A shredder control method, system and medium based on target identification for destruction

Abstract

The application relates to a shredder control method, system and medium based on destruction target identification, and belongs to the technical field of intelligent control. The control method comprises the following steps: collecting multi-modal sensing data of a target object; identifying a medium type, extracting physical property characteristics, and generating medium characteristic data; querying a preset parameter mapping table, combining temperature and humidity data to generate a dynamic compensation coefficient, and outputting initial shredding control parameters; driving a shredding execution mechanism and monitoring mechanical resistance values in real time during the shredding process; when the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold value, the initial shredding control parameters are adjusted in real time; performing spectral scanning on the residues after shredding, generating residue distribution data, verifying the destruction compliance, and outputting a verification report; storing the medium characteristic data, the shredding control parameters and the residue distribution data into a training set, and regularly updating the preset parameter mapping table. The application can meet the high standard requirements for physical medium disposal in modern high-security environments.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology, and in particular to a crusher control method, system and medium based on the identification of destruction targets. Background Technology

[0002] As the country's requirements for information security, data privacy protection, and hazardous waste management become increasingly stringent, the secure destruction of various physical media containing sensitive information (such as paper documents, hard drives, USB flash drives, encryption chips, etc.) at the end of their life cycle has become a rigid requirement in high-security scenarios such as government agencies, military units, financial institutions, and data centers.

[0003] Currently, most common crushing equipment relies on fixed process parameters and lacks the ability to actively identify the characteristics of materials. This makes it difficult to handle complex and ever-changing objects to be destroyed, leading to problems such as incomplete destruction, equipment damage, excessive energy consumption, or lack of operational traceability. Especially when dealing with disguised items, composite materials, or high-density electronic storage components, mechanical crushing using preset modes often cannot ensure the complete destruction of all information carriers. Residual fragments may still carry recoverable data traces, posing a potential risk of leakage and failing to meet the high standards required for physical media disposal in modern high-security environments. Summary of the Invention

[0004] To improve the safety of destruction operations, this application provides a shredder control method, system, and medium based on destruction target identification.

[0005] Firstly, this application provides a shredder control method based on destruction target identification, employing the following technical solution:

[0006] A shredder control method based on target identification for destruction, the control method comprising:

[0007] Multimodal sensing data of the target object is acquired through a multi-sensor array; wherein, the multimodal sensing data includes optical images, mass measurements, electronic tag information, and a three-dimensional point cloud coordinate set;

[0008] Based on the multimodal sensing data, the medium type of the target object is identified, and physical property features are extracted to generate medium feature data including a medium type identifier and physical property features.

[0009] Based on the medium characteristic data, a preset parameter mapping table is queried, and dynamic compensation coefficients are generated by combining the temperature and humidity data collected in real time by the environmental sensor, and the initial crushing control parameters are output.

[0010] The crushing actuator is driven according to the initial crushing control parameters, and the mechanical resistance value is monitored in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold, the initial crushing control parameters are adjusted in real time.

[0011] The spectral scan of the pulverized residue is performed to generate residue distribution data. The compliance of the disposal is verified based on the residue distribution data, and a verification report is generated.

[0012] The media characteristic data, crushing control parameters, and residue distribution data are stored in a training set. Based on the training set, the mapping relationship between the media characteristic data and the crushing control parameters is refitted to periodically update the preset parameter mapping table.

[0013] By adopting the above technical solution, based on multimodal sensor data fusion to identify target object features, the crushing parameters can be automatically optimized and adjusted in real time according to different materials, ensuring the efficiency and thoroughness of the destruction process. At the same time, the compliance of destruction is verified through residue detection, forming a data closed-loop feedback mechanism to achieve intelligent and adaptive crushing control, which significantly improves the safety, accuracy and traceability of destruction operations. It has important practical value and promotion significance in fields such as confidential document processing and hazardous waste disposal.

[0014] Optionally, the step of identifying the medium type of the target object based on the multimodal sensing data, extracting physical property features, and generating medium feature data including a medium type identifier and physical property features includes:

[0015] Receive multimodal sensing data acquired by a multi-sensor array; wherein, the multimodal sensing data includes optical images, quality measurements, electronic tag information, and a three-dimensional point cloud coordinate set;

[0016] The optical image is subjected to illumination compensation processing to generate standardized image data, and the electronic tag information is parsed to obtain the pre-stored medium identification code.

[0017] The standardized image data is input into a pre-trained surface material classification model, which outputs the probability distribution of material categories and calculates the geometric structure feature vector by combining the three-dimensional point cloud coordinate set.

[0018] By integrating the probability distribution of the material category, the geometric structure feature vector, and the quality measurement value, a medium type determination result is generated through a decision tree ensemble algorithm;

[0019] The physical property database is indexed based on the medium type determination result, and the physical property features of the corresponding medium are extracted; wherein, the physical property features include density threshold range and hardness benchmark value;

[0020] The geometric feature vector is combined to generate medium feature data containing medium type identifier and physical property features.

[0021] By adopting the above technical solutions, a complete processing chain was constructed, from multimodal data acquisition, feature extraction, intelligent recognition to physical attribute association, enabling accurate identification and comprehensive feature description of the target object's medium type. Through multi-level information processing and feature fusion, the accuracy and reliability of medium type identification were significantly improved, laying a solid data foundation and technical support for subsequent intelligent crushing control.

[0022] Optionally, the step of generating a medium type determination result by fusing the material category probability distribution, geometric structure feature vector, and quality measurement value using a decision tree ensemble algorithm includes:

[0023] Receive material category probability distribution data, geometric structure feature vectors, and mass measurement values; wherein, the material category probability distribution data contains probability value sequences of at least three materials;

[0024] Calculate the standardized deviation between the measured quality value and the preset medium quality benchmark;

[0025] Principal component dimensionality reduction is performed on the probability value sequence, and the top k principal components are extracted as a subset of material features;

[0026] The geometric feature vector is decomposed into spatial attribute components and structural stability components;

[0027] The standardized deviation is mapped to logarithmic space to generate a quality compensation factor.

[0028] The material feature subset, spatial attribute component, structural stability component, and mass compensation factor are combined to generate a joint feature vector;

[0029] The joint feature vectors are input in parallel into the pre-trained random forest sub-model and gradient boosting tree sub-model;

[0030] Determine whether the output types of the random forest sub-model and the gradient boosting tree sub-model are consistent; if so, directly output the type identifier as the medium type determination result; if not, activate the convolutional neural network arbitration module to generate the final type identifier as the medium type determination result.

[0031] By adopting the above technical solutions, the discriminative information of material probability is preserved based on PCA dimensionality reduction, geometric feature decoupling avoids interference from micro / macro features, and the parallel architecture of random forest and gradient boosting tree achieves efficient and stable classification decision function for complex media, greatly reducing the misclassification rate.

[0032] Optionally, the steps of querying a preset parameter mapping table based on the medium characteristic data, generating a dynamic compensation coefficient by combining the temperature and humidity data collected in real time by the environmental sensor, and outputting the initial crushing control parameters include:

[0033] Receive media characteristic data containing media type identifier and physical attribute characteristics;

[0034] The basic crushing parameter set is obtained by querying the preset parameter mapping table according to the medium type identifier; wherein, the basic crushing parameter set includes the reference speed value, the reference pressure value, and the reference duration value;

[0035] Real-time reading of temperature and humidity data collected by environmental sensors;

[0036] Based on the hardness benchmark value in the physical property characteristics and the temperature and humidity data, a dynamic compensation coefficient is generated through a dynamic compensation model; wherein, the dynamic compensation coefficient includes a rotational speed dynamic compensation coefficient and a pressure dynamic compensation coefficient.

[0037] The basic crushing parameter set is combined with the dynamic compensation coefficient input parameter optimization engine to calculate the compensation crushing control parameters;

[0038] Load the compensated crushing control parameters and the three-dimensional model of the medium into a virtual environment, perform a crushing process simulation, and generate risk prediction values.

[0039] When the predicted risk value exceeds the preset risk threshold, the compensation crushing control parameters are adjusted and the simulation is repeated until the predicted risk value meets the target, at which point the initial crushing control parameters are output.

[0040] By adopting the above technical solutions and deeply integrating the three core technical elements of the perception layer, cognition layer, and decision-making layer, a fundamental shift from passive response to proactive prediction has been achieved. By integrating information flows from various heterogeneous sensors and relying on data analysis and a highly efficient and reliable simulation verification toolchain, the robustness, flexibility, and reliability of the crushing system have been improved.

[0041] Optionally, the step of calculating the compensated crushing control parameters by combining the basic crushing parameter set with the dynamic compensation coefficient input parameter optimization engine includes:

[0042] Receive a set of basic crushing parameters and dynamic compensation coefficients; wherein, the set of basic crushing parameters includes a reference speed value, a reference pressure value and a reference duration value, and the dynamic compensation coefficients include a speed dynamic compensation coefficient and a pressure dynamic compensation coefficient;

[0043] Load the preset equipment safety boundary constraint set, including the maximum permissible speed threshold, pressure tolerance range, and bearing thermal load limit;

[0044] The reference speed value is multiplied by the speed dynamic compensation coefficient to obtain the initial speed candidate value, and the reference pressure value is multiplied by the pressure dynamic compensation coefficient to obtain the initial pressure candidate value.

[0045] Construct a weighted combined objective function of minimizing energy consumption and maximizing output; where the energy consumption weight is associated with the real-time electricity price factor, and the output weight is associated with the physical properties of the medium.

[0046] Under the constraints of the equipment safety boundary set, the optimal solution of the weighted combined objective function is solved by a sequential quadratic programming algorithm to generate optimized values ​​for rotational speed and pressure.

[0047] Boundary verification is performed on the optimized rotation speed and optimized pressure values, and a set of compensated crushing control parameters is output.

[0048] By adopting the above technical solutions, an intelligent optimization architecture with constraint-solving capabilities was constructed. This architecture enables refined reconstruction of crushing process parameters while ensuring system operational safety, thus maintaining efficient, energy-saving, and stable crushing performance even under highly variable actual operating conditions. More importantly, through the synergistic effect of sequential quadratic programming and boundary rebalancing mechanisms, it ensures that even under extreme conditions, it can output control parameters that are both efficient and safe. This significantly improves the reliability, energy efficiency ratio, and intelligence level of crushing operations, making it particularly suitable for high-end destruction scenarios with stringent requirements for information security, equipment lifespan, and operating costs.

[0049] Optionally, under the constraints of the equipment safety boundary set, the step of solving the optimal solution of the weighted combined objective function using a sequential quadratic programming algorithm to generate the optimized values ​​of rotational speed and pressure includes:

[0050] Obtain the initial speed candidate value, the initial pressure candidate value, and the weighted combined objective function;

[0051] Construct a joint speed-pressure constraint equation based on the equipment safety boundary constraint set;

[0052] Based on the sequential quadratic programming algorithm, the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure are used as the starting point for optimization. An iterative optimization loop is executed, and the current optimization variables of rotational speed and pressure are output when the optimal solution is reached.

[0053] The current speed optimization variable and pressure optimization variable are respectively used as the speed optimization value and pressure optimization value.

[0054] By adopting the above technical solution, starting with the compensated initial candidate values, the optimal solution of the weighted objective function is searched within strict safety boundaries using a sequential quadratic programming algorithm. Efficient iteration is achieved through the sequential quadratic programming algorithm, supplemented by a degradation mechanism to ensure system robustness. This technical solution not only solves the problem of balancing energy consumption and efficiency in traditional crushing control, but also incorporates latent failure risks such as bearing thermal load into the explicit modeling scope, improving the long-term reliability of the equipment.

[0055] Optionally, based on the sequential quadratic programming algorithm, using the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure as the starting point for optimization, executing an iterative optimization loop, and outputting the current optimized variables of rotational speed and pressure when the optimal solution is reached includes the following steps:

[0056] Calculate the gradient vector of the weighted combined objective function at the current iteration point;

[0057] The approximation of the Hessian matrix is ​​updated using the quasi-Newton method, generating the quadratic coefficient matrix of the quadratic programming subproblem.

[0058] Linearize the combined speed-pressure constraint equation at the current iteration point to generate a set of linear constraint equations;

[0059] Based on the quadratic term coefficient matrix, the objective function gradient vector, and the linear constraint equations, the quadratic programming subproblem is solved to obtain the search direction vector.

[0060] Update the rotation speed optimization variables and pressure optimization variables along the search direction vector to generate a new iteration point;

[0061] When the change in the objective function value between the new iteration point and the previous iteration point is less than the dynamic convergence threshold, the optimal solution is determined to have been reached, the loop is terminated, and the current speed optimization variable and pressure optimization variable are output.

[0062] By adopting the above technical solution, the crushing control parameters are refined and optimized under multiple objectives and constraints, forming a closed-loop, self-consistent, and highly intelligent decision-making chain. This technical solution not only considers the rigid limitations of the equipment, such as mechanical strength, thermal stability, and hydraulic load-bearing capacity, but also incorporates external economic and process factors such as real-time electricity price fluctuations and differences in material physical properties. This enables the control system to automatically balance the contradiction between energy consumption costs and processing efficiency while ensuring safety.

[0063] Optionally, after the steps of verifying the compliance of the destruction based on the residue distribution data and generating a verification report, the following may also be included:

[0064] When the destruction compliance verification fails, obtain the coordinate set of metal accumulation areas from the residue distribution data;

[0065] A three-dimensional geometric thermal map is generated based on the coordinate set of the metal accumulation region; wherein, the thermal value represents the local metal mass density;

[0066] In the three-dimensional geometric heat map, locate the thermal peak point, and use the thermal peak point as the seed point to execute the region growing algorithm to extract the isosurface boundary coordinate set;

[0067] Construct a Delaunay triangular mesh based on the isosurface boundary coordinate set, and calculate the curvature distribution characteristics of the mesh vertices;

[0068] Based on the curvature distribution characteristics, high stress concentration sub-regions are identified, and a laser energy mapping table is generated by combining a preset metal material yield strength database.

[0069] The laser focusing path is planned according to the laser energy mapping table, and a laser crushing control command is sent to control the laser to perform gradient energy output.

[0070] Re-execute the spectral scan to generate updated residue distribution data, and repeat the above steps until the residue distribution data passes the destruction compliance verification.

[0071] By adopting the above technical solutions, a deep destruction enhancement system has been constructed where conventional mechanical shredding cannot meet safety standards. By converting spectral information into a computable spatial field and mapping geometric features into controllable energy input, the reliability of destroying high-value sensitive information carriers is significantly improved. This system is particularly suitable for application scenarios with extremely high information security requirements, such as government agencies, military units, and data centers.

[0072] Secondly, this application provides a shredder control system based on the identification of destruction targets, which adopts the following technical solution:

[0073] A shredder control system based on target identification for destruction, the control system comprising:

[0074] The data acquisition module is used to acquire multimodal sensing data of the target object through a multi-sensor array; wherein, the multimodal sensing data includes optical images, mass measurement values, electronic tag information, and a three-dimensional point cloud coordinate set;

[0075] The medium feature generation module is used to identify the medium type of the target object based on the multimodal sensing data, extract physical attribute features, and generate medium feature data including a medium type identifier and physical attribute features.

[0076] The dynamic compensation module is used to query a preset parameter mapping table based on the medium characteristic data, generate a dynamic compensation coefficient by combining the temperature and humidity data collected in real time by the environmental sensor, and output the initial crushing control parameters.

[0077] The crushing control module is used to drive the crushing actuator according to the initial crushing control parameters, and monitor the mechanical resistance value in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold, the initial crushing control parameters are adjusted in real time.

[0078] The residue verification module is used to perform spectral scanning on the residue after pulverization, generate residue distribution data, verify the compliance of destruction based on the residue distribution data, and output a verification report.

[0079] The update feedback module is used to store the medium characteristic data, crushing control parameters and residue distribution data into the training set, and refit the mapping relationship between the medium characteristic data and the crushing control parameters based on the training set, so as to periodically update the preset parameter mapping table.

[0080] Thirdly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0081] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0082] In summary, this application includes at least one of the following beneficial technical effects: by using multimodal sensor data fusion to identify the medium type and physical characteristics of the target to be destroyed, dynamically adjusting the crushing parameters in combination with environmental factors, and monitoring and adjusting in real time during the crushing process, and finally verifying the destruction effect through residue analysis, a closed-loop control and continuous optimization mechanism is formed, realizing intelligent and precise crushing control, and improving destruction efficiency and safety. Attached Figure Description

[0083] Figure 1 This is a first flowchart illustrating a shredder control method based on destruction target identification, one embodiment of the application.

[0084] Figure 2 This is a second flowchart illustrating a shredder control method based on destruction target identification, which is one embodiment of the application.

[0085] Figure 3 This is a schematic diagram of the third process of a shredder control method based on destruction target identification, which is one embodiment of the application.

[0086] Figure 4 This is a schematic diagram of the fourth process of a shredder control method based on destruction target identification, which is one embodiment of the application.

[0087] Figure 5 This is a fifth flowchart of a shredder control method based on destruction target identification, which is one embodiment of the application.

[0088] Figure 6 This is a schematic diagram of the sixth process of a shredder control method based on destruction target identification, which is one embodiment of the application.

[0089] Figure 7 This is a schematic diagram of the seventh process of a shredder control method based on destruction target identification, which is one embodiment of the application.

[0090] Figure 8 This is a schematic diagram of the eighth process of a shredder control method based on destruction target identification, which is one embodiment of the application. Detailed Implementation

[0091] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-8 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0092] This application discloses a shredder control method based on the identification of destruction targets.

[0093] Reference Figure 1 A shredder control method based on target identification for destruction, the control method comprising:

[0094] Step S101: Collect multimodal sensing data of the target object through a multi-sensor array; wherein, the multimodal sensing data includes optical images, mass measurement values, electronic tag information, and a three-dimensional point cloud coordinate set;

[0095] Among them, multimodal sensing data refers to the use of multiple different types of sensors to simultaneously acquire information about different dimensions of the target object, thereby constructing a more comprehensive and accurate target description system.

[0096] Specifically, the system is equipped with various devices, including a visible light camera, an infrared depth sensor, a weighing sensor, and an RFID / NFC module. These devices are responsible for capturing the target object's appearance texture (optical image), spatial geometry (3D point cloud coordinate set), weight information (mass measurement value), and embedded identification (electronic tag information), respectively.

[0097] Understandably, this design ensures accurate identification even when dealing with complex and diverse items to be destroyed, such as paper documents, hard drives, USB flash drives, or other carriers containing sensitive information, through cross-verification. For example, a piece of paper might be disguised as other materials, but its density, thickness, and material composition can be inferred through image analysis and weighing, while the built-in RFID tag provides an additional identification mechanism. Therefore, the core objective of this stage is to establish a high-fidelity digital target model, providing a reliable input source for subsequent data processing.

[0098] Step S102: Identify the medium type of the target object based on multimodal sensing data, extract physical property features, and generate medium feature data including medium type identifier and physical property features;

[0099] During this process, the algorithm fuses various received raw signals. For example, it uses a convolutional neural network (CNN) to classify and identify optical images to initially determine the material type, then combines 3D point cloud reconstruction technology to analyze shape characteristics, and uses RFID decoding to obtain officially defined medium category tags. At the same time, by quantitatively modeling visual features such as image texture, surface reflectivity, and color histograms, the specific physical properties of the medium can be further refined, such as hardness, brittleness, and conductivity.

[0100] Furthermore, machine learning models can be introduced to assist in completing more complex nonlinear mapping tasks, that is, deriving potential state parameters from a set of observed variables. The resulting media characteristic data not only includes clear category classifications (such as "paper documents" and "solid-state drives"), but also encompasses rich quantifiable physical feature vectors, forming an important basis for subsequent control system adjustment strategies.

[0101] Step S103: Query the preset parameter mapping table based on the medium characteristic data, generate dynamic compensation coefficients by combining the temperature and humidity data collected in real time by the environmental sensor, and output the initial crushing control parameters.

[0102] The preset parameter mapping table is essentially a pre-trained database or rule engine that establishes the correspondence between various media and their optimal crushing processes. Each record represents the most effective combination of operating parameters under specific conditions, such as cutter speed, feed rate, and applied pressure.

[0103] However, real-world scenarios often involve numerous uncertainties, especially as changes in environmental conditions can significantly impact mechanical performance. To address this, the system introduces a temperature and humidity detection unit as an external feedback channel to assess in real-time the degree to which the current operating state deviates from ideal conditions. Then, using empirical formulas or regression models, corresponding dynamic compensation factors are calculated to correct the baseline control parameters, enabling them to better adapt to actual operating conditions.

[0104] For example, reduced motor efficiency in high-temperature environments may lead to insufficient actual torque, in which case the voltage supply should be appropriately increased; while in humid environments, a decreased coefficient of friction may exacerbate slippage, requiring increased clamping force to maintain stable cutting. This dynamic parameter adjustment mechanism greatly enhances the robustness and versatility of the system.

[0105] Step S104: Drive the crushing actuator according to the initial crushing control parameters, and monitor the mechanical resistance value in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than the dynamic threshold, adjust the initial crushing control parameters in real time.

[0106] Specifically, once the appropriate initial parameters are determined, the actuators begin to operate and perform the crushing operation. Simultaneously, devices such as force-sensitive strain gauges and piezoelectric crystals installed at key locations continuously monitor load changes. Because different materials have different mechanical response curves, the magnitude of the expected reaction force at a given moment can theoretically be predicted based on a theoretical model. If the measured values ​​exceed a reasonable range, i.e., the deviation is large, it indicates that the current settings are no longer applicable and adjustments must be made immediately.

[0107] Furthermore, based on the deviation and the crushing stage identifier, a speed correction factor and a pressure correction factor are generated through a multi-objective optimization algorithm. The speed correction factor is superimposed on the target speed value, and the pressure correction factor is superimposed on the target pressure value, so that the updated crushing control parameters can be output and the actuator can be driven.

[0108] The execution process of the multi-objective optimization algorithm includes: establishing a speed correction objective function (minimizing the bearing temperature rise rate) and a pressure correction objective function (maximizing fragment particle size uniformity), and solving for the optimal combination of correction factors through the Pareto front. In this embodiment, the speed correction factor acts on the PWM duty cycle of the motor drive signal, and the pressure correction factor is converted into a hydraulic valve opening increment command.

[0109] In particular, considering the differences in stress behavior among different types of media, embodiments of this application also set differentiated tolerance boundaries, or so-called dynamic thresholds, according to different media types. For more fragile electronic storage media, smaller errors are allowed, while for more resilient paper materials, larger fluctuations are acceptable. This fine-grained distinction helps avoid misjudgment and misadjustment, improving the sensitivity and stability of the entire system.

[0110] In some embodiments, when the medium type is electronic storage medium, the dynamic threshold is configured as expected value × 5%; when the medium type is paper medium, the dynamic threshold is configured as expected value × 10%.

[0111] Step S105: Perform spectral scanning on the pulverized residue to generate residue distribution data, verify the compliance of the destruction based on the residue distribution data, and output a verification report.

[0112] The residue distribution data includes the maximum residue size and metal residue density, with the maximum fragment size and metal residue density being used as labels for compliance with destruction standards.

[0113] Specifically, by conducting Raman spectroscopy, near-infrared absorption spectroscopy, or multi-band fluorescence detection on the residue sample, a series of microscopic information can be obtained, such as particle size distribution, organic / inorganic component ratio, and residual magnetic material content. This information can then be compared with established safety standards to determine whether the criteria for complete removal are met.

[0114] For example, for classified data carriers, the maximum size of fragments is typically required to be no more than 1 millimeter, and the metal residue rate must be less than one-thousandth. If these standards are not met, the system will initiate an anomaly backtracking procedure. On the one hand, it will alert higher authorities to request manual review; on the other hand, it will add this failure case to the knowledge base for future reference and optimization.

[0115] Step S106: Store the media characteristic data, crushing control parameters and residue distribution data into the training set, and refit the mapping relationship between the media characteristic data and the crushing control parameters based on the training set, so as to periodically update the preset parameter mapping table.

[0116] Specifically, the medium characteristic data includes at least a medium type identifier and a hardness correction value, and the residue distribution data includes the maximum residue size and the metal residue density. The training set is input into the incremental learning model, the neural network weights are updated through an online backpropagation algorithm, the optimized medium type-parameter mapping function is output, and the parameter mapping table entries are regenerated. When a new medium type identifier is detected, the mapping table dimension is automatically expanded.

[0117] The incremental learning model adopts a two-layer LSTM network structure. The first layer processes the temporal features of the physical properties of the medium, and the second layer integrates the environmental temperature fluctuation pattern. The mapping relationship of historical medium types can be retained through the elastic weight solidification algorithm.

[0118] Understandably, as more and more practical cases accumulate, the original mapping table may become outdated or incomplete. In such cases, statistical methods or deep learning tools are needed to iterate and upgrade it. The newly added historical data not only expands the original sample size but also creates conditions for uncovering hidden patterns.

[0119] For example, some novel composite materials that have not been encountered before may exhibit unique damage threshold ranges. After sufficient learning, new coping strategies can be automatically summarized and fixed into a mapping table for direct use next time. Through this iterative process, the generalization ability and accuracy of the entire system will be steadily improved.

[0120] In the above embodiments, based on the multimodal sensor data fusion to identify the characteristics of the target object, the crushing parameters can be automatically optimized and adjusted in real time according to different materials, ensuring the efficiency and thoroughness of the destruction process. At the same time, the compliance of the destruction is verified by the detection of residues, forming a data closed-loop feedback mechanism to achieve intelligent and adaptive crushing control. This significantly improves the safety, accuracy and traceability of the destruction operation, and has important practical value and promotion significance in the fields of confidential document processing and hazardous waste disposal.

[0121] Reference Figure 2 As one implementation of step S102, the step of identifying the medium type of the target object based on multimodal sensing data, extracting physical attribute features, and generating medium feature data including a medium type identifier and physical attribute features includes:

[0122] Step S201: Receive multimodal sensing data acquired by a multi-sensor array; wherein, the multimodal sensing data includes optical images, quality measurements, electronic tag information, and a three-dimensional point cloud coordinate set;

[0123] Understandably, this multi-sensor configuration not only overcomes the limitations of a single sensor in complex environments, but also improves the robustness of the recognition results through data redundancy. The optical image provides texture and color information, the quality measurement reflects the overall physical characteristics of the object, the electronic tag information carries a predefined identity, and the three-dimensional point cloud coordinate set describes the spatial geometry of the object.

[0124] Step S202: Perform illumination compensation processing on the optical image to generate standardized image data, and simultaneously parse the electronic tag information to obtain the pre-stored medium identification code;

[0125] Since changes in ambient lighting conditions significantly impact the stability of image features and the generalization ability of classification models, illumination normalization is necessary to eliminate the influence of illumination non-uniformity on subsequent analysis. Normalization typically involves white balance correction, contrast adjustment, and brightness equalization, techniques that effectively improve the quality and consistency of image data. Meanwhile, the parsing of electronic tag information utilizes the data decoding principle of radio frequency identification (RFID) technology. By reading the pre-stored medium identification code in the RFID chip, partial object identification information can be directly obtained. This provides prior knowledge support for subsequent medium type determination, forming a hybrid identification strategy based on a combination of known information and unknown features.

[0126] Step S203: Input standardized image data into the pre-trained surface material classification model, output the probability distribution of material categories, and calculate the geometric structure feature vector by combining the three-dimensional point cloud coordinate set;

[0127] This material classification model is typically based on a convolutional neural network architecture. It learns abstract representations of visual features such as texture, gloss, and graininess of different material surfaces through training on a large number of labeled samples. The material category probability distribution output by the model reflects the likelihood of the input image belonging to each material category. This probabilistic output provides a quantitative basis for subsequent uncertainty handling and multi-evidence fusion.

[0128] Meanwhile, by combining the geometric feature vectors of the three-dimensional point cloud coordinate set with the mathematical principles of computational geometry and differential geometry, and by performing operations such as principal component analysis, surface fitting and topological feature extraction on the point cloud data, a quantitative description of the object's shape can be obtained, including geometric properties such as size parameters, surface curvature and volume features. These features are of great significance for distinguishing materials with similar appearances but different structural properties.

[0129] Step S204: The probability distribution of material category, geometric structure feature vector and quality measurement value are fused, and the medium type determination result is generated by decision tree ensemble algorithm;

[0130] By organically fusing various heterogeneous features such as material category probability distribution, geometric structure feature vectors, and mass measurements, and leveraging the advantages of ensemble methods like random forests and gradient boosting trees, classification accuracy can be effectively improved while reducing the risk of overfitting. The ensemble algorithm constructs multiple base classifiers and integrates the outputs of each model using weighted voting or average prediction, fully utilizing the information value of different feature spaces. This multi-feature fusion strategy not only considers the visual characteristics of an object's surface but also combines geometric structure information and physical weight features, forming a comprehensive recognition criterion system.

[0131] Step S205: Index the physical property database according to the medium type determination result and extract the physical property features of the corresponding medium; wherein, the physical property features include density threshold range and hardness benchmark value;

[0132] Specifically, by establishing a mapping relationship between media type and physical property parameters, the system can automatically retrieve key physical parameters such as the standard density range and hardness benchmark value of the corresponding material. This database-based knowledge retrieval mechanism not only improves the system's practicality but also provides a scientific basis for subsequent crushing control parameter setting.

[0133] Step S206: Combine the geometric structure feature vector to generate medium feature data containing medium type identifier and physical property features.

[0134] In this process, the final medium characteristic data is generated by combining geometric structure feature vectors. In fact, it is an organic combination of qualitative identification results and quantitative physical parameters. The resulting structured data contains information in multiple dimensions such as medium type identifier, density range, and hardness correction value, providing a complete description of the target object for the entire crushing control system.

[0135] In the above embodiments, a complete processing chain is constructed, from multimodal data acquisition, feature extraction, intelligent recognition to physical attribute association, enabling accurate identification and comprehensive feature description of the target object's medium type. Through multi-level information processing and feature fusion, the accuracy and reliability of medium type identification are significantly improved, laying a solid data foundation and technical support for subsequent intelligent crushing control.

[0136] Reference Figure 3 As one implementation of step S204, the step of generating a medium type determination result by fusing material category probability distribution, geometric structure feature vector, and quality measurement values ​​through a decision tree ensemble algorithm includes:

[0137] Step S301: Receive material category probability distribution data, geometric structure feature vector, and mass measurement value; wherein, the material category probability distribution data contains probability value sequences of at least three materials;

[0138] Specifically, the material category probability distribution data reflects the likelihood that the current sample belongs to several known material types; for example, each material such as metal, plastic, and ceramic corresponds to a confidence score. The geometric feature vector, on the other hand, characterizes the object's morphological properties in three-dimensional space, including but not limited to spatial dimension ratios (such as the length-width-height ratio), surface curvature statistics (used to describe the degree of drastic local shape changes), and contour complexity indices (reflecting edge jaggedness or smoothness). The mass measurement value comes from a high-precision weighing sensor installed at the end of the conveyor belt or at the crushing inlet. This value needs to be further converted into a mass deviation relative to a standard medium for comparison and integration with other features within the same framework.

[0139] Step S302: Calculate the standardized deviation between the measured quality value and the preset medium quality benchmark;

[0140] The standardized deviation, or Z-score, is calculated by subtracting the standard mass of a typical medium from the measured mass and then dividing by its standard deviation. This is done to eliminate interference from differences in absolute mass between different materials, allowing for cross-sectional comparisons within the same evaluation system even between objects with similar densities but different volumes.

[0141] For example, for aluminum alloy products, if the theoretical mass is set to 500g ± 10%, and the actual measured mass is 480g, the corresponding Z-score is approximately -0.4. This standardized deviation value not only retains the directionality of the original information (positive values ​​represent heavier weight, and negative values ​​represent lighter weight), but also provides a good input basis for further introducing nonlinear mapping.

[0142] Step S303: Perform principal component dimensionality reduction on the probability value sequence and extract the first k principal components as a subset of material features;

[0143] Since the original probability distribution often contains redundant information and even noise, directly inputting it into subsequent modeling may lead to overfitting and increase computational burden. Therefore, principal component extraction (PCE) can be used to project the original high-dimensional discrete probability vector into a low-dimensional orthogonal subspace, retaining the k most representative directional components with the largest variance contribution as new feature representations. The value of k can be selected based on empirical rules, such as retaining the number of principal components with a cumulative contribution rate of over 90%. This approach not only effectively compresses the data size but also enhances the robustness of the model, preventing erroneous decisions due to minor fluctuations during subsequent fusion.

[0144] Step S304: Decompose the geometric structure feature vector into spatial attribute components and structural stability components;

[0145] The geometric structural features are divided into two components: one is the spatial attribute component reflecting the macroscopic appearance, mainly covering the relative proportions between length, width, and height; the other is the structural stability component reflecting microscopic mechanical properties, including but not limited to the variation in surface curvature and the degree of fluctuation in edge contours. This division helps to break the fuzzy boundary effect brought about by the traditional single geometric description, especially when dealing with materials with similar shapes but completely different appearances. For example, both are cubic in shape, but one is rigid steel and the other is flexible resin. They are almost indistinguishable in appearance, but they have significant differences in their stress response.

[0146] Step S305: Map the standardized deviation to the logarithmic space to generate the quality compensation factor;

[0147] In this method, the absolute value of the original Z-score is increased by one, the common logarithm is calculated, and then multiplied by its sign bit to form a new scalar factor. This factor occupies an independent position in the entire feature space and is used to fill the information gaps that other modes have not fully captured. Choosing a logarithmic function can appropriately increase the dynamic range of the middle segment while maintaining monotonically increasing behavior. This method is particularly suitable for distinguishing target materials that are relatively similar in nature but must be differentiated. For example, lightweight metals such as aluminum alloys and magnesium alloys are very suitable for fine-grained identification using this method.

[0148] Step S306: Merge the material feature subset, spatial attribute component, structural stability component, and mass compensation factor to generate a joint feature vector;

[0149] Step S307: Input the joint feature vectors into the pre-trained random forest sub-model and gradient boosting tree sub-model in parallel;

[0150] The random forest sub-model is an ensemble model composed of multiple CART regression trees, each independently trained on a different sample set. This approach not only mitigates the risk of individual outliers impacting the overall prediction results but also significantly improves the overall accuracy through majority voting. In contrast, the gradient boosting tree sub-model involves layering weak classifiers and continuously correcting residual errors to approximate the optimal solution. Its advantage lies in its ability to uncover highly nonlinear patterns hidden behind numerous low-order interactions.

[0151] In this embodiment, the random forest sub-model consists of 100 CART trees, which increase generalization through bootstrap sampling and are suitable for the rapid identification of common media (such as metals and plastics); the gradient boosting tree sub-model is trained iteratively with 500 weak decision trees and the loss function is optimized by gradient descent, which is good at handling the subtle features of complex media (such as carbon fiber composites).

[0152] Step S308: Determine whether the output types of the random forest sub-model and the gradient boosting tree sub-model are consistent; if yes, proceed to step S309; ​​if no, proceed to step S310.

[0153] Step S309: Directly output the type identifier as the media type determination result;

[0154] Step S310: Activate the convolutional neural network arbitration module to generate a final type identifier as the medium type determination result.

[0155] In this embodiment, the arbitration module is triggered when the Hamming distance between the output type identifiers of the two models is greater than 0 (i.e., the types are inconsistent). The joint feature vector is reshaped into an 8×8 feature matrix, with rows corresponding to feature categories and columns storing numerical values. A three-layer convolutional layer (3×3 kernel, stride 1) captures the spatial correlation between features (such as the co-variation pattern of size ratio and curvature). The output layer Softmax covers the original voting results, resolving misjudgments caused by data noise or feature conflicts. For example, when the features of metal and carbon fiber composite materials are similar, CNN can identify the nonlinear coupling relationship between curvature variance and mass compensation factor.

[0156] Understandably, combining the two sub-models creates a complementary structure: Random Forest focuses on breadth-first search for latent patterns, while Gradient Boosting Trees seek to deeply uncover implicit connections. When the two reach a consensus, the conclusion can be directly adopted; otherwise, a higher-level cognitive engine, namely the convolutional neural network arbitration module, needs to be activated for the final decision.

[0157] In the above implementation, the discriminative information of material probability is preserved based on PCA dimensionality reduction, geometric feature decoupling avoids interference from micro / macro features, and the parallel architecture of random forest and gradient boosting tree achieves efficient and stable classification decision function for complex media, greatly reducing the misclassification rate.

[0158] Reference Figure 4 As one implementation of step S103, the steps of querying a preset parameter mapping table based on medium characteristic data, generating a dynamic compensation coefficient by combining real-time temperature and humidity data collected by environmental sensors, and outputting initial crushing control parameters include:

[0159] Step S401: Receive media feature data containing media type identifier and physical attribute features;

[0160] Step S402: Query the preset parameter mapping table according to the medium type identifier to obtain the basic crushing parameter set; wherein, the basic crushing parameter set includes the reference speed value, the reference pressure value, and the reference duration value;

[0161] The basic crushing parameter set is obtained from a preset parameter mapping table based on media type identifiers (such as "solid-state drive", "paper document", "encrypted USB flash drive casing", etc.), and includes reference speed, reference pressure, and reference duration values ​​obtained through extensive experimental calibration for this type of material. These parameters represent the ideal operating point under standard temperature and humidity conditions (usually 25℃, 50%RH) and serve as the initial reference for the entire control process.

[0162] Specifically, the basic crushing parameter set includes, but is not limited to, the spindle rotation speed (reference speed), the magnitude of the force applied to the material (reference pressure), and the recommended operation duration (reference duration). This categorized data organization allows the control system to quickly respond to the needs of different types of materials, avoiding inefficiencies or even mechanical damage caused by blind settings.

[0163] Step S403: Read the temperature and humidity data collected by the environmental sensor in real time;

[0164] It is understandable that changes in temperature and humidity can significantly alter the brittleness, toughness, and internal stress state of materials, indirectly affecting various aspects such as crushing energy consumption, particle morphology distribution, and even tool wear rate. Therefore, deploying high-precision temperature and humidity sensors to continuously monitor on-site conditions can lay a solid data foundation for implementing precise compensation strategies.

[0165] Step S404: Based on the hardness benchmark value and temperature and humidity data in the physical property characteristics, a dynamic compensation coefficient is generated through a dynamic compensation model; wherein, the dynamic compensation coefficient includes the rotation speed dynamic compensation coefficient and the pressure dynamic compensation coefficient.

[0166] Specifically, the dynamic compensation model can derive a set of optimal adjustment ratio factors by comprehensively considering the current material hardness level, immediate environmental conditions, and historical experience patterns.

[0167] In this embodiment, the dynamic compensation model embeds two core functions: one is a temperature-speed compensation function.

[0168] ;

[0169] One is the humidity-pressure compensation function:

[0170] ;

[0171] Where α and β represent empirical constants indicating the material's sensitivity to external thermal disturbances; while T env and H env This indicates the actual measured temperature and air moisture content; T ref and H ref H is the reference point value. adj These represent the relative correction factors of the workpiece with respect to a specific hardness grade. Using these two nonlinear transformation formulas, the originally static and unchanging basic operating parameters can be transformed into a new generation of adjustable variables that better meet the needs of real-world working conditions.

[0172] It should be noted that the material attenuation factor (α, β) in the dynamic compensation model is updated in the following way: historical operation datasets are received periodically, including actual crushing effect scores and environmental parameters; the attenuation factor mapping relationship is optimized using a gradient descent algorithm.

[0173] In this embodiment, the speed dynamic compensation coefficient and the pressure dynamic compensation coefficient are used to correct for motor efficiency degradation caused by temperature fluctuations and material friction performance degradation caused by humidity changes, respectively. For example, when the ambient temperature rises above 40°C, the motor winding resistance increases, the demagnetization effect of the magnetic material becomes apparent, and the actual output torque decreases. At this time, the speed dynamic compensation coefficient may be greater than 1.0, meaning that the drive signal strength needs to be increased to maintain the original shearing speed. Conversely, in a high-humidity environment, paper absorbs water, softens, and becomes more pliable, requiring increased clamping pressure to prevent slippage, and the corresponding pressure dynamic compensation coefficient is also adjusted upwards. These two compensation coefficients are essentially feedforward corrections to external disturbances, reflecting the control system's proactive adaptability to external uncertainties.

[0174] Step S405: Input the basic crushing parameter set and dynamic compensation coefficient into the parameter optimization engine to calculate the compensation crushing control parameters;

[0175] Specifically, the parameter optimization engine can be viewed as a solver module integrating multiple constraints. It can comprehensively consider multiple objective functions such as minimizing energy consumption and maximizing output, and derive a more refined instruction sequence using the intermediate results obtained above. It should be noted that this process is not a simple multiplication and addition, but involves complex weight allocation and boundary limit judgment logic. For example, if a calculation shows that the recommended pressure is too high and exceeds the safe range, it will be forcibly reduced to within the permissible upper limit, even if it is theoretically reasonable.

[0176] Step S406: Load the compensation crushing control parameters and the three-dimensional model of the medium in the virtual environment, perform the crushing process simulation and generate risk prediction values;

[0177] The simulation process of the crushing process includes: simulating the particle size distribution of fragments based on the discrete element method (DEM); analyzing the deformation risk of the mechanical structure through stress cloud diagrams; and outputting the maximum residual size and the peak stress of the bearing as risk prediction values.

[0178] Specifically, after obtaining a precise geometric shape description file by reverse scanning and reconstructing the actual object to be processed, a highly realistic digital twin platform is built using finite element analysis software. This platform can reconstruct the geometry of the object to be processed based on CAD drawings and, in conjunction with advanced numerical methods such as discrete element method (DEM) and finite element analysis (FEA), reproduce particle motion trajectories, stress distribution, and other aspects. Based on this, the system can predict potential problems in advance, such as crack propagation caused by localized stress concentration and accelerated abnormal tool wear, and provide quantitative risk prediction values, such as the probability of exceeding the maximum residual size limit and whether the bearing peak load is approaching the fatigue limit.

[0179] Understandably, this platform allows users to repeatedly test the potential consequences of various candidate control strategies without activating real hardware. In particular, it enables the quantitative assessment of the probability of events that could lead to severe vibrations, localized overloads, or abnormal overheating, allowing for proactive preventative measures.

[0180] Step S407: When the risk prediction value exceeds the preset risk threshold, adjust the compensation crushing control parameters and re-simulate until the risk prediction value meets the standard, then output the initial crushing control parameters.

[0181] If any safety-related index is detected to exceed the allowable range, it indicates a significant safety hazard in the existing setup, necessitating immediate intervention and a corrective procedure. Typically, the initial fine-tuning involves slightly increasing the compression force within the crushing chamber while appropriately slowing the cutterhead rotation to mitigate the adverse effects of the instantaneous impact. If the initial attempt fails to achieve the desired improvement, the intervention is intensified until the danger signal is completely eliminated. This process is repeated until all warning indicators return to normal ranges, ensuring that the system operates reliably and under control even in complex and harsh working conditions.

[0182] In the above implementation, the three core technical elements of the perception layer, cognition layer, and decision-making layer are deeply integrated, achieving a fundamental shift from passive response to proactive prediction. By integrating information flows from multiple heterogeneous sensors and relying on data analysis and a highly efficient and reliable simulation verification toolchain, the robustness, flexibility, and reliability of the crushing system are improved.

[0183] Reference Figure 5 As one implementation of step S405, the step of calculating the compensated crushing control parameters by combining the basic crushing parameter set with the dynamic compensation coefficient input parameter optimization engine includes:

[0184] Step S501: Receive the basic crushing parameter set and dynamic compensation coefficients; wherein, the basic crushing parameter set includes the reference speed value, the reference pressure value and the reference duration value, and the dynamic compensation coefficients include the speed dynamic compensation coefficient and the pressure dynamic compensation coefficient.

[0185] Step S502: Load the preset equipment safety boundary constraint set, including the maximum permissible speed threshold, pressure tolerance range, and bearing thermal load limit;

[0186] Specifically, the maximum permissible speed threshold is determined by the critical speed of the spindle bearing and the dynamic balance level of the tool. Exceeding this value will cause severe vibration or even resonance damage. The pressure tolerance range is limited by the maximum output force of the hydraulic / pneumatic actuator and the yield strength of the stressed components. Applying excessive clamping force may cause roller deformation or housing cracking. The bearing thermal load limit is a safe upper limit set based on the lubricating oil film rupture temperature and temperature rise rate. Continuous overload will accelerate wear and shorten service life.

[0187] Understandably, these boundary conditions constitute insurmountable "hard constraints" in the parameter optimization process; any candidate solution must fall within this feasible region to be accepted. It is worth noting that these thresholds can be dynamically updated based on the equipment's service life. For example, the bearing thermal load limit of older equipment will gradually decrease with lubrication aging. The system can automatically adjust the constraint boundaries through historical operation and maintenance data analysis, achieving true full lifecycle management.

[0188] Step S503: Multiply the reference speed value by the speed dynamic compensation coefficient to obtain the initial speed candidate value, and multiply the reference pressure value by the pressure dynamic compensation coefficient to obtain the initial pressure candidate value.

[0189] Understandably, since the compensation coefficients are derived under specific assumptions (such as ignoring coupling effects and assuming linear material response), the initial candidate values ​​may have already exceeded the safety boundaries of the equipment, especially under extreme environments or abnormal material conditions. Therefore, the system does not immediately adopt these preliminary results as the final instruction, but instead introduces them as variables to be optimized into subsequent steps for further processing.

[0190] Step S504: Construct a weighted combined objective function of minimizing energy consumption and maximizing output; wherein, the energy consumption weight is associated with the real-time electricity price factor, and the output weight is associated with the physical properties of the medium.

[0191] Specifically, the energy consumption minimization objective function focuses on the relationship between electrical energy consumption and mechanical power consumption per unit time. Its expression is usually in the form of a power integral, i.e., ∫P(t)dt, where power P can be decomposed into the sum of the power consumption of the main motor, the hydraulic pump, and the auxiliary system. The output maximization objective function, on the other hand, measures the effective crushing mass per unit time, often expressed as m / t, which is the total mass of material processed divided by the operating time. Because there is an inherent contradiction between these two, increasing the rotational speed and pressure can increase output, but it leads to a non-linear increase in energy consumption. Therefore, it is not possible to pursue either one to its extreme; a compromise balance must be achieved through weighted allocation.

[0192] In this embodiment, by linking energy consumption weight to real-time electricity price factors, its proportion is automatically increased during peak electricity price periods, prompting the system to prioritize low-power modes. Meanwhile, output weight is associated with the physical properties of the medium. For example, for highly classified media (such as military-grade SSDs), which have a higher destruction priority, the system automatically increases the output weight to ensure complete pulverization in the shortest possible time. Conversely, for high-hardness, difficult-to-break media (such as ceramic-encapsulated chips), the system relatively decreases the output weight to avoid damaging the cutting tools or producing substandard fragments due to excessive pursuit of processing speed. This design, which incorporates both economic efficiency and task importance into the optimization objectives, enables the control system to possess cross-level decision-making intelligence.

[0193] Step S505: Under the constraints of the equipment safety boundary set, the optimal solution of the weighted combined objective function is solved by the sequential quadratic programming algorithm to generate the optimized values ​​of rotational speed and pressure.

[0194] Among them, the sequential quadratic programming algorithm is an iterative algorithm suitable for nonlinear constrained optimization problems, particularly well-suited for engineering scenarios with continuously differentiable objective functions and inequality constraints. In each iteration, Taylor expansion is used to locally linearize or quadratically approximate the original nonlinear problem, transforming the complex global optimization problem into a series of easily solvable quadratic programming subproblems. The solution to each subproblem constitutes the search direction, and the current variable value is updated along this direction until the convergence condition is met.

[0195] In this embodiment, by applying the Sequential Quadratic Programming (SQP) algorithm, not only can the optimal speed and pressure values ​​that satisfy all constraints be found quickly, but sensitivity information such as the Jacobian matrix can also be provided to evaluate the degree of influence of each parameter on the objective function, thus providing support for subsequent fault diagnosis and parameter tuning.

[0196] Step S506: Perform boundary verification on the optimized rotation speed and optimized pressure values, and output the set of compensated crushing control parameters.

[0197] The compensation crushing control parameter set includes the final compensated speed value, pressure value, and reference duration value.

[0198] Specifically, the boundary verification steps include: when the speed optimization value exceeds the maximum allowable speed threshold, resetting the speed optimization value to the threshold and recalculating the pressure optimization value; when the pressure optimization value exceeds the pressure tolerance range, scaling the speed optimization value proportionally to maintain power conservation.

[0199] In this embodiment, when the optimized rotational speed value exceeds the maximum allowable rotational speed threshold, the system does not simply discard the solution. Instead, it adopts a "reset-rebalancing" strategy: forcibly reducing the rotational speed to the upper limit of the threshold and recalculating the matching optimized pressure value to maintain a relatively constant overall power demand and avoid a sudden drop in processing capacity due to unilateral reduction. Similarly, when the optimized pressure value exceeds the pressure tolerance range, the system does not simply truncate the system. Instead, it adjusts the rotational speed in reverse according to the power conservation principle, that is, moderately reducing the rotational speed to compensate for the excessive pressure demand, so that the total input energy remains within a reasonable range. This linkage adjustment mechanism fully considers the energy conservation characteristics and dynamic coupling relationship of the mechanical system, avoiding the performance loss and control instability problems caused by traditional one-size-fits-all limiting.

[0200] In the above implementation, an intelligent optimization architecture with constraint-solving capabilities is constructed. While ensuring system operational safety, it achieves refined reconstruction of crushing process parameters, thereby maintaining efficient, energy-saving, and stable crushing performance even under highly variable actual operating conditions. More importantly, through the synergistic effect of sequential quadratic programming and boundary rebalancing mechanisms, it ensures that even under extreme conditions, it can output control parameters that are both efficient and safe, greatly improving the reliability, energy efficiency ratio, and intelligence level of crushing operations. This is particularly suitable for high-end destruction scenarios with stringent requirements for information security, equipment lifespan, and operating costs.

[0201] Reference Figure 6 As one implementation of step S505, the step of generating optimized speed and pressure values ​​by solving the optimal solution of the weighted combined objective function through a sequential quadratic programming algorithm under the constraints of the equipment safety boundary set includes:

[0202] Step S601: Obtain initial speed candidate values, initial pressure candidate values, and weighted combined objective function;

[0203] Step S602: Construct the joint constraint equation of rotational speed and pressure based on the equipment safety boundary constraint set;

[0204] The system constructs a joint speed-pressure constraint equation based on the equipment safety boundary constraint set, which is the fundamental guarantee for ensuring that the optimization results are executable and implementable. This constraint set not only includes the maximum allowable speed threshold (usually determined by the critical speed of the spindle bearing and the dynamic balance level of the cutterhead) and the pressure tolerance range (rated output force of the hydraulic cylinder and pressure resistance limit of the seals), but also introduces a key nonlinear thermodynamic constraint: the bearing thermal load limit. This limit cannot be simply expressed as an upper limit of an independent variable, but has a strong coupling relationship with speed and pressure.

[0205] Specifically, the frictional heat power of the bearing is approximately proportional to the product of the rotational speed n and the load pressure p. Therefore, the system transforms this into a nonlinear inequality constraint to prevent lubricating oil film rupture, ball annealing, or even seizure due to prolonged high-load operation. Furthermore, the pressure itself must also satisfy the double boundary constraint p. min ≤p≤p max To ensure clamping stability and actuator safety; the rotational speed is limited by n≤n max This is to avoid severe vibrations caused by the resonance zone.

[0206] Step S603: Based on the sequential quadratic programming algorithm, the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure are used as the starting point for optimization. An iterative optimization loop is executed, and the current optimization variables of rotational speed and pressure are output when the optimal solution is reached.

[0207] Among them, the sequential quadratic programming algorithm is essentially an efficient numerical method for solving nonlinear constrained optimization problems, and it is particularly suitable for engineering scenarios with smooth objective functions and complex constraints.

[0208] Step S604: The current speed optimization variable and pressure optimization variable are used as the speed optimization value and pressure optimization value, respectively.

[0209] If the iteration count reaches the preset upper limit and the iteration does not terminate, it degenerates into using the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure as the optimized values ​​of rotational speed and pressure.

[0210] Specifically, when the algorithm successfully converges, the system outputs the current speed optimization variable and pressure optimization variable as the speed optimization value and pressure optimization value, respectively, forming a set of optimal control commands that conform to physical laws and take into account economic benefits.

[0211] However, under extreme conditions (such as encountering unknown ultra-high hardness composite materials, abnormal sensor signals, or multiple constraint conflicts), the iterative process may fail to converge for a long time. To address this, the system presets a maximum upper limit for the number of iterations (usually set to 30-50 times, depending on the controller's computing power). Once this upper limit is reached and the termination condition is still not met, a "degradation mechanism" is triggered, i.e., further searching is abandoned, and the initial candidate values ​​for rotational speed and initial candidate values ​​for pressure are directly adopted as the final optimized values ​​for rotational speed and pressure.

[0212] This mechanism reflects the high importance placed on reliability and real-time performance in engineering practice: on the one hand, it prevents the control algorithm from falling into an infinite loop or oscillation state, leading to delayed instruction issuance and affecting the overall machine response performance; on the other hand, since the initial candidate values ​​have already been corrected by the dynamic compensation model and include prior knowledge of environmental and material properties, even if the global optimum is not reached, it is sufficient to ensure basic crushing effectiveness and equipment safety. For example, in the event of a sudden thunderstorm causing a sharp increase in humidity, if the optimization algorithm fails to converge due to strong nonlinear response, the degradation mechanism can ensure that the equipment can continue to operate with the compensated robust parameters, avoiding the risk of downtime.

[0213] In the above implementation, starting with the compensated initial candidate values, the optimal solution of the weighted objective function is searched within a strict safety boundary using a sequential quadratic programming algorithm. Efficient iteration is achieved through the sequential quadratic programming algorithm, supplemented by a degradation mechanism to ensure system robustness. This technical solution not only solves the problem of balancing energy consumption and efficiency in traditional crushing control, but also incorporates latent failure risks such as bearing thermal load into the explicit modeling scope, improving the long-term reliability of the equipment.

[0214] Reference Figure 7 As one implementation of step S603, based on the sequential quadratic programming algorithm, using the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure as the starting point for optimization, executing an iterative optimization loop, and outputting the current optimized variables of rotational speed and pressure when the optimal solution is reached, the steps include:

[0215] Step S701: Calculate the gradient vector of the weighted combined objective function at the current iteration point;

[0216] At the beginning of each iteration, the system calculates the gradient vector of the objective function at the current iteration point. This objective function is a weighted combination of two sub-objectives: minimizing energy consumption and maximizing output. Its gradient vector reflects the local sensitivity of the system's performance indicators to changes in rotational speed and pressure.

[0217] Specifically, each component in the gradient vector represents the rate of change of the overall target value caused by a unit change in rotational speed or a unit change in pressure. Since motor power typically increases approximately cubically with rotational speed (i.e., high rotational speed leads to a significant increase in power consumption), the gradient component in the rotational speed direction is often large, reflecting the system's high sensitivity to speed regulation. The impact of pressure on output depends on the mechanical properties of the medium itself. For high-hardness, dense materials (such as metal-encapsulated chips or ceramic substrates), a greater positive pressure is required to achieve effective crushing, in which case the gradient component in the pressure direction is more prominent.

[0218] Understandably, this gradient information not only reveals the weaknesses of the system under the current parameter configuration, but also provides clear mathematical guidance for the selection of subsequent search directions. For example, when the gradient shows that reducing the rotational speed can significantly reduce energy consumption while only slightly affecting output, the optimization path will tend to prioritize reducing the rotational speed to achieve the energy-saving goal, thereby ensuring that the entire optimization process always proceeds in the most efficient direction.

[0219] Step S702: Update the approximate value of the Hessian matrix using the quasi-Newton method to generate the quadratic term coefficient matrix of the quadratic programming subproblem;

[0220] The Hessian matrix is ​​essentially the second derivative matrix of the objective function at the current point, describing the local curvature characteristics of the target surface, i.e., the influence trend of parameter changes on the rate of change of the target. Directly and accurately calculating this matrix is ​​costly and difficult to implement in engineering applications. Therefore, this application introduces the quasi-Newton method, an efficient approximation strategy, to gradually correct the estimated value of the Hessian matrix by using the gradient difference and variable displacement difference between two iterations.

[0221] Specifically, the quasi-Newton method ensures that the constructed approximate matrix always remains positive definite, thus guaranteeing that the search direction obtained each time is the descent direction of the objective function. More importantly, this matrix implicitly captures the dynamic inertial characteristics of the crushing system and the nonlinear response behavior of the actuator: for example, when the speed increases rapidly, the motor exhibits torque ramp-up delay; the hydraulic system exhibits stiffness softening under high pressure. Although these complex physical effects are not explicitly modeled, they are indirectly reflected in the Hessian matrix update process through historical gradient information, giving the optimization model a certain "memory ability" and real-world adaptability.

[0222] Step S703: Linearize the joint constraint equation of rotational speed and pressure at the current iteration point to generate a set of linear constraint equations;

[0223] This step is a crucial bridge connecting the original nonlinear physical constraints with the computable mathematical model. Taking the bearing thermal load limit as an example, its original form is that the product of speed and pressure does not exceed a certain threshold, which is a typical nonlinear inequality constraint. Without processing, traditional optimization algorithms struggle to directly address such complex boundaries.

[0224] To address this, the system performs a first-order Taylor expansion of the constraint near the current iteration point, transforming it into a linear inequality concerning the speed and pressure increments. This is equivalent to locally approximating the original surface boundary with a tangent plane. Similarly, the maximum permissible speed and pressure tolerance ranges are also converted into standard linear or box-type constraints. This linearization strategy not only significantly reduces the difficulty of solving the problem but also allows for dynamic adjustment of the safety boundary position based on the latest parameters in each iteration, enabling real-time mapping and precise control of equipment operation risks. Especially when dealing with sudden abnormal materials (such as the introduction of high-density metallic impurities) leading to accelerated temperature rise, this mechanism can quickly tighten the operating window to prevent the system from entering a dangerous zone.

[0225] Step S704: Based on the quadratic term coefficient matrix, the objective function gradient vector, and the linear constraint equations, solve the quadratic programming subproblem to obtain the search direction vector;

[0226] The core of this quadratic programming subproblem lies in finding a parameter adjustment direction that maximizes the improvement of the objective function while satisfying all linear constraints. The solution process typically employs the effective set method, which involves identifying the currently active constraints (such as a pressure reaching its limit) and transforming the problem into a standard quadratic programming problem with equality constraints. The optimal search direction is then derived by solving the Karush-Kuhn-Tucker (KKT) equations. This direction vector not only contains numerical information but also determines how the speed and pressure should be adjusted in tandem. For example, if the results indicate that the speed should be reduced while the pressure is increased, it means that the current system is in a high-energy-consumption and low-efficiency range, and the processing capacity loss caused by the speed reduction needs to be compensated by increasing the shear force. This multi-variable linkage adjustment mechanism demonstrates the control system's deep understanding of complex operating conditions and its coordinated control capabilities, far exceeding the traditional mode of independent adjustment of a single parameter.

[0227] Step S705: Update the rotation speed optimization variable and pressure optimization variable along the search direction vector to generate a new iteration point;

[0228] Specifically, the update process does not simply proceed with the full step size, but introduces a damped step size control mechanism to prevent out-of-bounds errors or oscillations caused by excessively large step sizes. The initial step size is usually set to a moderate value (e.g., 0.5). If the newly generated parameter combination violates any safety constraints, a backtracking search procedure is triggered, gradually reducing the step size until a feasible new point is found. This approach ensures both convergence speed and avoids the risk of optimization failure due to blind jumps.

[0229] Step S706: When the change in the objective function value between the new iteration point and the previous iteration point is less than the dynamic convergence threshold, it is determined that the optimal solution has been reached, the loop is terminated, and the current speed optimization variable and pressure optimization variable are output.

[0230] The system continuously monitors the change in the objective function value between two consecutive iterations and compares it with a dynamically set convergence threshold. This threshold is not a fixed constant but is adaptively adjusted according to the magnitude of the initial objective function to ensure that it can reasonably determine whether the system is approaching stability under different scales and load conditions. When the change in the objective function between two consecutive iterations is less than this threshold, the system determines that a local optimum has been reached, terminates the loop, and outputs the current speed optimization variables and pressure optimization variables as the final result.

[0231] The above embodiments achieve refined optimization of crushing control parameters under multiple objectives and constraints, forming a closed-loop, self-consistent, and highly intelligent decision-making chain. This technical solution not only considers the rigid limitations of the equipment, such as mechanical strength, thermal stability, and hydraulic load-bearing capacity, but also incorporates external economic and process factors such as real-time electricity price fluctuations and differences in material physical properties, enabling the control system to automatically balance the contradiction between energy consumption costs and processing efficiency while ensuring safety.

[0232] The aforementioned technical solution, as the core optimization link in the crusher control method based on destruction target identification in this application, realizes a fundamental shift from empirical parameter setting to mathematically driven intelligent parameter tuning, playing a crucial role and having practical significance. By introducing a sequential quadratic programming algorithm, it refines the initial candidate parameters while comprehensively considering multiple factors such as energy consumption, output, equipment safety, and environmental disturbances, ensuring that the crushing process reaches its optimal operating state without exceeding mechanical strength, thermal load, and hydraulic limits. When optimization fails to converge due to extreme operating conditions, a degradation mechanism ensures that the system can still output robust parameters after environmental and material compensation, balancing intelligence and reliability. This mechanism not only significantly improves the safety, energy efficiency ratio, and processing quality consistency of the crushing operation but also enables the system to adapt to complex and variable materials and external conditions. It is a key support for achieving closed-loop intelligent control throughout the entire process and has significant engineering value for improving the thoroughness, traceability, and long-term stable operation of classified materials.

[0233] Understandably, while traditional physical shredding boasts high processing efficiency, it may leave some high-density, high-strength, or structurally complex electronic storage media (such as multi-layered metal-encapsulated chips in solid-state drives and embedded circuit boards inside encrypted USB flash drives) with areas that are not fully broken down, particularly exhibiting residual aggregation of metal components. In such cases, mechanical shearing alone is insufficient for complete destruction, necessitating the introduction of higher-precision, higher-energy-density auxiliary methods, namely laser-based refocusing shredding technology, for secondary destruction.

[0234] Based on this, refer to Figure 8As a further implementation of the shredder control method, after the step of verifying the compliance of destruction based on the residue distribution data and outputting a verification report, the method further includes:

[0235] Step S801: When the destruction compliance verification fails, obtain the coordinate set of metal accumulation areas in the residue distribution data;

[0236] This technical solution does not simply identify which locations contain metal elements. Instead, it uses multi-band spectral analysis techniques (such as Raman spectroscopy for identifying lattice vibration modes and near-infrared absorption spectroscopy for retrieving free electron concentration) to distinguish between organic residues and inorganic metal particles. Combined with a spatial point cloud registration algorithm, it maps the detected metal signals back to a three-dimensional physical coordinate system, forming a set of discrete metal mass points with clearly defined spatial locations. These coordinate points reflect the core of the information carrier that remains after the material is crushed, and are key to potential leaks.

[0237] Step S802: Generate a three-dimensional geometric thermal map based on the coordinate set of the metal accumulation area; wherein, the thermal value represents the local metal mass density.

[0238] This step is essentially a spatial kernel density estimation (KDE) process. By applying a Gaussian kernel function to each metal coordinate point and performing three-dimensional convolution, a continuous spatial field function is generated, the magnitude of which reflects the concentration of metallic material within a certain region. Compared to simple clustering, heatmaps can more realistically simulate the non-uniform distribution characteristics of materials in space.

[0239] Step S803: Locate the thermal peak point in the three-dimensional geometric heat map, and use the thermal peak point as the seed point to execute the region growing algorithm to extract the coordinate set of the isosurface boundary.

[0240] The thermal peak points, or local maxima, represent the centers of the most severe metal accumulation in the current residual system, typically corresponding to the most structurally robust parts of the original equipment. These thermal peak points are then used as "seed points" to execute a region growing algorithm. Starting from the seed points, the algorithm gradually expands outwards into the neighborhood based on a set similarity criterion (in this case, a thermal value difference threshold) until the thermal gradient at the boundary significantly decreases, thereby extracting a complete set of isosurface boundary coordinates. This boundary is not a simple bounding box or spherical region, but a complex curved surface contour that conforms to the actual metal cluster morphology, ensuring that the subsequent processing range is neither over-expanded nor misses key areas. This gradient-based dynamic boundary extraction method effectively avoids the errors caused by manually setting a fixed-size window, improving the accuracy of spatial focusing.

[0241] Step S804: Construct a Delaunay triangular mesh based on the isosurface boundary coordinate set, and calculate the curvature distribution characteristics of the mesh vertices;

[0242] Specifically, Delaunay triangulation, as an optimal spatial partitioning strategy, can generate a stable mesh structure that maximizes angles and avoids elongated triangles given a set of vertices, greatly improving the numerical stability of subsequent geometric calculations. Based on this mesh, the system calculates the curvature distribution characteristics at each vertex, including two indices: Gaussian curvature and mean curvature. The former reflects whether the local surface exhibits spherical convexity or saddle-shaped concavity, while the latter characterizes the overall strength of the surface curvature trend.

[0243] For metallic residues, areas with high curvature often correspond to stress concentration points, such as sharp corners, crack tips, or folded edges. These locations, due to abrupt geometric changes, are prone to localized plastic deformation or even fracture failure under stress. Therefore, curvature analysis is not only a tool for morphological description but also an important mechanical criterion for predicting weak points in materials.

[0244] Step S805: Identify high stress concentration sub-regions based on curvature distribution characteristics, and generate a laser energy mapping table by combining the preset metal material yield strength database;

[0245] Among these features, high stress concentration sub-regions are identified by marking mesh segments with curvature exceeding a preset threshold as potentially vulnerable areas. A preset metal material yield strength database stores key mechanical parameters of common electronic components' metal materials (such as copper, aluminum, gold wire, nickel-iron alloys, etc.), including elastic modulus, Poisson's ratio, yield strength, and tensile strength.

[0246] By associating the spatial location of high-curvature regions with their corresponding material types (obtainable through prior RFID tagging or spectral composition analysis), the system can calculate the critical conditions for plastic yielding or melting vaporization in these regions under different external energy inputs. This generates a "laser energy mapping table," a control command matrix that maps spatial coordinates to parameters such as required laser power density, pulse width, and repetition frequency. This mapping is not a uniform, constant output; instead, it employs a gradient configuration based on the curvature of the regions and differences in the heat resistance of the materials. For example, lower-energy, short pulses are used for fine peeling in regions with high curvature and low melting points, while high-energy, long pulses are used for rapid penetration in large, thick metal areas. This differentiated energy supply strategy balances processing efficiency and safety, preventing splashing or release of harmful gases due to excess energy.

[0247] Step S806: Plan the laser focusing path according to the laser energy mapping table and send the laser crushing control command to control the laser to perform gradient energy output;

[0248] The path planning can comprehensively consider the kinematic constraints of the laser head (such as acceleration limits and turning radius), the thermal accumulation effect between adjacent action points, and the overall scanning coverage. It uses a method similar to the Traveling Salesman Problem (TSP) optimization to generate a spatiotemporally optimal trajectory, ensuring that all areas to be processed are fully covered and the switching time is minimized.

[0249] Simultaneously, the system sends a timestamped laser crushing control command to the laser controller, triggering it to execute gradient energy output according to a predetermined sequence. During this process, the laser can fine-tune the focal position in real time based on feedback closed loop (via the Z-axis autofocus module) to adapt to changes in the height of irregular surfaces, ensuring that the energy focusing accuracy remains at the sub-millimeter level.

[0250] Step S807: Re-execute the spectral scan to generate updated residue distribution data, and repeat the above steps until the residue distribution data passes the destruction compliance verification.

[0251] After completing one round of laser refocusing, the system immediately restarts the spectral scanning process, collects updated residue distribution data, and verifies the compliance of the destruction process again. If it still fails to meet the standards, the entire process is repeated. This closed-loop iterative mechanism gives the system strong fault tolerance and ultimate assurance capabilities, enabling it to gradually approach complete destruction even when facing extremely complex or exceptionally stubborn targets through multiple progressive processes.

[0252] In the above embodiments, a deep destruction enhancement system was constructed when conventional mechanical shredding could not meet safety standards. By converting spectral information into a computable spatial field and mapping geometric features into controllable energy input, the reliability of destroying high-value sensitive information carriers was significantly improved. This system is particularly suitable for application scenarios with extremely high information security requirements, such as government agencies, military units, and data centers.

[0253] This application also discloses a shredder control system based on the identification of destruction targets.

[0254] A shredder control system based on target identification for destruction includes:

[0255] The data acquisition module is used to acquire multimodal sensing data of the target object through a multi-sensor array; the multimodal sensing data includes optical images, mass measurements, electronic tag information, and a three-dimensional point cloud coordinate set.

[0256] The medium feature generation module is used to identify the medium type of a target object based on multimodal sensing data, extract physical property features, and generate medium feature data including a medium type identifier and physical property features.

[0257] The dynamic compensation module is used to query the preset parameter mapping table based on the medium characteristic data, combine the temperature and humidity data collected in real time by the environmental sensor to generate dynamic compensation coefficients, and output the initial crushing control parameters.

[0258] The crushing control module is used to drive the crushing actuator according to the initial crushing control parameters and monitor the mechanical resistance value in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold, the initial crushing control parameters are adjusted in real time.

[0259] The residue verification module is used to perform spectral scanning on the residue after pulverization, generate residue distribution data, verify the compliance of destruction based on the residue distribution data, and output a verification report.

[0260] The update feedback module is used to store media characteristic data, crushing control parameters and residue distribution data into the training set, and refit the mapping relationship between media characteristic data and crushing control parameters based on the training set, so as to update the preset parameter mapping table periodically.

[0261] The crusher control system based on destruction target identification in this application embodiment can implement any of the above-mentioned crusher control methods, and the specific working process of each module in the crusher control system can refer to the corresponding process in the above-mentioned method embodiment.

[0262] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0263] This application also discloses a computer-readable storage medium.

[0264] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the methods of a shredder control method based on destruction target identification.

[0265] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0266] In this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0267] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A shredder control method based on target identification for destruction, characterized in that, The control method includes: Multimodal sensing data of the target object is acquired through a multi-sensor array; wherein, the multimodal sensing data includes optical images, mass measurements, electronic tag information, and a three-dimensional point cloud coordinate set; Based on the multimodal sensing data, the medium type of the target object is identified, and physical property features are extracted to generate medium feature data including a medium type identifier and physical property features. Based on the medium characteristic data, a preset parameter mapping table is queried, and dynamic compensation coefficients are generated by combining the temperature and humidity data collected in real time by the environmental sensor, and the initial crushing control parameters are output. The crushing actuator is driven according to the initial crushing control parameters, and the mechanical resistance value is monitored in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold, the initial crushing control parameters are adjusted in real time. The spectral scan of the pulverized residue is performed to generate residue distribution data. The compliance of the disposal is verified based on the residue distribution data, and a verification report is generated. The media characteristic data, initial crushing control parameters, and residue distribution data are stored in the training set. Based on the training set, the mapping relationship between the media characteristic data and the initial crushing control parameters is refitted to periodically update the preset parameter mapping table. The steps of identifying the medium type of the target object based on the multimodal sensing data, extracting physical property features, and generating medium feature data including a medium type identifier and physical property features include: Receives multimodal sensing data acquired by a multi-sensor array; The optical image is subjected to illumination compensation processing to generate standardized image data, and the electronic tag information is parsed to obtain the pre-stored medium identification code. The standardized image data is input into a pre-trained surface material classification model, which outputs material category probability distribution data and calculates geometric structure feature vectors by combining the three-dimensional point cloud coordinate set. By integrating the probability distribution data of the material category, the geometric structure feature vector, and the quality measurement value, a medium type determination result is generated through a decision tree ensemble algorithm; The physical property database is indexed based on the medium type determination result, and the physical property features of the corresponding medium are extracted; wherein, the physical property features include density threshold range and hardness benchmark value; The geometric structure feature vector is combined to generate medium feature data containing medium type identifier and physical property features; The steps of querying a preset parameter mapping table based on the medium characteristic data, generating a dynamic compensation coefficient by combining the real-time temperature and humidity data collected by the environmental sensor, and outputting the initial crushing control parameters include: Receive media characteristic data containing media type identifier and physical attribute characteristics; The basic crushing parameter set is obtained by querying the preset parameter mapping table according to the medium type identifier; wherein, the basic crushing parameter set includes the reference speed value, the reference pressure value, and the reference duration value; Real-time reading of temperature and humidity data collected by environmental sensors; Based on the hardness benchmark value in the physical property characteristics and the temperature and humidity data, a dynamic compensation coefficient is generated through a dynamic compensation model; wherein, the dynamic compensation coefficient includes a rotational speed dynamic compensation coefficient and a pressure dynamic compensation coefficient. The basic crushing parameter set is combined with the dynamic compensation coefficient input parameter optimization engine to calculate the compensation crushing control parameter set; Load the set of compensated crushing control parameters and the three-dimensional model of the medium into a virtual environment, perform a crushing process simulation, and generate risk prediction values. When the predicted risk value exceeds the preset risk threshold, the compensation crushing control parameter set is adjusted and the simulation is repeated until the predicted risk value meets the target, at which point the initial crushing control parameters are output.

2. The shredder control method based on target identification for destruction according to claim 1, characterized in that, The steps for generating a media type determination result by integrating the material category probability distribution data, geometric structure feature vectors, and quality measurement values ​​using a decision tree ensemble algorithm include: Receive material category probability distribution data, geometric structure feature vectors, and mass measurement values; wherein, the material category probability distribution data contains probability value sequences of at least three materials; Calculate the standardized deviation between the measured quality value and the preset medium quality benchmark; Principal component dimensionality reduction is performed on the probability value sequence, and the top k principal components are extracted as a subset of material features; The geometric feature vector is decomposed into spatial attribute components and structural stability components; The standardized deviation is mapped to logarithmic space to generate a quality compensation factor. The material feature subset, spatial attribute component, structural stability component, and mass compensation factor are combined to generate a joint feature vector; The joint feature vectors are input in parallel into the pre-trained random forest sub-model and gradient boosting tree sub-model; Determine whether the output types of the random forest sub-model and the gradient boosting tree sub-model are consistent; if so, directly output the type identifier as the medium type determination result; if not, activate the convolutional neural network arbitration module to generate the final type identifier as the medium type determination result.

3. The shredder control method based on target identification for destruction according to claim 1, characterized in that, The steps of calculating the compensated crushing control parameter set by combining the basic crushing parameter set with the dynamic compensation coefficient input parameter optimization engine include: Receive the basic crushing parameter set and dynamic compensation coefficient; Load the preset equipment safety boundary constraint set, including the maximum permissible speed threshold, pressure tolerance range, and bearing thermal load limit; The reference speed value is multiplied by the speed dynamic compensation coefficient to obtain the initial speed candidate value, and the reference pressure value is multiplied by the pressure dynamic compensation coefficient to obtain the initial pressure candidate value. Construct a weighted combined objective function of minimizing energy consumption and maximizing output; where the energy consumption weight is associated with the real-time electricity price factor, and the output weight is associated with the physical properties of the medium. Under the constraints of the equipment safety boundary set, the optimal solution of the weighted combined objective function is solved by a sequential quadratic programming algorithm to generate optimized values ​​for rotational speed and pressure. Boundary verification is performed on the optimized rotation speed and optimized pressure values, and a set of compensated crushing control parameters is output.

4. The shredder control method based on target identification for destruction according to claim 3, characterized in that, Under the constraints of the equipment safety boundary set, the steps of solving the optimal solution of the weighted combined objective function using a sequential quadratic programming algorithm to generate the optimized values ​​of rotational speed and pressure include: Obtain the initial speed candidate value, the initial pressure candidate value, and the weighted combined objective function; Construct a joint speed-pressure constraint equation based on the equipment safety boundary constraint set; Based on the sequential quadratic programming algorithm, the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure are used as the starting point for optimization. An iterative optimization loop is executed, and the current optimization variables of rotational speed and pressure are output when the optimal solution is reached. The current speed optimization variable and pressure optimization variable are respectively used as the speed optimization value and pressure optimization value.

5. A shredder control method based on target identification for destruction according to claim 4, characterized in that, Based on the sequential quadratic programming algorithm, using the initial candidate values ​​of rotational speed and initial candidate values ​​of pressure as the starting point for optimization, the steps of executing an iterative optimization loop and outputting the current optimized variables of rotational speed and pressure when the optimal solution is reached include: Calculate the gradient vector of the weighted combined objective function at the current iteration point; The approximation of the Hessian matrix is ​​updated using the quasi-Newton method, generating the quadratic coefficient matrix of the quadratic programming subproblem. Linearize the combined speed-pressure constraint equation at the current iteration point to generate a set of linear constraint equations; Based on the quadratic term coefficient matrix, the objective function gradient vector, and the linear constraint equations, the quadratic programming subproblem is solved to obtain the search direction vector. Update the rotation speed optimization variables and pressure optimization variables along the search direction vector to generate a new iteration point; When the change in the objective function value between the new iteration point and the previous iteration point is less than the dynamic convergence threshold, the optimal solution is determined to have been reached, the loop is terminated, and the current speed optimization variable and pressure optimization variable are output.

6. A shredder control method based on destruction target identification according to any one of claims 1 to 5, characterized in that, After verifying the compliance of the disposal based on the residue distribution data and generating a verification report, the following steps are also included: When the destruction compliance verification fails, obtain the coordinate set of metal accumulation areas from the residue distribution data; A three-dimensional geometric thermal map is generated based on the coordinate set of the metal accumulation region; wherein, the thermal value represents the local metal mass density; In the three-dimensional geometric heat map, locate the thermal peak point, and use the thermal peak point as the seed point to execute the region growing algorithm to extract the isosurface boundary coordinate set; Construct a Delaunay triangular mesh based on the isosurface boundary coordinate set, and calculate the curvature distribution characteristics of the mesh vertices; Based on the curvature distribution characteristics, high stress concentration sub-regions are identified, and a laser energy mapping table is generated by combining a preset metal material yield strength database. The laser focusing path is planned according to the laser energy mapping table, and a laser crushing control command is sent to control the laser to perform gradient energy output. Re-execute the spectral scan to generate updated residue distribution data, and repeat the above steps until the residue distribution data passes the destruction compliance verification.

7. A shredder control system based on target identification for destruction, characterized in that, A shredder control method based on destruction target identification as described in any one of claims 1 to 6, the control system comprising: The data acquisition module is used to acquire multimodal sensing data of the target object through a multi-sensor array; wherein, the multimodal sensing data includes optical images, mass measurement values, electronic tag information, and a three-dimensional point cloud coordinate set; The medium feature generation module is used to identify the medium type of the target object based on the multimodal sensing data, extract physical attribute features, and generate medium feature data including a medium type identifier and physical attribute features. The dynamic compensation module is used to query a preset parameter mapping table based on the medium characteristic data, generate a dynamic compensation coefficient by combining the temperature and humidity data collected in real time by the environmental sensor, and output the initial crushing control parameters. The crushing control module is used to drive the crushing actuator according to the initial crushing control parameters, and monitor the mechanical resistance value in real time during the crushing process. When the monitored mechanical resistance value deviates from the expected value by more than a dynamic threshold, the initial crushing control parameters are adjusted in real time. The residue verification module is used to perform spectral scanning on the residue after pulverization, generate residue distribution data, verify the compliance of destruction based on the residue distribution data, and output a verification report. The update feedback module is used to store the medium characteristic data, initial crushing control parameters and residue distribution data into the training set, and refit the mapping relationship between the medium characteristic data and the initial crushing control parameters based on the training set, so as to periodically update the preset parameter mapping table.

8. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 6.

Images