A wire cutting and removal auxiliary system for additively molded parts for rocket thrust chambers

Abstract

The present invention relates to the field of rocket thrust chamber manufacturing technology, and discloses a wire cutting and picking auxiliary system for additively molded parts for rocket thrust chambers. The system includes a multi-source data acquisition module, which collects geometric morphology data of additively molded parts through multiple sensors; a cutting path planning module, which generates a global cutting path sequence based on an improved genetic algorithm; a dynamic deformation compensation module, which uses a fuzzy adaptive observer to compensate for thermal deformation and vibration interference in real time; and a multi-axis collaborative control module, which constructs a distributed constraint optimization model to achieve precise control of multi-axis motion. The system can accurately collect data, reasonably plan cutting paths, effectively compensate for deformation, and ensure multi-axis collaborative work, significantly improving the quality and efficiency of wire cutting and picking of additively molded parts for rocket thrust chambers, reducing production costs, and providing strong technical support for aerospace manufacturing.

Description

Technical Field

[0001] The present invention relates to the technical field of rocket thrust chamber manufacturing, and in particular to a wire cutting and removing auxiliary system for additively molded parts for rocket thrust chambers. Background Art

[0002] In modern aerospace, rocket thrust chambers are core components of rocket engines, and their manufacturing precision and quality directly impact the performance and reliability of the rocket. With the rapid development of additive manufacturing (3D printing), its advantages, such as the ability to create complex structures, save materials, and shorten manufacturing cycles, have led to its increasing application in rocket thrust chamber manufacturing. Additive manufacturing can rapidly produce rocket thrust chamber components with intricate internal structures and complex shapes that are difficult to achieve using traditional machining methods.

[0003] However, after the additive manufacturing process is completed, the parts need to be removed from the printing platform, and wire cutting is a commonly used method for removing the parts. However, there are many difficulties in the actual operation process. First of all, the geometric morphology of the rocket thrust chamber additive manufacturing parts is complex and diverse, and traditional data acquisition methods are difficult to obtain its detailed geometric feature information comprehensively and accurately. This leads to a lack of accurate data support when planning the wire cutting path, and the structural characteristics of the molded parts cannot be fully considered, which can easily lead to unreasonable cutting paths and affect cutting efficiency and product quality. For example, if the complex curved surface shape of the molded part cannot be accurately grasped, the cutting path may have unnecessary detours or difficulty in reaching certain key parts, which will extend the cutting time and may even damage the molded part.

[0004] Secondly, the wire cutting process generates thermal deformation and vibration interference. The high temperature during cutting causes local expansion of the molded part, which then contracts after cooling. This thermal deformation reduces cutting accuracy and results in large dimensional deviations in the cut product, failing to meet the high-precision component requirements of rocket thrust chambers. Furthermore, vibration during the cutting process can cause the cutting tool to shake, further exacerbating cutting errors and potentially creating chatter marks on the molded part, affecting the surface quality and performance of the product.

[0005] Furthermore, the coordination of multi-axis motion control in wire-cutting equipment is a key issue. To achieve high-precision cutting, multiple motion axes must precisely coordinate and move along a predetermined trajectory. However, due to the varying dynamic characteristics of each axis, inter-axis synchronization errors can easily occur during motion, causing the cutting path to deviate from the ideal trajectory. Furthermore, in actual production environments, different batches of additively molded parts may exhibit certain differences, making it difficult for existing cutting systems to quickly adapt to these changes, making it impossible to achieve efficient and stable cutting operations.

[0006] Furthermore, traditional cutting path planning algorithms are typically based on simple geometric models and fixed constraints, failing to fully consider the unique requirements of rocket thrust chamber additive manufacturing, such as cutting energy consumption and corner smoothness. This not only increases cutting costs but can also cause excessive wear on cutting tools, shortening their service life. Furthermore, traditional deformation compensation methods, which mostly rely on empirical evidence or simple mathematical models, are unable to accurately and in real time compensate for thermal deformation and vibration disturbances during the cutting process, making them unable to meet the stringent high-precision requirements of modern aerospace manufacturing. Summary of the Invention

[0007] The purpose of the present invention is to provide a wire cutting and removing auxiliary system for additively molded parts for rocket thrust chambers to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a wire cutting and removing auxiliary system for additively molded parts for rocket thrust chambers, the system comprising:

[0009] Multi-source data acquisition module: used to collect geometric data of additively molded parts through multiple types of sensors;

[0010] Cutting path planning module: performing topological optimization analysis on the geometric shape data based on an improved genetic algorithm to generate a global cutting path sequence;

[0011] Dynamic deformation compensation module: uses fuzzy adaptive observer to model thermal deformation and vibration interference in the cutting process in real time and output compensation control value;

[0012] Multi-axis collaborative control module: Based on the global cutting path sequence and compensation control amount, a distributed constraint optimization model is constructed to generate multi-axis motion control instructions.

[0013] Preferably, the multi-source data acquisition module includes:

[0014] Multiple types of sensors include laser displacement sensors, infrared thermal imagers, acoustic emission sensors, and visual cameras;

[0015] Perform 3D point cloud registration on laser displacement sensor data and visual camera data to construct a surface topography model of the additively molded part; perform wavelet packet decomposition on infrared thermal imager data and acoustic emission sensor data to extract thermal stress distribution and material peeling characteristics;

[0016] A two-branch feature extraction network was constructed. The first branch used a three-dimensional convolutional neural network to extract the local geometric features of the surface morphology model, and the second branch used a temporal attention network to extract the dynamic change characteristics of the thermal stress distribution.

[0017] The local geometric features and dynamic change features are fused through a cross-modal feature alignment mechanism to generate a joint feature tensor. The joint feature tensor is spatially correlated and modeled based on a gated graph neural network to output comprehensive monitoring features including deformation trend, cutting residual stress, and material peeling status.

[0018] Preferably, the improved genetic algorithm integrates a dynamic mutation operator and an elite retention strategy, specifically including:

[0019] The cutting path planning is modeled as a multi-constrained traveling salesman problem, where the decision variables include the cutting starting point, transition path, and cutting order.

[0020] Initialize the population and calculate the fitness function, which includes the total path length, corner smoothness, and cutting energy consumption weight terms;

[0021] In the mutation phase, the mutation probability is dynamically adjusted according to the thermal conductivity of the material; in the crossover phase, a segmented reorganization strategy is used to retain the local optimal path segments;

[0022] The Pareto front screening mechanism is introduced to iteratively optimize the non-inferior solution set and finally output the global cutting path sequence.

[0023] Preferably, the fuzzy adaptive observer adopts a double closed-loop compensation structure, specifically including:

[0024] Construct a thermal-mechanical coupling state equation, taking thermal deformation, vibration acceleration, and material removal rate as state variables;

[0025] Design a fuzzy rule base to map state variables into membership function activation weights;

[0026] The confidence of fuzzy rules is dynamically updated through online learning mechanism, and the membership function parameters are optimized using gradient descent method;

[0027] The output includes the compensation control amount of axial compensation displacement and speed correction amount.

[0028] Preferably, the distributed constraint optimization model adopts a hierarchical interactive architecture, specifically including:

[0029] Construct a multi-axis motion constraint graph, where nodes represent the control units of each motion axis and edges represent the synchronization error constraints between axes;

[0030] The alternating direction multiplier method is used to decompose the global optimization problem into local subproblems. Each subproblem contains axis speed limit, acceleration smoothing term and synchronization error penalty term.

[0031] Introducing virtual synchronization variables at the coordination layer to balance conflicts in control instructions between axes, and implementing distributed iterative solutions through a consistency protocol;

[0032] Finally, multi-axis motion control instructions that meet dynamic accuracy and synchronization are generated.

[0033] Preferably, the three-dimensional convolutional neural network adopts a hollow residual structure, specifically including:

[0034] The surface topography model is divided into a multi-resolution voxel grid, each voxel stores curvature, normal vector and height gradient statistics;

[0035] In the convolution stage, a dilation rate of 3 is used to capture large-scale geometric features, and the residual connection stage fuses shallow details with deep semantic features.

[0036] A spatial attention module is introduced to adaptively weight feature channels.

[0037] Preferably, the dynamic mutation operator is implemented based on an environment response mechanism, specifically including:

[0038] The heat accumulation and path curvature during the historical cutting process are collected as variation factors;

[0039] Construct a radial basis function neural network to fit the mapping relationship between mutation probability and environmental parameters;

[0040] The mutation intensity is dynamically adjusted through an online reinforcement learning strategy, and a directed mutation operation is triggered when local overheating is detected.

[0041] Preferably, the dual closed-loop compensation structure includes a feedforward compensation and a feedback correction loop, specifically including:

[0042] The feedforward loop predicts the deformation based on the heat conduction equation and uses a sliding mode controller to generate pre-compensation;

[0043] The feedback loop fuses the vibration sensor data with the visual measurement data through a Kalman filter to generate an error correction;

[0044] A hybrid compensation strategy is designed to output the final compensation control amount after weighted superposition of the pre-compensation amount and the error correction amount.

[0045] Preferably, the virtual synchronization variable adopts a time lag compensation mechanism, specifically including:

[0046] Measure the transmission delay of each axis control command and build a time-varying communication topology matrix;

[0047] A time delay compensation term is introduced into the consensus protocol, and the compensation coefficient is determined by the Lyapunov stability criterion.

[0048] Preferably, the spatial attention module adopts a channel-position joint focusing mechanism, specifically including:

[0049] Calculate the information entropy weight of the feature map in the channel dimension and select high-information feature channels;

[0050] Construct a Gaussian mixture model in the spatial dimension to locate key geometric defect areas;

[0051] The channel weights and spatial weights are fused via the Hadamard product.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The wire cutting and picking auxiliary system for additively molded parts for rocket thrust chambers of the present invention has shown significant beneficial effects in many aspects in practical applications. At the data acquisition and analysis level, the multi-source data acquisition module uses various types of sensors such as laser displacement sensors, infrared thermal imagers, acoustic emission sensors and visual cameras to achieve comprehensive and accurate acquisition of geometric morphology data of additively molded parts. By constructing a surface morphology model through three-dimensional point cloud registration and wavelet packet decomposition of infrared thermal imager and acoustic emission sensor data, it is possible to deeply explore the thermal stress distribution and material peeling characteristics. On this basis, the dual-branch feature extraction network and the cross-modal feature alignment mechanism further integrate and strengthen the data features, providing a solid data foundation for subsequent cutting path planning and deformation compensation. This ensures that the system accurately grasps the status of additively molded parts and greatly improves the reliability and scientificity of pre-cutting preparations.

[0054] The cutting path planning module uses an improved genetic algorithm to model cutting path planning as a multi-constrained traveling salesman problem, comprehensively considering factors such as total path length, corner smoothness, and cutting energy consumption. The introduction of a dynamic mutation operator and an elite retention strategy enables the algorithm to more efficiently search for the globally optimal cutting path sequence in a complex solution space. In actual cutting operations, a shorter total path length means a significant reduction in cutting time, improving production efficiency; optimizing corner smoothness effectively reduces cutting tool wear, extends its service life, and reduces production costs; and reasonable control of cutting energy consumption not only saves energy, but also reduces the potential impact of overheating on the material properties of the molded part, thereby ensuring product quality.

[0055] The dynamic deformation compensation module utilizes a dual closed-loop compensation structure based on a fuzzy adaptive observer to accurately model and compensate for thermal deformation and vibration disturbances during the cutting process in real time. Feedforward compensation predicts deformation based on the heat conduction equation, while the feedback correction loop generates error corrections by fusing multiple measurement data using a Kalman filter. The combined output of these two factors allows for timely adjustment of the cutting tool's position and speed during the cutting process, effectively reducing the impact of thermal deformation and vibration disturbances on cutting accuracy. Experimental data shows that the module improves cutting accuracy by [X]% compared to traditional methods, significantly improving the dimensional accuracy and surface quality of the cut product, ensuring that the additively molded parts of the rocket thrust chamber meet the high-precision requirements of aerospace grade.

[0056] The distributed constraint optimization model constructed by the multi-axis collaborative control module adopts a hierarchical interactive architecture that fully considers the speed limits, acceleration smoothing terms, and synchronization error penalties of each motion axis. The global optimization problem is decomposed into local subproblems for solution using the alternating direction multiplier method, and the introduction of virtual synchronization variables and time-lag compensation mechanisms effectively resolves the synchronization error problem in multi-axis motion control. This enables the various motion axes of the wire cutting equipment to work closely together and move along precise trajectories, further ensuring cutting accuracy. At the same time, the architecture also has good adaptability and can quickly respond to the differences between different additively molded parts, achieving stable and efficient cutting operations.

[0057] Overall, the system of the present invention organically combines multiple key modules to form a complete and efficient wire-cutting component removal assistance system. The modules collaborate and complement each other, comprehensively improving the quality and efficiency of wire-cutting component removal for additively molded parts used in rocket thrust chambers, from data acquisition, path planning, deformation compensation, to multi-axis coordinated control. This not only helps reduce production costs and improve production efficiency, but also provides strong technical support for the manufacture of rocket thrust chambers in the aerospace field, promoting the further development of aerospace manufacturing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a working principle diagram of the wire cutting and removing auxiliary system for additively molded parts for rocket thrust chambers according to the present invention;

[0059] Figure 2 This is the working principle diagram of the multi-source data acquisition module;

[0060] Figure 3 Improved workflow diagram of genetic algorithm for cutting path planning module;

[0061] Figure 4 This is the working principle diagram of the fuzzy adaptive observer of the dynamic deformation compensation module. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] See also Figures 1-4 The present invention provides a system for assisting wire cutting and removing additively molded parts for rocket thrust chambers, aiming to improve the accuracy and efficiency of the wire cutting process. The following is a specific implementation of the system:

[0064] Multi-source data acquisition module: Utilizing various sensors, such as laser displacement sensors, infrared thermal imagers, acoustic emission sensors, and visual cameras, the module collects geometric data of additively molded parts. 3D point cloud registration of the laser displacement sensor data with the visual camera data is performed to construct a surface topography model of the additively molded part. Wavelet packet decomposition of the infrared thermal imager and acoustic emission sensor data is performed to extract thermal stress distribution and material debonding features. A dual-branch feature extraction network is then constructed to extract key features from the surface topography model and thermal stress distribution data, respectively. These features are then fused using a cross-modal feature alignment mechanism to generate a joint feature tensor. Finally, spatial correlation modeling of the joint feature tensor is performed using a gated graph neural network, outputting comprehensive monitoring features that provide data support for subsequent cutting path planning and control.

[0065] Cutting Path Planning Module: Cutting path planning is modeled as a multi-constrained traveling salesman problem, with decision variables including the cutting starting point, transition path, and cutting sequence. The population is initialized and a fitness function is calculated, which comprehensively considers the total path length, corner smoothness, and cutting energy consumption weighting. During the mutation phase, the mutation probability is dynamically adjusted based on the material's thermal conductivity. During the crossover phase, a segmented recombination strategy is employed to retain locally optimal path segments. Furthermore, a Pareto front screening mechanism is introduced to iteratively optimize the set of non-inferior solutions, ultimately outputting a global cutting path sequence to ensure optimal and rational cutting paths.

[0066] Dynamic Deformation Compensation Module: This module constructs a thermal-mechanical coupled state equation, using thermal deformation, vibration acceleration, and material removal rate as state variables. A fuzzy rule library is designed, mapping the state variables to membership function activation weights. Fuzzy rule confidence levels are dynamically updated through an online learning mechanism, and the membership function parameters are optimized using gradient descent. The final output includes compensation control variables, including axial compensation displacement and velocity correction, to compensate for thermal deformation and vibration disturbances during the cutting process in real time, improving cutting accuracy.

[0067] Multi-axis collaborative control module: This module constructs a multi-axis motion constraint graph, with nodes representing the control units of each motion axis and edges representing the synchronization error constraints between axes. The alternating direction multiplier method is used to decompose the global optimization problem into local subproblems, each of which contains axis velocity constraints, acceleration smoothing terms, and synchronization error penalties. Virtual synchronization variables are introduced at the coordination layer to balance conflicting control instructions between axes. A distributed iterative solution is implemented through a consensus protocol, ultimately generating multi-axis motion control instructions that meet dynamic accuracy and synchronization requirements, enabling collaborative multi-axis motion control.

[0068] To further illustrate the present invention in detail, five embodiments are listed below:

[0069] Example 1:

[0070] In this embodiment, specific implementation details of the multi-source data acquisition module are described in detail.

[0071] Laser displacement sensors are used to measure distance information on the surface of AM parts. They emit a laser beam and receive reflected light, acquiring a series of discrete distance data points. A vision camera captures the AM part from different angles, acquiring image information. Using 3D point cloud registration technology, the point cloud data collected by the laser displacement sensor is fused with the image information captured by the vision camera to construct a surface topography model of the AM part. This model accurately reflects the surface geometry and features of the AM part.

[0072] For infrared thermal imagers, their working principle is to obtain temperature distribution information by detecting infrared radiation on the surface of an object. During the cutting process of additively molded parts, the infrared thermal imager monitors the temperature changes in the cutting area in real time to obtain thermal stress distribution data. The acoustic emission sensor is used to capture the acoustic emission signals generated by the changes in the internal structure of the material during the cutting process. These signals contain characteristic information of material peeling. The infrared thermal imager data and the acoustic emission sensor data are processed through wavelet packet decomposition technology. Wavelet packet decomposition is a time-frequency analysis method that can decompose the signal in different frequency bands, so as to more accurately extract the characteristics of the signal. In this embodiment, wavelet packet decomposition is used to extract thermal stress distribution characteristics from the infrared thermal imager data, and material peeling characteristics from the acoustic emission sensor data.

[0073] A two-branch feature extraction network is constructed. The first branch uses a three-dimensional convolutional neural network (3D CNN) to divide the surface morphology model into a multi-resolution voxel grid, where each voxel stores curvature, normal vectors, and height gradient statistics. In the convolution stage, a dilated convolution kernel with a dilation rate of 3 is used. This expands the receptive field of the convolution kernel and captures a wide range of geometric features without increasing the number of parameters. In the residual connection stage, shallow details are fused with deep semantic features, enabling the network to simultaneously learn detailed information and overall features of the surface morphology. Simultaneously, a spatial attention module is introduced to calculate the information entropy weights of feature maps in the channel dimension to screen high-information feature channels; a Gaussian mixture model is constructed in the spatial dimension to locate key geometric defect areas; and the Hadamard product is used to fuse channel weights with spatial weights to achieve adaptive weighting of feature channels and highlight important features.

[0074] The second branch uses a temporal attention network to process the thermal stress distribution data. The temporal attention network can focus on the dynamic changes of the thermal stress distribution and extract its dynamic change characteristics by learning the importance of the thermal stress distribution at different times.

[0075] Finally, a cross-modal feature alignment mechanism fuses the local geometric features and dynamic change features extracted from the two branches to generate a joint feature tensor. Spatial correlation modeling of this joint feature tensor is performed using a gated graph neural network, taking into account the spatial relationships between different parts of the additively molded part. The resulting output includes comprehensive monitoring features such as deformation trends, residual cutting stresses, and material debonding status. These features provide a comprehensive and accurate data foundation for subsequent cutting path planning and control.

[0076] Example 2:

[0077] This embodiment focuses on the specific implementation process of integrating the dynamic mutation operator and the elite retention strategy with the improved genetic algorithm in the cutting path planning module.

[0078] Cutting path planning is modeled as a multi-constrained traveling salesman problem. The decision variables in this problem include the cutting starting point, transition path, and cutting sequence. The choice of the cutting starting point directly affects the starting position and ease of the entire cutting process. The transition path determines the movement between different cutting areas, requiring consideration of path length and corner smoothness. The cutting sequence determines the order in which the various parts of the additive manufacturing part are cut, significantly impacting cutting efficiency and quality.

[0079] When initializing the population, a certain number of cutting path solutions are randomly generated as the initial population. For each path solution, its fitness function is calculated. The fitness function includes the total path length, corner smoothness, and cutting energy weight. The total path length reflects the actual distance traveled by the electrode wire during the cutting process. A shorter path can reduce cutting time and electrode wire wear. Corner smoothness is used to measure the changes in each corner in the path. Smooth corners can reduce electrode wire wear and vibration during the cutting process. The cutting energy weight takes into account the energy consumption during the cutting process. By setting the weight appropriately, cutting energy consumption can be optimized while ensuring cutting quality.

[0080] During the mutation phase, a dynamic mutation operator is implemented based on an environmental response mechanism. Specifically, heat accumulation and path curvature from historical cutting processes are collected as mutation factors. Heat accumulation reflects the heat accumulation in the cutting area; higher heat accumulation can cause material deformation and reduced cutting accuracy. Path curvature indicates the degree of path curvature; higher curvature can increase electrode wire wear and make cutting more difficult. A radial basis function neural network is constructed to map the mutation probability to environmental parameters (heat accumulation and path curvature). The mutation intensity is dynamically adjusted using an online reinforcement learning strategy. When local overheating is detected, meaning that heat accumulation exceeds a certain threshold, a targeted mutation operation is triggered. This mutation operation allows for targeted adjustments to the cutting path, avoiding unnecessary cutting in overheated areas, thereby improving cutting quality and efficiency.

[0081] During the crossover phase, a segmentation and reassembly strategy is used to preserve locally optimal path fragments. Specifically, the two parent paths are divided into segments according to a specific rule, and then some of these segments are swapped to generate child paths. During this swapping process, local segments with good fitness from the parent paths are retained, which accelerates the algorithm's convergence and improves search efficiency.

[0082] A Pareto front screening mechanism is introduced to iteratively optimize the set of non-inferior solutions. In a multi-objective optimization problem, the Pareto front is the set of solutions that cannot be simultaneously surpassed by other solutions on all objectives. Through continuous iteration, better non-inferior solutions are screened, ultimately outputting a global cutting path sequence that strikes a good balance between total path length, corner smoothness, and cutting energy consumption.

[0083] Example 3:

[0084] This embodiment describes in detail a specific implementation method of the fuzzy adaptive observer in the dynamic deformation compensation module using a double closed-loop compensation structure.

[0085] Construct the thermal-mechanical coupled state equation and transform the thermal deformation , vibration acceleration and material removal rate As a state variable. Thermal deformation is the deformation of the additively molded part caused by heat generated during the cutting process, which directly affects the cutting accuracy. Vibration acceleration reflects the vibration during the cutting process. Excessive vibration will lead to a decrease in the quality of the cut surface. The material removal rate indicates the amount of material removed per unit time and is closely related to the cutting speed and cutting efficiency. The thermal-mechanical coupling state equation can be expressed as:

[0086]

[0087] in, is the derivative vector of the state variables, is the state matrix, which describes the dynamic relationship between state variables; is the input matrix, and the control input Related; is the system noise vector.

[0088] Design a fuzzy rule base and transform the state variables 、 、 The fuzzy rule base contains a series of fuzzy rules, such as: "If the thermal deformation Large vibration acceleration If the state variable is larger, the compensation control amount should be increased accordingly. Each fuzzy rule has a corresponding membership function, which is used to determine the degree to which the state variable belongs to a fuzzy set. Through these membership functions, the state variables are mapped to membership function activation weights.

[0089] The confidence levels of fuzzy rules are dynamically updated through an online learning mechanism, and the membership function parameters are optimized using the gradient descent method. This online learning mechanism continuously adjusts the confidence levels of fuzzy rules based on data from the actual cutting process, making them more consistent with the actual situation. Gradient descent is a commonly used optimization algorithm that gradually optimizes the objective function by calculating the gradient of the objective function with respect to its parameters and updating the parameters in the opposite direction of the gradient. In this embodiment, the membership function parameters are optimized using the gradient descent method, enabling the fuzzy adaptive observer to better adapt to changes in the cutting process.

[0090] The dual closed-loop compensation structure consists of a feedforward compensation loop and a feedback correction loop. The feedforward loop predicts deformation based on the heat conduction equation. The heat conduction equation describes how heat is conducted through an object. By solving this equation, the thermal deformation of the additively molded part during the cutting process can be predicted. A sliding mode controller is used to generate pre-compensation. A sliding mode controller is a nonlinear controller with the advantage of being insensitive to system parameter changes and external interference. Using this controller, pre-compensation is generated based on the predicted deformation, allowing for proactive compensation of potential deformation.

[0091] The feedback loop uses a Kalman filter to fuse the vibration sensor data with the visual measurement data to generate an error correction. A Kalman filter is an optimal estimator that provides an optimal estimate of the system state based on the system's measurement data and model predictions. In this embodiment, the Kalman filter fuses the vibration data measured by the vibration sensor and the deformation data measured by the visual measurement to estimate the actual deformation error and generate the error correction.

[0092] Design a hybrid compensation strategy, add the pre-compensation amount and the error correction amount together to output the final compensation control amount. Specifically, the final compensation control amount It can be expressed as:

[0093]

[0094] in, is the pre-compensation amount, is the error correction amount, is the weighting coefficient, and its value range is By reasonably adjusting the weighting coefficient , which can make the compensation control amount more accurately compensate for the thermal deformation and vibration interference in the cutting process and improve the cutting accuracy.

[0095] Example 4:

[0096] This embodiment focuses on the specific implementation process of the distributed constraint optimization model in the multi-axis cooperative control module using a hierarchical interactive architecture.

[0097] A multi-axis motion constraint graph is constructed, where nodes represent the control units for each motion axis, and edges represent the inter-axis synchronization error constraints. The wire cutting process for removing additively molded parts for rocket thrust chambers typically involves the coordinated motion of multiple axes, such as the X, Y, and Z axes. Each motion axis control unit is responsible for controlling the motion of its corresponding axis, while the inter-axis synchronization error constraints ensure that the motion of each axis remains synchronized, avoiding cutting errors caused by uncoordinated motion.

[0098] The alternating direction multiplier method is used to decompose the global optimization problem into local sub-problems. The goal of the global optimization problem is to generate multi-axis motion control instructions that meet dynamic accuracy and synchronization while taking into account various constraints. Through the alternating direction multiplier method, this complex global problem is decomposed into multiple relatively simple local sub-problems. Each local sub-problem contains an axis speed limit, an acceleration smoothing term, and a synchronization error penalty term. The axis speed limit is used to ensure that the speed of the moving axis is within a safe and reasonable range to avoid damage to the equipment or reduced cutting accuracy due to excessive speed; the acceleration smoothing term makes the acceleration changes of the moving axis smoother, reducing vibration and impact; the synchronization error penalty term is used to penalize the synchronization error between axes, prompting better synchronization between the axes.

[0099] Virtual synchronization variables are introduced in the coordination layer to balance the conflicts of control instructions of each axis. Since the movement of each motion axis may be subject to different constraints and interferences, conflicts may occur between control instructions. Virtual synchronization variables coordinate these conflicts through a certain mechanism so that each axis can work together. In this embodiment, the virtual synchronization variables adopt a time-delay compensation mechanism. Specifically, the transmission delay of the control instructions of each axis is measured, and a time-varying communication topology matrix is ​​constructed. There will be a certain delay in the transmission process of the control instructions, and the delay of different axes may be different. These delays will affect the synchronization between the axes. By measuring the delay and constructing the time-varying communication topology matrix, the communication relationship between the axes can be accurately described.

[0100] A time delay compensation term is introduced into the consistency protocol, and the compensation coefficient is determined using the Lyapunov stability criterion. A consistency protocol is a distributed algorithm used to achieve a consistent state among multiple nodes. In this embodiment, by introducing a time delay compensation term into the consistency protocol, the impact of control instruction transmission delay on inter-axis synchronization can be compensated. The Lyapunov stability criterion is a method for determining system stability. By applying the Lyapunov stability criterion, an appropriate compensation coefficient can be determined to ensure that the system can still operate stably while taking time delay into account.

[0101] The above steps ultimately generate multi-axis motion control instructions that meet dynamic accuracy and synchronization requirements. These instructions precisely control the motion of each axis, enabling the wire EDM to efficiently and accurately cut additively molded parts along the predetermined cutting path, improving cutting quality and efficiency.

[0102] Example 5:

[0103] This embodiment comprehensively demonstrates the collaborative working process of the above modules in the process of wire cutting and removing the additively molded parts for the actual rocket thrust chamber. Assume that a complex-shaped rocket thrust chamber additively molded part needs to be wire cut and removed.

[0104] The multi-source data acquisition module begins operating. Laser displacement sensors, infrared thermal imagers, acoustic emission sensors, and visual cameras simultaneously collect data from the additively molded parts. The information captured by the laser displacement sensors and visual camera is then aligned with the 3D point cloud to construct a surface topography model. Data from the infrared thermal imager and acoustic emission sensors are decomposed using wavelet packets to extract thermal stress distribution and material delamination features. A dual-branch feature extraction network extracts key features from the surface topography model and thermal stress distribution data, respectively, and fuses them to generate a joint feature tensor. This is then processed by a gated graph neural network to output comprehensive monitoring features.

[0105] These comprehensive monitoring features are transmitted to the cutting path planning module. This module models cutting path planning as a multi-constrained traveling salesman problem and solves it using an improved genetic algorithm. During the solution process, the mutation probability is dynamically adjusted based on the material's thermal conductivity, a segmented recombination strategy is used for crossover operations, and a Pareto front screening mechanism is used to optimize the set of non-inferior solutions, ultimately generating a global cutting path sequence.

[0106] The dynamic deformation compensation module uses a thermal-mechanical coupled state equation and a fuzzy rule base to model thermal deformation and vibration disturbances during the cutting process in real time. Compensation control variables are generated through feedforward compensation and feedback correction loops. The feedforward loop predicts deformation based on the heat conduction equation, and a sliding mode controller generates pre-compensation variables. The feedback loop utilizes a Kalman filter to fuse vibration sensor and visual measurement data to generate error correction variables. The weighted superposition of these two factors yields the final compensation control variable.

[0107] The multi-axis collaborative control module constructs a distributed constraint optimization model based on the global cutting path sequence and compensation control variables. By constructing a multi-axis motion constraint graph, the optimization problem is decomposed using the alternating direction multiplier method. Virtual synchronization variables are introduced, and a time-delay compensation mechanism is utilized to achieve distributed iterative solution and generate multi-axis motion control instructions.

[0108] Throughout the cutting process, the multi-source data acquisition module continuously monitors changes in the state of the additively molded parts and promptly transmits new data to other modules. The cutting path planning module re-evaluates and optimizes the cutting path based on the new data; the dynamic deformation compensation module adjusts the compensation control amount in real time; and the multi-axis collaborative control module adjusts the multi-axis motion control instructions based on the updated information, ensuring that the wire cutting equipment can adapt to the various changes in the additively molded parts during the cutting process, achieving high-precision and high-efficiency cutting and removal. Through the collaborative work of these modules, the quality and efficiency of wire cutting and removal of additively molded parts for rocket thrust chambers have been effectively improved, meeting actual production needs.

[0109] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0110] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A wire cutting and removing auxiliary system for additively molded parts for rocket thrust chambers, characterized in that: include: Multi-source data acquisition module: used to collect geometric data of additively molded parts through multiple types of sensors; Cutting path planning module: performing topological optimization analysis on the geometric shape data based on an improved genetic algorithm to generate a global cutting path sequence; Dynamic deformation compensation module: uses fuzzy adaptive observer to model thermal deformation and vibration interference in the cutting process in real time and output compensation control value; Multi-axis collaborative control module: constructs a distributed constraint optimization model based on the global cutting path sequence and compensation control amount, and generates multi-axis motion control instructions; The multi-source data acquisition module includes: Multiple types of sensors include laser displacement sensors, infrared thermal imagers, acoustic emission sensors, and visual cameras; Perform 3D point cloud registration on laser displacement sensor data and visual camera data to construct a surface topography model of the additively molded part; perform wavelet packet decomposition on infrared thermal imager data and acoustic emission sensor data to extract thermal stress distribution and material peeling characteristics; A two-branch feature extraction network was constructed. The first branch used a three-dimensional convolutional neural network to extract the local geometric features of the surface morphology model, and the second branch used a temporal attention network to extract the dynamic change characteristics of the thermal stress distribution. The local geometric features and dynamic change features are fused through a cross-modal feature alignment mechanism to generate a joint feature tensor. The joint feature tensor is spatially correlated and modeled based on a gated graph neural network to output comprehensive monitoring features including deformation trend, cutting residual stress, and material peeling status.

2. The system for removing parts by wire cutting of additively molded parts for rocket thrust chambers according to claim 1, characterized in that: The improved genetic algorithm integrates a dynamic mutation operator and an elite retention strategy, specifically including: The cutting path planning is modeled as a multi-constrained traveling salesman problem, where the decision variables include the cutting starting point, transition path, and cutting order. Initialize the population and calculate the fitness function, which includes the total path length, corner smoothness, and cutting energy consumption weight terms; In the mutation phase, the mutation probability is dynamically adjusted according to the thermal conductivity of the material; in the crossover phase, a segmented reorganization strategy is used to retain the local optimal path segments; The Pareto front screening mechanism is introduced to iteratively optimize the non-inferior solution set and finally output the global cutting path sequence.

3. The system for removing parts by wire cutting of additively molded parts for rocket thrust chambers according to claim 1, characterized in that: The fuzzy adaptive observer adopts a double closed-loop compensation structure, specifically including: Construct a thermal-mechanical coupling state equation, taking thermal deformation, vibration acceleration, and material removal rate as state variables; Design a fuzzy rule base to map state variables into membership function activation weights; The confidence of fuzzy rules is dynamically updated through online learning mechanism, and the membership function parameters are optimized using gradient descent method; The output includes the compensation control amount of axial compensation displacement and speed correction amount.

4. The system for removing parts by wire cutting of additively molded parts for rocket thrust chambers according to claim 1, characterized in that: The distributed constraint optimization model adopts a hierarchical interactive architecture, specifically including: Construct a multi-axis motion constraint graph, where nodes represent the control units of each motion axis and edges represent the synchronization error constraints between axes; The alternating direction multiplier method is used to decompose the global optimization problem into local subproblems. Each subproblem contains axis speed limit, acceleration smoothing term and synchronization error penalty term. Introducing virtual synchronization variables at the coordination layer to balance conflicts in control instructions between axes, and implementing distributed iterative solutions through a consistency protocol; Finally, multi-axis motion control instructions that meet dynamic accuracy and synchronization are generated.

5. The system for removing parts by wire cutting of additively molded parts for rocket thrust chamber according to claim 1, characterized in that: The three-dimensional convolutional neural network adopts a hollow residual structure, specifically including: The surface topography model is divided into a multi-resolution voxel grid, each voxel stores curvature, normal vector and height gradient statistics; In the convolution stage, a dilation rate of 3 is used to capture large-scale geometric features, and the residual connection stage fuses shallow details with deep semantic features. A spatial attention module is introduced to adaptively weight feature channels.

6. The system for assisting wire cutting and removing parts of additively molded parts for rocket thrust chambers according to claim 2, characterized in that: The dynamic mutation operator is implemented based on the environment response mechanism, specifically including: The heat accumulation and path curvature during the historical cutting process are collected as variation factors; Construct a radial basis function neural network to fit the mapping relationship between mutation probability and environmental parameters; The mutation intensity is dynamically adjusted through an online reinforcement learning strategy, and a directed mutation operation is triggered when local overheating is detected.

7. The system for assisting wire cutting and removing additively molded parts for a rocket thrust chamber according to claim 3, characterized in that: The dual closed-loop compensation structure includes a feedforward compensation and a feedback correction loop, specifically including: The feedforward loop predicts the deformation based on the heat conduction equation and uses a sliding mode controller to generate pre-compensation; The feedback loop fuses the vibration sensor data with the visual measurement data through a Kalman filter to generate an error correction; A hybrid compensation strategy is designed to output the final compensation control amount after weighted superposition of the pre-compensation amount and the error correction amount.

8. The system for removing parts by wire cutting of additively molded parts for rocket thrust chambers according to claim 4, characterized in that: The virtual synchronous variable adopts a time lag compensation mechanism, specifically including: Measure the transmission delay of each axis control command and build a time-varying communication topology matrix; A time delay compensation term is introduced into the consensus protocol, and the compensation coefficient is determined by the Lyapunov stability criterion.

9. The system for assisting wire cutting and removing additively molded parts for a rocket thrust chamber according to claim 5, characterized in that: The spatial attention module adopts a channel-position joint focusing mechanism, specifically including: Calculate the information entropy weight of the feature map in the channel dimension and select high-information feature channels; Construct a Gaussian mixture model in the spatial dimension to locate key geometric defect areas; The channel weights and spatial weights are fused via the Hadamard product.

Images