caused by unstable grinding force in the traditional method.

[0044] Through multi-modal data fusion, self-supervised learning, and real-time adaptive control technologies, the present invention realizes precise control of the grinding force of carbon fiber sheets, not only solving problems such as lag in processing parameter adjustment and large fluctuations in grinding force in traditional control methods, but also significantly improving processing accuracy and efficiency, providing a new technical solution for the precision processing of high-performance materials in the aerospace field.

[0045] The above is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A carbon fiber plate grinding force adaptive control system, characterized in that: include: S1, a grinding execution module, used for performing a grinding operation on a carbon fiber plate, the grinding execution module comprising an adjustable speed grinding wheel, a motor drive unit, a processing depth control unit, and a force sensing unit connected to a grinding device; S2, a multimodal data acquisition module, which acquires multimodal data in the processing process in real time through sensors, wherein the sensors include vibration sensors, temperature sensors, and acoustic emission sensors; S3, multimodal data fusion module, dynamically fuses multimodal data to generate a high-dimensional processing state representation vector; S4, a dynamic modeling module based on self-supervised learning, which uses contrastive learning technology to construct a processing state feature mapping model through a high-dimensional processing state representation vector, captures the anisotropy and high hardness processing characteristics of carbon fiber plates, and dynamically predicts the optimal grinding parameters; S5, a real-time adaptive control module, which adjusts the processing parameters of the grinding execution module in real time according to the optimal grinding parameters output by the dynamic modeling module, so that the grinding force is always within the target range; S6, closed-loop feedback module, collects multimodal data that changes in real time during the processing process, and feeds it back to the multimodal data fusion module and the dynamic modeling module to form a dynamic data cycle.

2. The carbon fiber plate grinding force adaptive control system according to claim 1, characterized in that: The S1 specifically includes: S11, adjustable speed grinding wheel, used to dynamically adjust the linear speed of the grinding wheel during the grinding process Processing of carbon fiber sheets, where the line speed The adjustment is based on the real-time force signal collected from the force sensing unit; S12, the motor drive unit is connected to the adjustable speed grinding wheel to provide power for driving the grinding operation, and the motor drive unit supports a speed adjustment function based on variable frequency control, and its speed range is ,in, Indicates the instantaneous speed provided by the motor drive unit, Indicates the minimum speed allowed by the motor drive unit. Indicates the maximum speed allowed by the motor drive unit; S13, the processing depth control unit is linked with the motor drive unit to control the processing depth of the adjustable speed grinding wheel and the surface of the carbon fiber plate The processing depth control unit includes a mechanical feeding mechanism and a position feedback device, which collects contact displacement signals in real time to control the processing depth. In the set range Internal dynamic adjustment, where and Respectively represent the minimum and maximum values ​​of the machining depth; S14, force sensing unit, installed between the grinding device and the workpiece, used to monitor the grinding force generated during the grinding process in real time, where the grinding force includes the normal force and tangential force .

3. The carbon fiber plate grinding force adaptive control system according to claim 2, characterized in that: The S3 specifically includes: S31, a data receiving unit, receiving the multimodal data collected by the multimodal data acquisition module, including the normal force collected by the force sensing unit and tangential force , vibration signal collected by vibration sensor , the temperature signal collected by the temperature sensor And the acoustic emission signal collected by the acoustic emission sensor , and perform preliminary dynamic fusion of multimodal data to generate preliminary fusion results ; S32, data alignment unit, for preliminary fusion results The data in the time dimension are aligned to make the initial fusion result Keep the timestamps of each data in the same, and generate a time-aligned data set ; S33, data denoising unit, uses adaptive filtering algorithm to denoise the time-aligned data set Perform denoising and minimize the residual error: ； in, Indicates that the time-aligned dataset is in time The value of is the weight parameter of the filter, is the sampling length, is the lag order, and the weight parameters of the filter are optimized iteratively The value of , eliminates the noise signal, and finally generates a valid data set after denoising ; S34, data standardization unit, for the effective data set after denoising Perform standardization to generate a standardized data set ; S35, state representation generation unit, based on the standardized data set Constructing a high-dimensional processing state representation vector , the high-dimensional processing state representation vector is defined as: ； in, Represents the feature mapping function extracted from the corresponding multimodal data.

4. The carbon fiber plate grinding force adaptive control system according to claim 3, characterized in that: The S4 specifically includes: S41, feature data preprocessing unit, used for processing the high-dimensional processing state representation vector generated by the multimodal data fusion module Screening and extraction of key feature components , and construct the feature input matrix : ； in, Indicates In the processing status samples eigenvalues, is the sample size, is the feature dimension; S42, comparison sample generation unit, from the feature input matrix Generate training samples for contrastive learning; S43, a feature mapping model building unit, builds a processing state feature mapping model based on a contrastive learning algorithm, and optimizes a contrastive loss function: ； in, represents the contrast loss function, and is the positive sample pair in the training sample, is the negative sample in the training sample, represents the similarity between samples, is the temperature coefficient, M is the total number of samples, and a feature mapping model that can capture the characteristics of the processing state is generated; S44, dynamic modeling unit, combining feature mapping model and high-dimensional processing state representation vector ,dynamically model the processing state of carbon fiber sheet, and generate a dynamic prediction model of processing state by identifying nonlinear characteristics and multivariate dynamic relationships; S45, optimal parameter calculation unit, uses the dynamic prediction model of machining state to calculate the target grinding parameter set ,in is the grinding wheel linear speed, is the feed speed, is the processing depth.

5. The carbon fiber plate grinding force adaptive control system according to claim 4, characterized in that: The S42 specifically includes: S421, dynamic processing parameter sampling unit, dynamically adjusts the target grinding parameter set according to the real-time processing status , calculate the high-dimensional processing state representation vector The timing change rate : ； in, is the sampling time interval; S422, intelligent sample generation unit, from the dynamically adjusted target grinding parameter set Construct positive sample pairs and negative sample pairs; Positive sample pairs are generated based on the similarity of processing state feature vectors: ； in, is the similarity threshold of the positive sample pair, represents the similarity function between feature vectors, Represents the feature vector The corresponding processing parameter combination, Represents the feature vector Corresponding processing parameter combination; Negative sample pairs are generated based on the difference of processing state feature vectors: ； in, is the similarity threshold of negative sample pairs, Represents the feature vector Corresponding processing parameter combination; S423, a multi-dimensional feature weight adjustment unit dynamically assigns weights based on feature distribution of sample pairs: ； in, Positive sample Or negative sample pair The weight of is the adjustment factor of weight distribution, is the square of the Euclidean distance between samples; S424. Generate final training samples by optimizing the ratio of positive and negative samples and combining the distribution of sample pairs after weight adjustment.

6. The carbon fiber plate grinding force adaptive control system according to claim 5, characterized in that: The S44 specifically includes: S441, dynamic feature relationship analysis unit, based on feature mapping model and high-dimensional processing state representation vector , through nonlinear regression analysis to explore the dynamic relationship between processing state features: ； in, is the dynamic characteristic mapping function of the processing state, is the weight matrix, is the bias vector, represents the feature activation function; S442, multivariate feature recognition unit, combined with dynamic feature mapping function and target grinding parameter set , capturing the anisotropic properties and high stiffness characteristics of carbon fiber sheets through multivariate relationship modeling: ； in, is the current processing state feature vector, and are the state transfer matrix and parameter influence matrix respectively, is the random disturbance term, Indicates time; S443, processing state prediction unit, combined with the processing state representation vector input in real time and target grinding parameter set , generate a dynamic prediction model for estimating the processing state at future moments: ； in, is the predicted processing state vector.

7. The carbon fiber plate grinding force adaptive control system according to claim 6, characterized in that: The S45 specifically includes: S451, parameter range generation unit, according to the predicted processing state vector and the set target processing state, generating the target parameter range through dynamic analysis ; Determine the target processing status based on feature similarity: ； in, is the target processing state feature vector, is the similarity threshold; Dynamically adjust the target parameter range based on the timing changes of the processing status : ； in, is the initial parameter range, is the parameter increment, Indicates the sensitivity of the processing state to the parameters; S452, construct objective function based on multi-objective optimization principle of processing efficiency and processing stability : ； in, is the objective function, represents the normalized error between the machining state prediction and the target state, Indicates the processing time, and are the weight factors of machining stability and efficiency, and the target grinding parameter set is the variable to be optimized; S453, adaptive optimization unit, based on the objective function and target parameter range , a hybrid optimization algorithm, including a combination of local gradient search and global heuristic search, is used to iteratively calculate the target grinding parameter set , achieve fast convergence through local gradient search, and avoid falling into local optimum through global heuristic search: 。 8. The carbon fiber plate grinding force adaptive control system according to claim 7, characterized in that: The S5 specifically includes: S51, a processing parameter receiving unit receives a target grinding parameter set output by a dynamic modeling module ; S52, real-time force monitoring unit, real-time monitoring of the normal force generated during the grinding process and tangential force , and the collected grinding force Transmitted to the grinding force adjustment unit; S53, grinding force adjustment unit, based on the preset target grinding force range and real-time monitoring of grinding forces , dynamic calculation of grinding force deviation: ； Adjust grinding parameters according to grinding force deviation: ； in, Adjust the gain matrix for grinding forces, represents the adjusted processing parameter set; S54, the processing parameter execution unit, according to the adjusted processing parameter set , real-time control of the grinding execution module, including dynamic adjustment of the grinding wheel linear speed , Feed speed and processing depth , so that the processing process meets the target range.

Images