A design platform and method for a high-toughness vertical shaft cutting head

By incorporating modules for anomaly identification, risk prediction, and fault-tolerant recovery, the problems of difficult parameter anomaly identification and unpredictable simulation risks in the design of vertical shaft cutting heads have been solved, thus achieving an efficient and reliable design verification process.

CN123116879APending Publication Date: 2026-11-17ZHEJIANG UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611612104.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing automated design process for vertical shaft cutting heads lacks an effective pre-verification and risk prediction mechanism, making it difficult to identify abnormal parameters in a timely manner, and simulation risks cannot be predicted in advance. After an error occurs, the entire process needs to be restarted, affecting the efficiency and stability of design verification.

Method used

It employs an anomaly identification module, an anomaly risk prediction module, a recovery and fault tolerance module, and a data template storage module. Through neural networks, it performs parameter verification, risk prediction, and fault tolerance recovery, thereby enabling the identification and rapid repair of potential anomalies.

Benefits of technology

It improves the efficiency and stability of design verification, reduces the cost of repeated trial and error, ensures the continuity and reliability of the design process, and is suitable for rapid modeling and simulation of vertical shaft cutting heads under multiple working conditions and parameters.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116879A_ABST
    Figure CN123116879A_ABST
Patent Text Reader

Abstract

This invention discloses a high-toughness shaft cutting head design platform and method, belonging to the field of underground coal mine transportation technology. The design platform includes an anomaly identification module, an anomaly risk prediction module, a recovery and fault tolerance module, and a data template storage module. The anomaly identification module is used to verify and correct the input parameters for modeling and simulation; the anomaly risk prediction module is used to predict mesh quality before mesh generation, and to predict simulation error risks and simulation performance index trends before simulation calculation; the recovery and fault tolerance module is used to roll back to the previous historical feasible parameters or the previous error-free step when an error occurs, or to adjust the parameters; the data template storage module is used to classify, store, and manage parameters at each stage, providing data support for retrieval, historical traceability, and model updates. This invention significantly improves the reliability and controllability of the modeling and simulation process, and is suitable for the rapid development and performance evaluation of shaft cutting head structures under multiple working conditions and parameters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of underground coal mine transportation equipment technology, specifically to a high-toughness vertical shaft cutting head design platform and method. Background Technology

[0002] The current structural design field employs a "parametric script + external simulation software" approach. In the design of vertical shaft cutting heads, this means: the script automatically generates geometric models based on user input (automatic modeling and automatic network generation), and then batches the models and boundary conditions to solvers such as CFD / DEM / finite element methods for calculation (automatic simulation). After simulation, key indicators such as slag volume, velocity / pressure field, and particle throughput are summarized for engineers to compare schemes and perform preliminary optimization (data analysis). This type of process typically relies on the secondary development interfaces (API / macro commands / batch scripts) of commercial software to form a streamlined operation with no or weak interface, enabling batch calculations and parametric studies without changing the original commercial software.

[0003] In actual implementation, the script typically completes operations in a predetermined order: assigning geometric parameters, updating component assembly positions, setting mesh strategies and element sizes, arranging fluid inlets and particle sources, importing boundary and load conditions, setting solution control parameters, and exporting result files. This workflow allows designers to complete simulation comparisons of multiple cutting head structure schemes in a short time, offering a degree of automation and reusability, making it the most advanced design verification path currently available.

[0004] However, the automated design and simulation of shaft cutting heads is not a simple parametric verification of general mechanical structures. Their models typically include multiple complex components such as the cutting head, slag bin, slag discharge port, and flow guiding structure, exhibiting irregular geometric boundaries, numerous small local features, and complex surface intersections. During automatic modeling and mesh generation, parameter variations can easily cause distortion in thin-walled regions, excessively small local gaps, surface self-intersections, and component interference, leading to mesh generation failures, incorrect contact surface identification, or abnormal computational domain generation. Simultaneously, the simulation of slag discharge performance often involves solid-liquid coupling, particle flow, and complex boundary condition coupling. Slight deviations in the settings of the inlet, outlet, particle source, rotating region, and slag discharge channel can result in non-convergence, abnormal particle accumulation, flow field divergence, or inability to output the correct result file. Therefore, compared to the automated simulation process of ordinary high-toughness mechanical equipment, shaft cutting head design platforms require higher robustness for complex irregular geometry, multi-physics coupled simulation, and automatic recovery from abnormal operating conditions.

[0005] Therefore, the above design verification path has the following shortcomings: First, the automatic modeling and simulation stages lack pre-verification of "physical realizability." In engineering practice, automated modeling and simulation often encounter problems such as spatial interference of components, dimensional out-of-bounds issues, unreasonable inlet / outlet settings, and particle sources deviating from the effective computational domain. Some of these problems will trigger solver errors, requiring the entire process to be reset and executed, severely impacting computational efficiency; some problems may not even trigger explicit errors at the script level, but due to geometric unmanufacturability, unrealizable boundary conditions, and other issues, the results will be severely distorted, ultimately leading to situations where the program appears normal but is not feasible in engineering. These problems waste computational resources and delay optimization time.

[0006] Secondly, before modeling or simulation is completed, users struggle to promptly discern the trend of results and the rationality of parameters, lacking the ability to "predict" potential divergences, poor quality, or substandard performance. Furthermore, because the entire process is coupled with external finite element / CFD software through hidden calls, if an anomaly occurs at any step, users cannot intervene in detail or troubleshoot breakpoints on the solver interface; they must revert to the beginning of the process, reconfigure parameters, and recalculate. These problems result in two main issues: firstly, a lack of rapid recovery of historical inputs, requiring a complex reset process for each error; and secondly, a lack of automatic fine-tuning capabilities for minor anomalies, leading to high costs for troubleshooting and trial and error. Summary of the Invention

[0007] This invention proposes a high-toughness vertical shaft cutting head design platform and method, aiming to solve problems in existing automated design and verification processes, such as the difficulty in timely identification of parameter anomalies, the inability to predict simulation risks in advance, and the need for a complete restart after an error. Existing methods lack effective pre-verification and risk prediction mechanisms during automatic modeling and simulation, leading to difficulties in timely detection of issues such as component size conflicts, misalignment of flow field inlets and particle sources, and unmet environmental parameters. Even if the program runs without errors, the actual model or calculation results may be unusable. Furthermore, once an error occurs in the current system, users can only set parameters and recalculate from scratch, lacking a mechanism for rapid recovery based on historical operations and fault-tolerant adjustment for minor anomalies, severely impacting the efficiency and stability of design verification.

[0008] To address the above problems, the present invention adopts the following technical solution:

[0009] A high-toughness shaft cutting head design platform is used for parameter verification, risk prediction, and fault tolerance recovery in the automatic modeling, automatic simulation, and design verification process of shaft cutting heads. It includes an abnormal state identification module, an abnormal risk prediction module, a recovery and fault tolerance module, and a data template storage module.

[0010] The abnormal state identification module is used to determine whether the cutting head size parameters meet the spatial assembly constraints. If not, the cutting head size parameters are corrected; if they are met, flow field and particle source verification is performed. It is also used to determine whether the particle source coordinates during the cutting head operation are within the slag discharge medium flow field calculation domain, i.e., flow field and particle source verification. If they are not within the slag discharge medium flow field calculation domain, an adjustment prompt is given. If the particle source coordinates are within the slag discharge medium flow field calculation domain, the module enters the abnormal risk prediction module to perform mesh generation and simulation calculation, and stores the component size or position parameters, particle source coordinates, etc., in the data template storage module.

[0011] The anomaly risk prediction module incorporates a mesh generation neural network and a simulation calculation neural network. The mesh generation neural network takes the cutting head size parameters and mesh generation settings as inputs, and outputs mesh error prediction results and mesh quality scores, used to predict mesh quality and report errors before mesh generation. The simulation calculation neural network takes the cutting head size parameters, boundary conditions, mesh generation settings, mesh feature parameters, and simulation settings as inputs, and outputs simulation error prediction results and simulation performance index prediction trends, used to predict simulation error risks and report errors before simulation calculation.

[0012] The recovery and fault tolerance module is configured to: when an error occurs, recall historical data to roll back to the previous historical feasible parameters or the previous error-free step, or adjust the parameters and record them as a repair strategy;

[0013] The data template storage module is used to classify, store, and manage historical parameters and results at each stage, and to summarize stable and effective parameter combinations and repair strategies that have been verified in practice into a reusable template library.

[0014] This invention also provides a design method for a high-toughness vertical shaft cutting head based on the aforementioned platform, comprising the following steps:

[0015] S1: Input the cutting head size parameters, cutting head material parameters, boundary conditions and mesh setting parameters. The abnormal state identification module performs legality, spatial interference relationship and physical feasibility verification and corrects infeasible parameters.

[0016] S2: The mesh generation neural network predicts the mesh quality based on the cutting head size parameters and mesh generation settings parameters; if the mesh quality does not meet the preset standard, an error is reported and the process proceeds to S4; otherwise, mesh generation is performed and S3 is executed.

[0017] S3: After the mesh is generated, the simulation calculation neural network predicts the simulation error risk based on the cutting head size parameters, boundary conditions, mesh generation settings parameters, mesh feature parameters, and simulation settings parameters, and outputs the simulation error risk prediction and simulation performance index prediction trend; if the risk or performance index exceeds the preset threshold, an error is reported and the process proceeds to S4, otherwise the simulation calculation is submitted and S5 is executed.

[0018] S4: When an error occurs, the recovery and fault tolerance module will roll back to the previous historical feasible parameters or the previous error-free step, or adjust the cutting head size parameters, cutting head material parameters, boundary conditions, mesh generation settings parameters, and simulation settings parameters; then return to the step before the error occurred and continue execution.

[0019] S5: After the simulation calculation is completed, the data template storage module classifies, stores and manages the parameters of each stage, the mesh generation results and the simulation results.

[0020] S6: Iteratively update the two neural networks based on the data stored in S5.

[0021] Compared with existing technologies, the high-toughness vertical shaft cutting head design platform and method proposed in this invention have the following significant features and advantages:

[0022] First, an abnormal operation identification module was constructed to pre-judge and constrain potentially infeasible parameters during automatic modeling and simulation. Through spatial interference verification, threshold limits for flow field inlet and particle source positions, and control of key parameter ranges, obviously unreasonable parameter combinations can be automatically blocked before simulation calculations, avoiding the problem of modeling or simulation failure due to incorrect parameter settings in traditional methods, and effectively improving the reliability of modeling and simulation.

[0023] Second, an anomaly risk prediction neural network module is introduced to predict potential error types, mesh quality issues, and performance anomalies during the mesh generation and simulation calculation stages. By calling the network model in real time to assess the risk of user input parameters, it provides scientific guidance for parameter adjustment, enabling designers to anticipate potential problems before or early in the simulation process. This reduces the cost of repeated trial and error and significantly improves simulation efficiency and experimental reliability.

[0024] Third, a recovery and fault-tolerance mechanism module was designed to enable rapid rollback and fine-tuning iteration of historical operation records. The system can automatically save the most recent parameter operations, allowing users to quickly restore to the valid configuration and make local adjustments to key parameters when errors occur. For minor anomalies, the system can automatically fine-tune and iteratively repair simulation tasks, thereby reducing repeated reset operations and ensuring the continuity and efficiency of the modeling and simulation process.

[0025] Fourth, a highly efficient automated design method with a closed-loop process was constructed, which realizes full-process automation and modular data transfer from parameter input, anomaly detection, mesh generation, simulation calculation, fault tolerance processing to template saving, eliminating manual intervention and repetitive configuration.

[0026] Fifth, a highly resilient, modular, and scalable software development methodology has been developed: reducing the risk of failure in a single link spreading to the entire design process, improving the reliability, continuity, recoverability, and efficiency of the design verification process, applicable to rapid modeling and simulation of vertical shaft cutting heads under multiple working conditions and parameters, and can also be extended to the design and optimization of key structures of other tunneling equipment, possessing general engineering application value. Attached Figure Description

[0027] Figure 1 Flowchart for designing a high-toughness vertical shaft cutting head software.

[0028] Figure 2 This is a comparison chart showing the average single design cycle between the conventional design process and the high-toughness design platform of this invention. Detailed Implementation

[0029] The present invention will be further described and illustrated below with reference to specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly, provided that there is no mutual conflict.

[0030] To address the problems in existing automatic modeling and simulation processes, such as unreasonable parameter settings leading to structural interference, flow field boundary misalignment, and particle sources escaping the computational domain, as well as the difficulty in predicting the various causes of simulation errors and the need to reset all parameters once an error occurs, resulting in a large amount of repetitive work, this invention proposes a high-toughness vertical shaft cutting head design platform and method.

[0031] like Figure 1 The design platform consists of four main modules: Module a: Abnormal state identification module, Module b: Abnormal risk prediction module, Module c: Recovery and fault tolerance module, and Module d: Data template storage module.

[0032] Module a: Abnormal State Recognition Module

[0033] Module A's function is to add logical judgment statements to key nodes in the automatic modeling and simulation scripts to perform constraint verification on the user-input cutting head structural parameters (including dimensional parameters, etc.), cutting head material parameters, and boundary conditions, and to verify the physical feasibility of the intermediate model (mesh model, etc.) generated from these parameters. The purpose of this module is to prevent obviously infeasible or erroneous parameter combinations before model generation or formal simulation calculations.

[0034] Specifically, this module includes a spatial relationship and interference verification unit, and a flow field and particle source verification unit:

[0035] The spatial relationship and interference verification unit, designed for the automatic modeling stage, performs a spatial interference pre-check on each pair of components in the vertical shaft cutting head that have assembly or relative motion relationships. Once their dimensions or positional parameters are determined, the unit is based on preset constraints such as minimum assembly clearance, manufacturing tolerances, and functional safety margins. It automatically determines whether each pair of components meets the spatial assembly constraints. If not, an error message is returned, and the system automatically corrects the error to the most recent valid value (users can choose whether to correct it). The interference area is highlighted for user confirmation. If the interference is satisfied, the system proceeds to the flow field and particle source verification unit for flow field and particle source verification.

[0036] Specifically, spatial interferometry pre-detection employs a physical realizability assessment method based on constraint margins. Taking the automatic modeling stage as an example, given the structural parameters of the vertical shaft cutting head, the material parameters of the cutting head, the boundary conditions, and the mesh setting parameters, these parameters collectively constitute a parameter vector. .

[0037]

[0038] Under this set of parameters, for any two components i and j in the vertical shaft cutting head that have an assembly or relative motion relationship, according to the parameter vector Calculate its minimum spatial distance And according to the preset minimum assembly gap Manufacturing tolerances , and functional safety margin Constructing the interferometric discriminant function :

[0039]

[0040] when When, it is determined that the set of parameters satisfies the spatial assembly constraints; when If an interference or insufficient clearance risk is detected, the system outputs the corresponding component number, interference location, and parameter correction direction.

[0041] The flow field and particle source verification unit, designed for the automatic simulation phase, constrains the flow field inlet and particle source positions during the operation of the vertical shaft cutting head. Once the flow field inlet position and flow domain boundary are set, a legal spatial domain for the particle source is automatically established by calculating the projection of the flow field inlet surface. This ensures that the coordinates of each particle are contained within the convergent range of the flow field, preventing simulation failures caused by particles escaping the computational domain. If the particle source coordinates are within the convergent range of the flow field, the simulation proceeds to module b for mesh generation and calculation, storing the component dimensions or position parameters, the flow field inlet position, and the flow domain boundary in module d.

[0042] Furthermore, before performing spatial relationship and interference verification, module a imposes threshold restrictions on the overall structural dimensions, material and environmental parameters of the shaft cutting head (such as the maximum overall dimensions, the recommended range of minimum mesh element size, fluid property parameters, and inlet / outlet boundary condition parameters) to ensure that the entire modeling and simulation process remains within the physically and numerically feasible range. For example, for a single dimension parameter... Using interval constraints Perform a validity check, among which and These are the lower and upper limits, respectively, determined by factors such as structural strength, manufacturing process, slag discharge channel dimensions, and mesh subdivision. If... If the value exceeds the constraint range, module a will correct it to a valid value. The parameter values ​​before and after the correction are written to the log file.

[0043] In summary, module a establishes an abnormal operation identification mechanism for the design of high-toughness vertical shaft cutting heads. During the automatic modeling stage, it can output the structural parameters of the cutting head, the material parameters of the cutting head, the legality judgment results of the boundary conditions, automatic parameter correction suggestions and the corrected parameter values, as well as log files that record all errors / warnings, providing a basic guarantee for the stable operation of subsequent modules.

[0044] Module b: Anomaly Risk Prediction Module

[0045] The function of module b is to predict possible error types, mesh quality problems, simulation anomaly indicators and results based on neural networks during the mesh generation and simulation calculation stages, thereby providing risk warnings before mesh generation and simulation execution or in the early stages of operation.

[0046] Module b incorporates a mesh generation neural network (Mesh NN) and a simulation calculation neural network (Solver NN). The Mesh NN takes the cutter head structure parameters and mesh generation settings as input, and outputs a mesh quality evaluation (score) and potential mesh generation error types. The Solver NN takes the cutter head structure parameters, boundary conditions, mesh generation settings, post-mesh characteristic parameters, and simulation settings as input, and outputs simulation error predictions (whether errors will occur or not), simulation error types, and prediction trends for various performance indicators (such as residuals, continuity, etc.). Based on these two neural networks, users can anticipate potential risks in mesh generation and anomalies in simulation calculations, allowing them to adjust parameters accordingly. Both neural networks iteratively update based on data from each mesh generation and simulation calculation, improving prediction accuracy.

[0047] Furthermore, both Mesh NN and Solver NN employ a feedforward neural network structure consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives the corresponding cut-head structure parameters, boundary conditions, mesh generation settings, mesh feature parameters, and simulation settings. The hidden layers extract the nonlinear correlation features between the input parameters. The output layer includes a classification output and a regression output. Specifically, the classification output of Mesh NN outputs the mesh error prediction results (including whether mesh generation errors occurred and the potential types of mesh generation errors), while the regression output outputs the mesh quality score. Similarly, the classification output of Solver NN outputs the simulation error prediction results (including whether simulation calculation errors occurred and the potential types of simulation errors), and the regression output predicts the trends in simulation performance indicators such as residuals and continuity.

[0048] The training data for both neural networks originates from data records generated during the platform's historical mesh generation and simulation calculations. Specifically, the training samples for the Mesh NN include historical cut-head structure parameters, historical mesh generation settings, and their corresponding historical mesh error prediction results and quality assessment results. The training samples for the Solver NN include historical cut-head structure parameters, historical boundary conditions, historical mesh generation settings, historical mesh feature parameters, historical simulation settings, and their corresponding historical simulation error prediction results and historical simulation performance index prediction trends. Before training, the input parameters are normalized. Historical mesh error prediction results are used as the classification training labels for the Mesh NN, historical simulation error prediction results as the classification training labels for the Solver NN, historical mesh quality scores as the regression training labels for the Mesh NN, and historical simulation performance indices such as historical residuals and historical continuity as the regression training labels for the Solver NN. Network parameters are updated through supervised learning. After each mesh generation or simulation calculation, the platform adds the newly generated data to the training dataset and incrementally trains and updates both neural networks according to a preset period or sample size threshold.

[0049] In summary, module b can output predicted results and risk warnings for mesh quality and simulation anomalies, and record relevant prediction and error logs, providing a fundamental guarantee for the stable operation of subsequent modules. Module b introduces a neural network for mesh generation and simulation calculation to predict error types and performance anomaly trends in real time, providing a scientific basis for parameter adjustment and reducing the cost of repeated trial and error.

[0050] Module c: Recovery and Fault Tolerance Module

[0051] Module C's function is to provide the ability to quickly restore to a historically valid state and automatically fine-tune for minor anomalies when an anomaly occurs in the overall process (either an actual error occurs or a risk warning is issued by Module B), thereby reducing the cost of redoing from scratch. Module C employs recovery and fault-tolerance mechanisms.

[0052] Module C categorizes anomalies into minor anomalies and serious errors based on their impact on the current design verification process, the integrity of the calculation process, and the recoverability of the anomaly. Minor anomalies refer to situations where the automatic modeling, meshing, or simulation calculation process has not been interrupted, the generated model files, intermediate calculation results, or historical parameters remain valid, and the anomaly index only exceeds the preset warning threshold but not the preset error threshold; or, based on judgment, can be restored to normal operation by adjusting one or more relevant parameters within a preset fine-tuning range and conducting short-term simulation verification. Serious errors refer to situations where automatic modeling, meshing, or simulation calculation cannot continue, resulting in model generation failure, meshing failure, solver interruption, calculation non-convergence, missing key result files, or anomaly indexes exceeding the preset error threshold; it also includes situations where minor anomalies fail to recover after a preset number of parameter fine-tunings and short-term simulation verifications. For minor anomalies, Module C activates a fault-tolerance mechanism; for serious errors, Module C activates a recovery mechanism, reverting to the most recent historical valid parameter configuration or the previous step without errors. The "anomaly index" mentioned in this paper refers broadly to numerical evaluation parameters that can be monitored in real time during the modeling and simulation process.

[0053] The recovery mechanism addresses more severe errors by retrieving various information from module d, including historical cutting head structure parameters, historical cutting head material parameters, historical boundary conditions, historical mesh generation, historical simulation settings parameters, and module files generated from simulation calculations. When an error occurs, it can revert to the most recent valid parameter configuration; or, if an error occurs in some stages, it can partially revert to the last successful step. For example, if there is no error in the mesh generation process but an error occurs in the simulation calculation, the mesh generation results can be retained, and the program can be restored to the simulation calculation stage.

[0054] The fault tolerance mechanism automatically fine-tunes minor anomalies that are close to normal: it sets the fine-tuning range for the cutting head structural parameters, cutting head material parameters, boundary conditions, mesh generation settings parameters, and simulation settings parameters, and verifies the repair effect after fine-tuning by conducting short-term simulations. If successful, it continues to run and records this adjustment as a repair strategy; otherwise, it prompts the user for manual intervention.

[0055] Furthermore, the recovery and fault tolerance mechanisms of module c adopt a parameter fine-tuning strategy of "hierarchical perturbation - short-term verification - strategy solidification". Module c first classifies errors or risk warnings into mesh-type anomalies, boundary-type anomalies, particle-source-type anomalies, and solution-type anomalies according to the anomaly type, and sets corresponding adjustable parameter sets for different anomaly types.

[0056] For example, for mesh-related anomalies, priority is given to adjusting local mesh size, curvature refinement coefficient, and minimum element mass threshold; for boundary-related anomalies, priority is given to adjusting inlet / outlet positions, watershed envelope, and rotation radius; for particle source-related anomalies, priority is given to adjusting particle source projection position, particle release area, and initial particle velocity; and for solution-related anomalies, priority is given to adjusting time step, iteration number, relaxation factor, and convergence residual threshold. Mesh-related anomalies correspond to errors or risk warnings such as mesh generation failure predicted by the mesh generation neural network and mesh quality not meeting preset standards; boundary-related anomalies include anomalies such as unreasonable inlet or outlet positions of the flow field or improper watershed boundary settings causing boundary conditions to be unrealizable; particle source-related anomalies include particle source coordinates exceeding the legal spatial domain, particles escaping the computational domain, abnormal particle accumulation, or abnormal passage states; and solution-related anomalies include residuals or continuity predicted by the simulation calculation neural network exceeding preset thresholds, as well as actual problems such as solver interruption, computational non-convergence, flow field divergence, or missing key result files.

[0057] For any parameter to be adjusted Module c is within the preset fine-tuning range Candidate repair parameters are generated internally, and short-time simulations are used to verify the candidate repair parameters; This is the preset maximum single-step fine-tuning range for this parameter, the specific value of which is determined by the magnitude of each parameter and engineering experience. If the mesh quality, residual changes, particle passage status, and intermediate result file output in the short-term simulation all meet the preset conditions, the fine-tuning scheme is deemed effective, and subsequent complete simulations continue—this is the fault-tolerance mechanism. If multiple consecutive fine-tuning attempts still fail to meet the conditions, the system will revert to the most recent effective state or prompt the user for manual intervention—this is the recovery mechanism. Verified effective fine-tuning schemes will be stored as repair strategies in module d for automatic recall in the event of similar anomalies.

[0058] Module c provides both automatic execution and semi-automatic interactive modes, and ensures that all processes are traceable and backtrackable.

[0059] Module d: Data template storage module

[0060] The function of module d is to classify, store and manage the key data, operation records and stable and effective parameter templates generated at each stage after the design verification process is completed, so as to provide data support for subsequent calls, historical traceability and model updates.

[0061] Specifically, module d provides structured storage for the cutting head structural parameters, material parameters, boundary conditions, mesh generation settings, simulation settings, solution results, error logs, risk warning information, and recovery and fault tolerance records for each stage. Simultaneously, it summarizes and organizes stable and effective combinations of cutting head structural and material parameters, boundary conditions, mesh generation settings, simulation settings, and repair strategies that have been verified in practice, forming a reusable template library. Subsequent similar design verification tasks can directly call upon the corresponding historical data and templates, reducing redundant settings and improving design verification efficiency and stability.

[0062] In summary, module d enables the categorized storage of key data, the tracing of historical records, and the summarization and retrieval of effective templates, providing a fundamental guarantee for continuous optimization and rapid execution of subsequent processes.

[0063] Overall, the four modules provide strong support for the stable operation of the shaft cutting head design verification software, covering the entire process from parameter input to simulation output. First, after the user inputs the cutting head structural parameters and boundary conditions, module a verifies the validity of the parameters and spatial interference relationships. After passing the pre-check, it enters the modeling and simulation stage. During the design verification process, module b predicts potential error risks in mesh generation and simulation calculations based on Mesh NN and Solver NN, and provides early warnings for possible anomalies. When module b issues a risk warning or the program actually encounters an error, the system switches to module c: for minor anomalies that are close to normal, module c automatically fine-tunes each parameter within a set range and verifies the repair effect through a short-term simulation. If the repair is successful, it continues running and records this adjustment as a repair strategy; for more serious problems, it backtracks to the most recent valid parameter configuration, or partially backtracks to the previous successful step, and prompts the user for manual intervention if necessary. After completing a simulation, module d stores and records the data from each stage. Thus, the four modules cooperate to ensure the continuity, stability, and traceability of the design verification process.

[0064] The method of this invention constructs a closed-loop workflow through the collaborative operation of four modules. The process can be described as follows:

[0065] S1. Parameter Input and Pre-check: Users input the cutting head structure parameters, cutting head material parameters, boundary conditions and mesh settings parameters through the interface or batch import (Excel / CSV / JSON); first, module a performs a legality check, spatial interference pre-check and physical feasibility check on the input parameters, and automatically corrects obvious out-of-bounds items or prompts the user to make interactive modifications.

[0066] S2. Mesh Prediction and Mesh Generation: After passing the pre-check, the model enters the mesh generation stage. The Mesh NN in module b predicts the mesh quality based on the cut-head structure parameters and mesh generation settings, outputting a mesh quality evaluation and potential error risks. If the prediction results indicate a high risk, it directly proceeds to S4 for recovery or fault tolerance processing; if the risk is within an acceptable range, it submits for mesh generation and executes S3.

[0067] S3. Simulation Submission and Early Monitoring: After mesh generation, before simulation startup, the Solver NN in module b predicts simulation error risks and performance trend based on cutter head structure parameters, boundary conditions, mesh generation settings, mesh feature parameters, and simulation settings. If the risk or performance index is lower than a preset threshold, an external solver is submitted for simulation calculation, and S5 is executed. If the risk or performance index is slightly higher than the preset threshold, the user is advised to modify the settings or execute a low-risk solution (e.g., reduce parallelism, increase iteration limit). If the score is significantly higher than the preset threshold, proceed to S4. During simulation runtime, residual values ​​and intermediate outputs are monitored in real time. If early abnormal signs appear, an early stop or fault-tolerant process is triggered, also proceeding to S4.

[0068] S4. Recovery and Fault Tolerance: When module a or b issues a risk warning or an actual program error occurs, module c is invoked. For minor anomalies that are close to normal, the cutting head structural parameters, cutting head material parameters, boundary conditions, mesh generation settings, and simulation settings are automatically fine-tuned within the set range. The repair effect is verified through short-term simulation. If the repair is successful, the program continues to run and records this adjustment as a repair strategy. If the repair fails or the problem is more serious, the program reverts to the most recent valid parameter configuration or partially reverts to the previous successful step, prompting the user for manual intervention if necessary. Module c can be executed automatically or semi-automatically, and the entire process is traceable and reversible.

[0069] S5. Result Extraction and Data Recording: After the simulation is completed, or after recovery and fault tolerance processing, module d extracts the simulation performance indicators and synchronously records key data such as the input values ​​of each parameter, mesh generation results, simulation setting parameters, solver logs, risk warning information, recovery and fault tolerance processing process and its results, forming structured data and storing it in the historical database.

[0070] S6. Model Update and Process Optimization: Based on the continuously accumulated mesh generation, simulation calculation, risk warning, recovery and repair, and final result data in module d, Mesh NN and Solver NN are iteratively updated to continuously improve risk prediction and anomaly identification capabilities. At the same time, the stable and effective combinations of cutting head structure parameters and material parameters, boundary conditions, mesh generation settings, simulation settings, and repair strategies verified in practice are summarized and saved in module d, providing a basis for subsequent design verification processes, thereby achieving continuous optimization of the overall process.

[0071] In summary, the pre-judgment of module a, the risk prediction of module b, the recovery and fault tolerance of module c, and the data storage of module d work together to form a highly resilient design verification platform that can both prevent infeasible parameter combinations in the early stage and realize risk warning and automatic repair during the simulation execution stage.

[0072] To quantitatively evaluate the improvement effect of this invention on the high toughness and operational efficiency of the vertical shaft cutting head design verification process, Figure 2 To statistically compare the results of 25 simulation optimization tasks in the cutting head design process, calculations were performed using both conventional manual modeling and simulation processes and the design platform described in this invention. The average time consumed in each stage—parameter input and modification, modeling, mesh generation, simulation calculation, rollback and resubmission after errors, and manual confirmation and result verification—was statistically analyzed. The comparison results show that the mesh generation time and simulation calculation time are basically the same for both methods, indicating that this invention does not gain a time advantage by reducing mesh accuracy or compressing the solution process. However, using this invention significantly reduces the time for parameter input and modification, modeling, and the time for rollback, reset, and resubmission due to errors, resulting in an average design cycle reduction of more than 20% compared to the conventional process.

[0073] The aforementioned improvements primarily stem from the high resilience mechanism of the design platform of this invention: the anomaly identification module (module a) can block physically infeasible or numerically unreasonable parameter combinations before modeling and simulation execution; the anomaly risk prediction module (module b) can identify mesh quality issues and simulation error risks in advance; when actual errors or risk warnings occur, the recovery and fault tolerance module (module c) can perform restricted parameter fine-tuning, short-term verification, historical valid state rollback, or successful step recovery according to the degree of anomaly, avoiding the complete failure of existing models, meshes, and intermediate results; the data template storage module (module d) can save valid parameters, repair strategies, and historical running information, providing a basis for the rapid handling of similar anomalies in the future. Through the collaboration of the above modules, even under disturbances such as parameter anomalies, mesh partitioning risks, or unstable solution processes, the platform can still maintain the continuous operation of the design verification process and possess the capabilities of anomaly pre-identification, local fault tolerance, rapid recovery, and result traceability.

[0074] Therefore, the technical effects of this invention are not only reflected in the shortening of the average design cycle, but more importantly, it improves the anti-disturbance capability and self-recovery capability of the automatic modeling and simulation process of the vertical shaft cutting head under abnormal conditions, reduces the risk of the failure of a single link spreading to the entire design process, and realizes the simultaneous improvement of the reliability, continuity, recoverability and efficiency of the design verification process, fully demonstrating the high toughness advantage of the platform of this invention.

[0075] In summary, this invention establishes a high-toughness shaft cutting head design and verification platform through modular design, anomaly identification and prediction, and recovery-tolerant closed-loop control. This method significantly improves the reliability and controllability of the modeling and simulation process, reduces the need for human intervention, shortens the design verification cycle, and ensures the scientific and efficient nature of parameter adjustment and performance optimization. It is suitable for the rapid development and performance evaluation of shaft cutting head structures under multiple working conditions and parameters.

[0076] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. Those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A high-toughness shaft cutting head design platform, used for parameter verification, risk prediction, and fault-tolerant recovery in the automatic modeling, automatic simulation, and design verification process of shaft cutting heads, characterized in that, include: The abnormal state identification module is used to determine whether the cutting head size parameters meet the spatial assembly constraints. If not, the cutting head size parameters are corrected. And it is used to determine whether the coordinates of the particle source during the cutting head operation are within the calculation domain of the slag discharge medium flow field. If they are not within the calculation domain of the slag discharge medium flow field, an adjustment is prompted. The anomaly risk prediction module incorporates a mesh generation neural network and a simulation calculation neural network. The mesh generation neural network takes the cutting head size parameters and mesh generation settings as inputs, and outputs mesh error prediction results and mesh quality scores, used to predict mesh quality and report errors before mesh generation. The simulation calculation neural network takes the cutting head size parameters, boundary conditions, mesh generation settings, mesh feature parameters, and simulation settings as inputs, and outputs simulation error prediction results and simulation performance index prediction trends, used to predict simulation error risks and report errors before simulation calculation. The recovery and fault tolerance module is configured to: when an error occurs, call historical data to roll back to the previous historical feasible parameters or the previous error-free step, or adjust the parameters and record them as a repair strategy; The data template storage module is used to classify, store, and manage historical parameters and results at each stage, and to summarize stable and effective parameter combinations and repair strategies that have been verified in practice into a reusable template library.

2. The design platform according to claim 1, characterized in that, The recovery and fault tolerance module employs recovery and fault tolerance mechanisms. The recovery mechanism calls the historical cutting head size parameters, historical cutting head material parameters, historical boundary conditions, historical mesh generation settings parameters, and historical simulation settings parameters stored in the data template storage module, and rolls back to the previous historical feasible parameters or the previous step without errors; The fault tolerance mechanism sets adjustment ranges for cutting head size parameters, cutting head material parameters, boundary conditions, mesh generation settings parameters, and simulation settings parameters. Within the adjustment range, the parameters are adjusted, and a short-term simulation is performed to verify the effect of the adjustment. If no error is reported, the adjustment is recorded as a repair strategy; otherwise, manual intervention is prompted.

3. The design platform according to claim 1, characterized in that, Both the mesh generation neural network and the simulation calculation neural network adopt a feedforward neural network structure consisting of an input layer, multiple hidden layers, and an output layer. The input layer is used to receive the corresponding cut-head structure parameters, boundary conditions, mesh generation setting parameters, mesh feature parameters, and simulation setting parameters. The hidden layer is used to extract the nonlinear correlation features between the input parameters. The output layer includes a classification output and a regression output; The classification output of the grid partitioning neural network is used to output the grid error prediction result, which includes whether a grid partitioning error has occurred and the potential grid partitioning error type; the regression output is used to output the grid quality score. The classification output of the simulation computational neural network is used to output simulation error prediction results, which include whether a simulation error has occurred and the potential simulation error type. The regression output is used to predict trends in simulation performance indicators, which include residuals and continuity.

4. The design platform according to claim 3, characterized in that, The training data for the mesh partitioning neural network includes historical cutter head structure parameters, historical mesh partitioning setting parameters, historical mesh error prediction results, and historical mesh quality evaluation results; the mesh partitioning neural network uses the mesh error prediction results as classification training labels and the historical mesh quality scores as regression training labels. The training data of the simulation computational neural network includes historical cutter head structure parameters, historical boundary conditions, historical mesh partitioning settings, historical mesh feature parameters, historical simulation settings, historical simulation error prediction results, and historical simulation performance indicators. The simulation computational neural network uses historical simulation error prediction results as classification training labels and historical simulation performance indicators as regression training labels. The mesh partitioning neural network and the simulation computational neural network update network parameters through supervised learning.

5. The design platform according to claim 4, characterized in that, After each mesh partitioning or simulation calculation is completed, the design platform adds the newly generated data to the training dataset and performs incremental training and updates on the mesh partitioning neural network and the simulation calculation neural network according to a preset period or sample number threshold.

6. The design platform according to claim 1, characterized in that, The mesh generation neural network reports an error when the predicted mesh quality does not meet the preset standard; the simulation calculation neural network reports an error when the predicted simulation error risk or the predicted trend of simulation performance indicators exceeds the preset threshold.

7. The design platform according to claim 1, characterized in that, The abnormal state identification module includes a spatial relationship and interference verification unit and a flow field and particle source verification unit. The spatial relationship and interference verification unit is used to determine whether the cutting head size parameters meet the spatial assembly constraints. Specifically, for each pair of components in the vertical shaft cutting head that have an assembly or relative motion relationship, once their size or position parameters are determined, a spatial interference pre-check is performed based on the preset minimum assembly gap, manufacturing tolerance, and functional safety margin to automatically determine whether each pair of components meets the spatial assembly constraints. If not, the cutting head size parameters are corrected. The flow field and particle source verification unit is used to determine whether the coordinates of the particle source are within the calculation domain of the slag discharge medium flow field during the operation of the cutting head. Specifically, after setting the inlet position and flow domain boundary of the vertical shaft cutting head during the operation of the flow field, the legal space domain of the particle source is automatically established by calculating the inlet surface projection of the flow field, ensuring that the coordinates of each particle are wrapped within the convergent range of the flow field, and avoiding the situation where particles run out of the calculation domain and cause simulation failure.

8. The design platform according to claim 7, characterized in that, The spatial interference pre-detection adopts a physical realizability discrimination method based on constraint margin. A parameter vector is constructed based on the given cutting head structural parameters, cutting head material parameters, boundary conditions, and mesh setting parameters. Under the parameter vector, for any two components in the vertical shaft cutting head that have an assembly or relative motion relationship, the minimum spatial distance is calculated based on the parameter vector, and an interference discrimination function is constructed according to the preset minimum assembly gap, manufacturing tolerance, and functional safety margin. When the function value is greater than or equal to 0, it is determined that the parameter vector satisfies the spatial assembly constraint; otherwise, it is determined that there is a risk of interference or insufficient clearance.

9. A design method for a high-toughness vertical shaft cutting head based on the platform described in any one of claims 1 to 8, characterized in that, Includes the following steps: S1: Input the cutting head size parameters, cutting head material parameters, boundary conditions and mesh setting parameters. The abnormal state identification module performs legality, spatial interference relationship and physical feasibility verification and corrects infeasible parameters. S2: The mesh generation neural network predicts mesh quality based on the cutting head size parameters and mesh generation settings parameters; if the mesh quality does not meet the preset standard, an error is reported and the process proceeds to S4; otherwise, mesh generation is performed and S3 is executed. S3: After the mesh is generated, the simulation computational neural network predicts the simulation error risk based on the cutting head size parameters, boundary conditions, mesh generation settings parameters, mesh feature parameters, and simulation settings parameters, and outputs the simulation error risk prediction and simulation performance index prediction trend. If the risk or performance indicator exceeds the preset threshold, an error will be reported and the process will proceed to S4; otherwise, the simulation calculation will be submitted and S5 will be executed. S4: When an error occurs, the recovery and fault tolerance module will roll back to the previous historical feasible parameters or the previous error-free step, or adjust the cutting head size parameters, cutting head material parameters, boundary conditions, mesh generation settings parameters, and simulation settings parameters; then return to the step before the error occurred and continue execution. S5: After the simulation calculation is completed, the data template storage module classifies, stores and manages the parameters of each stage, the mesh generation results and the simulation results. S6: Iteratively update the two neural networks based on the data stored in S5.

10. The design method according to claim 9, characterized in that, The S6 also includes a data template storage module that summarizes the stable and effective combinations of cutting head size parameters and cutting head material parameters, boundary conditions, mesh generation setting parameters, simulation setting parameters and repair strategies into a reusable template library.