Forming bin flue gas exhaust system for additive manufacturing of rocket engine thrust chamber

Abstract

The invention relates to the technical field of rocket engine thrust chamber additive manufacturing, and discloses a forming bin smoke exhaust system for rocket engine thrust chamber additive manufacturing. According to the system, the multi-source data acquisition module acquires smoke multi-dimensional parameters, the smoke feature extraction module decomposes features, the dynamic filtering module generates a filtering strategy and a gas flow optimization matrix, the cooperative control module coordinates equipment operation parameters to achieve energy consumption balance, and the abnormity processing module identifies abnormity and triggers a self-repairing mechanism. The system can accurately collect and analyze flue gas data, dynamically optimize filtering and airflow paths, efficiently cooperate with equipment operation, handle abnormities in time, improve the flue gas removal effect, reduce energy consumption, guarantee safety of operators and stable operation of equipment, and promote development of the rocket engine thrust chamber additive manufacturing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing of rocket engine thrust chambers, and specifically to a flue gas exhaust system for an additive manufacturing forming chamber of a rocket engine thrust chamber. Background Technique

[0002] In the modern aerospace field, as a core power device, the performance and quality of rocket engines play a decisive role in the success or failure of space missions. With the continuous progress of technology, additive manufacturing technology, with its unique advantages such as strong complex structure manufacturing ability, high material utilization rate, and short production cycle, has gradually been applied in the manufacturing of rocket engine thrust chambers. However, a large amount of complex flue gas is generated during this manufacturing process, bringing many challenges to the environmental control in the forming chamber and the operation of equipment, which makes an efficient flue gas exhaust system for the forming chamber a key link in the development of additive manufacturing technology for rocket engine thrust chambers.

[0003] The additive manufacturing process of a rocket engine thrust chamber involves the high-temperature melting and layer-by-layer stacking of various materials, and this process will generate flue gas containing complex components. From the perspective of temperature, the temperature distribution in the forming chamber is extremely uneven, with a significant temperature gradient. The high temperature in different regions will not only affect the forming quality of materials, but may also cause thermal damage to the key components of the equipment, reducing the service life of the equipment. For example, during the additive manufacturing of some metal materials, excessive local temperature may lead to coarse grains of the material, affecting the mechanical properties of the product.

[0004] In terms of particulate matter, the concentration and particle size distribution of particulate matter in the flue gas are complex. Small particle size particulate matter is easily adsorbed on the surface of the equipment, clogging the precision components of the equipment, such as sensors, filters, etc., resulting in a decrease in the measurement accuracy of the equipment or even failure. Large particle size particulate matter may deposit at the bottom of the forming chamber under the action of gravity, affecting the normal operation of the equipment and increasing the cleaning and maintenance costs.

[0005] The gas components cannot be ignored either. During the additive manufacturing process, various harmful gases such as carbon monoxide and nitrogen oxides are generated. These gases not only pose a threat to the health of operators, but may also react chemically with the metal components in the forming chamber, accelerating the corrosion of the components and reducing the reliability and safety of the equipment.

[0006] Existing flue gas exhaust technologies have many deficiencies in dealing with these problems. Traditional filtration methods often use fixed filter media and a single filtration structure, and cannot be dynamically adjusted according to the real-time changes of the flue gas. When the particulate matter concentration in the flue gas suddenly increases or the gas components change, the filtration effect will drop significantly, making it difficult to meet the production requirements. Moreover, when controlling the exhaust fan, filtration device, and cooling unit in the traditional system, there is a lack of an effective coordination mechanism, and the operating parameters between various equipment cannot achieve an optimal match, resulting in high system energy consumption and low operating efficiency.

[0007] In addition, in terms of abnormal situation handling, most of the existing technologies rely on manual inspections and post-fault repairs. When abnormal situations such as flue gas leakage and equipment failures occur, they cannot be detected and handled in a timely and accurate manner, which is likely to cause production interruptions, increase production costs and production cycles. Moreover, for some potential risks, such as the life prediction of filter media and the prevention of toxic gases, the existing technologies also lack effective means and it is difficult to ensure the safe and stable operation of the production process. In summary, it is extremely urgent to develop an efficient, intelligent and stable flue gas exhaust system for the forming chamber used in the additive manufacturing of rocket engine thrust chambers. Summary of the Invention

[0008] The purpose of the present invention is to provide a flue gas exhaust system for the forming chamber used in the additive manufacturing of rocket engine thrust chambers to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solution: A flue gas exhaust system for the forming chamber used in the additive manufacturing of rocket engine thrust chambers, the system includes: Multi-source data acquisition module: used to obtain multi-dimensional parameters of the flue gas in the forming chamber in real time, including temperature gradient distribution, particulate matter concentration spectrum, gas component ratio and air flow velocity field; Flue gas feature extraction module: perform non-linear feature decomposition on the multi-dimensional parameters to generate flue gas diffusion mode features, particulate matter aggregation trend features and thermodynamic coupling response features; Dynamic filtration module: construct a multi-layer filtration network based on an improved adaptive filtering algorithm, input the features into the dynamic weight allocation unit, and generate a filter medium switching strategy and an air flow path optimization matrix; Collaborative control module: based on the filter medium switching strategy and the air flow path optimization matrix, coordinate the operating parameters of the exhaust fan, filtration device and cooling unit through a distributed reinforcement learning model, and generate a global energy consumption balance control instruction; Abnormal handling module: identify abnormal fluctuations in the flue gas exhaust process based on the density peak detection algorithm, trigger the self-repair mechanism and update the topology structure of the filtration network.

[0010] Preferably, the multi-source data acquisition module includes: Use a distributed optical fiber sensing array to capture the temperature gradient distribution and reconstruct the three-dimensional thermal field evolution map through a spatio-temporal interpolation algorithm; Use polarization scattering measurement technology to separate the distribution density of particles with different particle sizes for the particulate matter concentration spectrum, and combine the Kalman filtering algorithm to correct the measurement error; Use the spectral absorption feature fusion method to extract the proportion of key components for the gas component ratio, and predict the gas diffusion boundary based on the chaotic mapping model.

[0011] Preferably, the flue gas characteristic extraction module further includes: Using a vortex decomposition algorithm to extract the turbulent core region from the air velocity field, and correlating the particulate matter movement trajectory through phase-locking technology; Modeling the temperature-airflow interaction of the thermodynamic coupling response characteristics by using a tensor decomposition method, and generating a multi-scale energy transfer path characteristic matrix.

[0012] Preferably, the dynamic filtration module further includes: Constructing a prediction model for the service life of the filter medium based on a double-layer gated recurrent unit, and calculating the attenuation rate of the porosity of the filter layer in real time; Optimizing the filter medium switching strategy through the gradient backpropagation algorithm, and generating a dynamic pressure drop compensation instruction by fusing the airflow path optimization matrix.

[0013] Preferably, the collaborative control module further includes: Defining the rotational speed of the exhaust fan, the pressure difference threshold of the filtration device, and the power threshold of the cooling unit as the joint action space, and using a multi-objective Pareto optimization algorithm to solve the global optimal parameter combination; Balancing the conflict of local control instructions through an asynchronous policy update mechanism.

[0014] Preferably, the exception handling module further includes: Dividing the risk level of the abnormal fluctuation by using a density clustering algorithm, generating a graded warning signal and associating the trigger condition of the self-repair mechanism; When the particulate matter concentration is detected to exceed the limit, activating the redundant filter layer deployment strategy and reconstructing the priority weight of the airflow path optimization matrix.

[0015] Preferably, the self-repair mechanism further includes: Modeling the coupling relationship between the filtration device and the exhaust fan by using a bidirectional graph neural network, extracting the fault conduction characteristics from the forward path, and extracting the redundant resource scheduling characteristics from the reverse path; Generating a repair plan through attention weight fusion, and updating the risk assessment parameters of the dynamic filtration network.

[0016] Preferably, the system further includes: Constructing a flue gas toxicity index prediction model, based on historical process data and real-time gas component parameters, using a temporal convolutional network to fuse the generation law of thermal degradation products, and dynamically adjusting the chemical adsorption priority of the filter medium.

[0017] Preferably, the flue gas toxicity index prediction model further includes: Using a generative adversarial network to simulate the gas mixing distribution under extreme conditions, generating a toxicity risk probability map and embedding it into the redundant filter layer deployment strategy.

[0018] Preferably, the treatment of the turbulent core region further includes: An adaptive vortex tracking algorithm is used to balance the measurement accuracy of high-speed airflows and low-speed stagnant zones. The nonlinear characteristics of vortex evolution are captured through a residual convolutional network, and an airflow stability evaluation matrix is constructed and integrated into the weight assignment unit of the dynamic filtering network.

[0019] Compared with the prior art, the beneficial effects of the present invention are: The flue gas exhaust system for the additive manufacturing forming chamber of the rocket engine thrust chamber of the present invention has many significant beneficial effects. In terms of data acquisition and analysis, the multi-source data acquisition module can real-time obtain multi-dimensional parameters such as the temperature gradient distribution, particulate concentration spectrum, gas component ratio, and airflow velocity field of the flue gas in the forming chamber. The distributed optical fiber sensing array is used to capture the temperature gradient distribution and reconstruct the three-dimensional thermal field evolution map, which can accurately grasp the temperature change trend in the forming chamber, provide strong data support for preventing material forming defects and equipment thermal damage caused by uneven temperature, and ensure the stability of the additive manufacturing process and product quality. Using the polarization scattering measurement technology combined with the Kalman filtering algorithm to measure the particulate concentration spectrum can more accurately master the particulate distribution, effectively avoid failures caused by particulate blockage of equipment components, and reduce equipment maintenance costs and downtime. The spectral absorption feature fusion method combined with the chaotic mapping model is used to obtain the gas component ratio and predict the diffusion boundary, which helps to take preventive measures in advance to ensure the safety of operators and the normal operation of equipment.

[0020] In the filtering and control link, the dynamic filtering module constructs a multi-layer filtering network based on an improved adaptive filtering algorithm, and generates a filtering medium switching strategy and an airflow path optimization matrix according to the flue gas characteristics. By constructing a filtering medium life prediction model based on a double-layer gated recurrent unit, the porosity decay rate of the filtering layer can be calculated in real time, the filtering medium can be switched in time, the high-efficiency filtering performance can be maintained, and the service life of the filtering device can be extended. The gradient backpropagation algorithm optimizes the filtering medium switching strategy and generates a dynamic pressure drop compensation instruction, which can ensure the stable operation of the system under different working conditions and reduce energy consumption. The cooperative control module coordinates the operating parameters of the exhaust fan, filtering device, and cooling unit with the help of a distributed reinforcement learning model, and uses the multi-objective Pareto optimization algorithm to solve the global optimal parameter combination, realizing global energy consumption balanced control, improving the overall operating efficiency of the system, effectively reducing energy consumption while ensuring the flue gas exhaust effect, and conforming to the concept of green manufacturing.

[0021] The ability to handle exceptions is also a major advantage of this system. The exception handling module identifies abnormal fluctuations based on the density peak detection algorithm, uses the density clustering algorithm to classify risk levels and generate hierarchical warning signals, and correlates with the trigger conditions of the self-healing mechanism. Once an anomaly is detected, the self-healing mechanism is quickly activated, and the bidirectional graph neural network models the coupling relationship between the filtering device and the exhaust fan, generates a repair plan and updates the risk assessment parameters, ensuring that the system can still operate stably under abnormal conditions, reducing the number of production interruptions, and improving production efficiency. At the same time, the flue gas toxicity index prediction model dynamically adjusts the chemical adsorption priority of the filtering medium according to real-time data, uses a generative adversarial network to simulate the gas mixing distribution under extreme conditions, generates a toxicity risk probability map and embeds it into the redundant filtering layer deployment strategy, greatly enhancing the system's ability to prevent toxic gases and ensuring the safety of operators and the reliability of the production environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is the working principle diagram of a flue gas exhaust system for an additive manufacturing forming chamber of a rocket engine thrust chamber according to the present invention; Figure 2 is the working flow chart of the multi-source data acquisition module; Figure 3 is the flow chart of the flue gas feature extraction module for processing the coupling response characteristics of the air flow velocity field and thermodynamics; Figure 4 is the working flow chart of the self-healing mechanism. DETAILED DESCRIPTION OF THE INVENTION

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] Please refer to Figures 1 - 4 , the present invention relates to a flue gas exhaust system for an additive manufacturing forming chamber of a rocket engine thrust chamber, and its specific implementation manner will be elaborated in detail below.

[0025] The multi-source data acquisition module is responsible for obtaining multi-dimensional parameters of the flue gas in the forming chamber in real time, including temperature gradient distribution, particulate matter concentration spectrum, gas component ratio, and air flow velocity field. Through the acquisition of these parameters, basic data is provided for subsequent analysis and processing.

[0026] The collected data is transmitted to the flue gas feature extraction module. This module performs non-linear feature decomposition on multi-dimensional parameters to generate flue gas diffusion pattern features, particulate matter aggregation trend features, and thermodynamic coupling response features, thereby excavating the key feature information of the flue gas.

[0027] The dynamic filtering module constructs a multi-layer filtering network based on an improved adaptive filtering algorithm. The features extracted above are input into the dynamic weight allocation unit to generate a filtering medium switching strategy and an air flow path optimization matrix, thereby achieving efficient filtering of the flue gas and reasonable planning of the air flow path.

[0028] The collaborative control module, based on the filtering medium switching strategy and the air flow path optimization matrix, coordinates the operating parameters of the exhaust fan, the filtering device, and the cooling unit through a distributed reinforcement learning model to generate a global energy consumption balance control instruction, ensuring reasonable distribution of energy consumption while the system operates efficiently.

[0029] The anomaly handling module identifies abnormal fluctuations during the flue gas exhaust process based on the density peak detection algorithm. Once an anomaly is detected, it triggers a self-repair mechanism and updates the filtering network topology structure to ensure the stable operation of the system.

[0030] The present invention will be further described in detail below through specific embodiments.

[0031] Embodiment 1: The implementation method of the multi-source data acquisition module includes: For the acquisition of the temperature gradient distribution, a distributed optical fiber sensing array is used. The distributed optical fiber sensing array can perform multi-point temperature monitoring in the forming bin. Its working principle is to utilize the optical time domain reflection characteristics of the optical fiber. When light propagates in the optical fiber, a change in the reflected light intensity will occur when encountering a temperature change. By analyzing the reflected light intensity, the temperature information at different positions can be obtained. In order to more intuitively understand the temperature distribution and change trend, it is necessary to reconstruct the three-dimensional thermal field evolution map through a spatio-temporal interpolation algorithm. The spatio-temporal interpolation algorithm is based on the collected discrete temperature data points and uses the methods of spatial interpolation and time series analysis to construct a map of the temperature change with time in a three-dimensional space. Assume that the collected temperature data points are , where represents the temperature of the th data point, represents the spatial coordinates of this data point in the forming bin, represents the acquisition time. The spatio-temporal interpolation algorithm obtains the temperature at any position in the entire forming bin at any time through fitting and extrapolation of these discrete points, thereby realizing the reconstruction of the three-dimensional thermal field evolution map.

[0032] In the measurement of the particulate matter concentration spectrum, the polarization scattering measurement technique is adopted. The polarization scattering measurement technique utilizes the differences in the scattering characteristics of polarized light by particles of different sizes to separate the distribution densities of particles of different sizes. When polarized light irradiates on particles, particles of different sizes will generate scattered light with different polarization states and intensities. By analyzing the polarization state and intensity of the scattered light, the distribution of particles of different sizes can be determined. To improve the measurement accuracy, the Kalman filtering algorithm is combined to correct the measurement error. The Kalman filtering algorithm is a filtering method based on linear minimum mean square estimation, which can perform optimal estimation on the measurement data according to the state equation and observation equation of the system. Assume that the true value of the particulate matter concentration is , the measured value is , and the measurement error is . The Kalman filtering algorithm continuously updates and predicts the measurement data to obtain an estimate closer to the true value, and its core formula is: Prediction equation: where is the predicted value at time based on the estimated value at time, is the state transition matrix, is the optimal estimated value at time, is the control matrix,

[0033] Update equation: where is the optimal estimated value at time, is the Kalman gain, is the measured value at time,

[0034] For the acquisition of the gas component ratio, the spectral absorption feature fusion method is adopted. Different gases have unique absorption characteristics within specific spectral ranges. By measuring the absorption intensities of flue gas in multiple spectral bands and combining the spectral absorption feature fusion algorithm, the proportion of key components can be extracted. For example, for a certain gas component , its absorption intensities at wavelengths are respectively , and through a specific fusion algorithm the proportion . To understand the gas diffusion range in advance, the gas diffusion boundary is predicted based on the chaotic mapping model. The chaotic mapping model is a non-linear dynamics model that can describe the complex non-linear behavior during the gas diffusion process. Assume the initial state of gas diffusion is , through the chaotic mapping function , the diffusion states at different times can be iteratively calculated , thus predicting the gas diffusion boundary.

[0035] Example 2: This example mainly focuses on the processing of the air flow velocity field and the modeling of the thermodynamic coupling response characteristics in the flue gas characteristic extraction module.

[0036] In the processing of the air flow velocity field, the vortex decomposition algorithm is used to extract the turbulent core region. The vortex decomposition algorithm is based on the vorticity concept in fluid mechanics. By calculating the curl of the air flow velocity field, the position and intensity of the vortices are determined, thereby identifying the turbulent core region. By analyzing the vorticity field, the region with larger vorticity is found, which is the turbulent core region.

[0037] To further understand the movement of particulate matter in the turbulent core region, the particulate matter movement trajectory is correlated through the phase-locking technique. The phase-locking technique uses a certain periodic signal in the air flow as a reference and synchronously analyzes the movement of the particulate matter with this periodic signal. For example, taking the rotation period of the vortex as the reference signal, the position change of the particulate matter in each period is recorded, thereby obtaining the movement trajectory of the particulate matter in the turbulent core region. Assume the rotation period of the vortex is , in the th period, the position of the particulate matter at time is . By continuously recording the positions of the particulate matter in multiple periods, its movement trajectory can be plotted.

[0038] For the thermodynamic coupling response characteristics, the tensor decomposition method is used to model the temperature-air flow interaction. The tensor decomposition method can decompose the high-dimensional temperature-air flow data into a combination of multiple low-dimensional tensors, thus more clearly describing the interaction relationship between them. Assume the temperature field is , the air flow velocity field is , and they are combined into a tensor . Through the tensor decomposition algorithm, is decomposed into , where is the rank of the decomposition, are the coefficients, , , , are respectively in the space , , and time low-dimensional tensors in the direction. Through this decomposition, a multi-scale energy transfer path feature matrix can be generated, which reflects the energy transfer relationship between temperature and air flow at different scales and provides an important basis for subsequent analysis and processing.

[0039] Example 3: In the dynamic filtration module, a prediction model for the service life of the filtration medium based on a double-layer gated recurrent unit is constructed. The gated recurrent unit (GRU) is a special structure of recurrent neural network that can effectively handle the long-term dependence problem in time series data. The double-layer gated recurrent unit further enhances the learning ability of the model. This model takes the historical operation data of the filtration layer as input, including information such as filtration time, flue gas volume passed, particulate matter concentration, etc., and calculates the attenuation rate of the porosity of the filtration layer in real time. Assume that the porosity of the filtration layer is , the time is , and the attenuation rate of the porosity of the filtration layer can be calculated by the double-layer gated recurrent unit model , where represents the input historical operation data.

[0040] To optimize the filtration medium switching strategy, it is optimized by the gradient backpropagation algorithm. The gradient backpropagation algorithm is an effective method for training neural networks. It calculates the gradient of the loss function with respect to the model parameters and updates the parameters along the opposite direction of the gradient, so that the loss function gradually decreases. In this system, taking filtration efficiency and filtration cost as optimization objectives, a loss function is constructed. Assume that the parameters related to the filtration medium switching strategy are , calculate by the gradient backpropagation algorithm, and update the parameter according to this gradient, so as to optimize the filtration medium switching strategy.

[0041] While optimizing the filtration medium switching strategy, a dynamic pressure drop compensation instruction is generated by fusing the air flow path optimization matrix. The purpose of the dynamic pressure drop compensation instruction is to ensure that during the filtration process, the pressure drop of the system remains within a reasonable range, and to avoid excessive or too small pressure drop caused by the switching of the filtration medium or the change of the air flow path. Assume that the air flow path optimization matrix is , and a dynamic pressure drop compensation instruction is generated through a specific fusion algorithm , and this instruction is used to adjust the working parameters of the filtration device, such as filtration speed, filtration area, etc., to achieve dynamic pressure drop compensation.

[0042] Example 4: The key task of the collaborative control module is to coordinate the operating parameters of the exhaust fan, filtration device, and cooling unit to achieve global energy consumption balance control. First, define the rotational speed of the exhaust fan , the differential pressure threshold of the filtration device , and the power threshold of the cooling unit as the joint action space. The rotational speed of the exhaust fan directly affects the exhaust speed of the flue gas. The higher the rotational speed, the faster the exhaust speed, but the energy consumption also increases accordingly; the differential pressure threshold of the filtration device determines the working state of the filtration device. When the differential pressure exceeds this threshold, it may be necessary to switch the filtration medium or adjust the operating parameters of the filtration device; the power threshold of the cooling unit controls the refrigeration capacity of the cooling unit to ensure that the system operates within an appropriate temperature range.

[0043] To solve for the global optimal parameter combination, a multi-objective Pareto optimization algorithm is adopted. The multi-objective Pareto optimization algorithm can find a set of optimal solutions among multiple conflicting objectives, and these solutions achieve a balance among different objectives. In this system, the objectives include reducing energy consumption, improving filtration efficiency, ensuring system stability, etc. Assume that the energy consumption is , the filtration efficiency is , and the system stability index is . By searching in the joint action space through the multi-objective Pareto optimization algorithm, a set of optimal parameter combinations is obtained, such that under these parameters, each objective can be better balanced.

[0044] During the actual operation process, since there may be conflicts among the control commands of different devices, an asynchronous policy update mechanism is needed to balance local control command conflicts. The asynchronous policy update mechanism allows each device to update its control policy at different time points according to its own situation, rather than synchronously. For example, when the exhaust fan detects a sudden increase in flue gas flow, it can immediately adjust the rotational speed without waiting for the filtration device and the cooling unit to adjust simultaneously. During the asynchronous update process, through a certain coordination mechanism, such as message passing and conflict detection algorithms, it is ensured that the control policy updates of each device do not conflict with each other, thus achieving the stable operation of the system. Assume that device updates its control policy at time based on its own local information . Through the coordination mechanism , where is the total number of devices, it is ensured that the control policy updates of each device can be coordinated and consistent, avoiding conflicts.

[0045] Example 5: This embodiment mainly introduces the specific implementation of the exception handling module, the related self-healing mechanism, and the flue gas toxicity index prediction model.

[0046] The exception handling module identifies abnormal fluctuations during the flue gas exhaust process based on the density peak detection algorithm. The density peak detection algorithm identifies data points with high local density and far distance from other high-density points by calculating the local density of data points and the distance to other high-density points. These points usually represent outliers. Assume the data point is , and its local density The calculation formula is: where is the distance between data points and , is the cut-off distance, is the total number of data points. The distance between data point and other high-density points, where is the local density of data point j. By analyzing and , the abnormal fluctuation points are determined.

[0047] When abnormal fluctuations are detected, the density clustering algorithm is used to divide the risk levels. The density clustering algorithm divides data points into different clusters according to the density distribution of data points, and each cluster represents a risk level. Generate a hierarchical warning signal according to the risk level and associate the trigger conditions of the self-healing mechanism. For example, when the risk level reaches a relatively high level, the self-healing mechanism is triggered.

[0048] The self-healing mechanism uses a bidirectional graph neural network to model the coupling relationship between the filtering device and the exhaust fan. The bidirectional graph neural network can learn the relationship between nodes from both the forward and reverse paths. In this system, the filtering device and the exhaust fan are used as the nodes of the graph, and the connection between them represents their coupling relationship. Extract the fault conduction characteristics from the forward path and the redundant resource scheduling characteristics from the reverse path. Assume the graph neural network model is , where is the adjacency matrix of the graph, describing the connection relationship between nodes, is the feature matrix of the nodes. Obtain the fault conduction characteristics and the redundant resource scheduling characteristics through the bidirectional graph neural network, generate a repair plan through attention weight fusion, and update the risk assessment parameters of the dynamic filtering network.

[0049] In terms of predicting the flue gas toxicity index, a prediction model for the flue gas toxicity index is constructed. This model is based on historical process data and real-time gas component parameters, and uses a temporal convolutional network to fuse the generation law of thermal degradation products. Assume that the historical process data is , and the real-time gas component parameters are . The generation law of thermal degradation products is represented by the function . The temporal convolutional network model is . The flue gas toxicity index is predicted through this model. To more comprehensively evaluate the toxicity risk, a generative adversarial network is used to simulate the gas mixing distribution under extreme conditions, generate a toxicity risk probability map, and embed a redundant filtering layer deployment strategy. The generative adversarial network consists of a generator and a discriminator . The generator generates a simulated gas mixing distribution based on random noise. The discriminator judges whether the generated distribution is real. Through the adversarial training of the two, a gas mixing distribution closer to the real extreme conditions is generated, so as to obtain a more accurate toxicity risk probability map and provide a more reliable basis for the redundant filtering layer deployment strategy.

[0050] Through the collaborative work of the exception handling module, self-healing mechanism, and the flue gas toxicity index prediction model, abnormal situations during the system operation can be detected and processed in a timely manner, ensuring the safe and stable operation of the system, and at the same time improving the ability to respond to flue gas toxicity.

[0051] It should be noted that in this article, relational terms such as first and second are only used 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 "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0052] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A flue gas exhaust system for an additive manufacturing forming chamber of a rocket engine thrust chamber, characterized in that, Including: Multi-source data acquisition module: used to obtain multi-dimensional parameters of the flue gas in the forming chamber in real time, including temperature gradient distribution, particulate matter concentration spectrum, gas component ratio, and air flow velocity field; Flue gas feature extraction module: perform non-linear feature decomposition on the multi-dimensional parameters to generate flue gas diffusion mode features, particulate matter aggregation trend features, and thermodynamic coupling response features; Dynamic filtering module: construct a multi-layer filtering network based on an improved adaptive filtering algorithm, input the features into a dynamic weight allocation unit, and generate a filtering medium switching strategy and an air flow path optimization matrix; Collaborative control module: based on the filtering medium switching strategy and the air flow path optimization matrix, coordinate the operating parameters of the exhaust fan, filtering device, and cooling unit through a distributed reinforcement learning model, and generate a global energy consumption balance control instruction; Abnormal handling module: identify abnormal fluctuations during the flue gas exhaust process based on the density peak detection algorithm, trigger the self-repair mechanism, and update the topology structure of the filtering network.

2. The flue gas exhaust system for the forming chamber according to claim 1, characterized in that, The multi-source data acquisition module includes: Adopt a distributed optical fiber sensing array to capture the temperature gradient distribution, and reconstruct the three-dimensional thermal field evolution map through a spatio-temporal interpolation algorithm; Use polarization scattering measurement technology to separate the distribution density of particles with different particle sizes in the particulate matter concentration spectrum, and correct the measurement error in combination with the Kalman filtering algorithm; Use the spectral absorption feature fusion method to extract the proportion of key components in the gas component ratio, and predict the gas diffusion boundary based on the chaotic mapping model.

3. The flue gas exhaust system for the forming chamber according to claim 2, characterized in that, The flue gas feature extraction module further includes: Adopt a vortex decomposition algorithm to extract the turbulent core region of the air flow velocity field, and correlate the particulate matter movement trajectory through phase-locking technology; Use the tensor decomposition method to model the temperature-air flow interaction of the thermodynamic coupling response feature, and generate a multi-scale energy transfer path feature matrix.

4. The flue gas exhaust system for the forming chamber according to claim 1, characterized in that, The dynamic filtering module further includes: Construct a filtering medium life prediction model based on a double-layer gated recurrent unit to calculate the attenuation rate of the porosity of the filtering layer in real time; Optimize the filtering medium switching strategy through the gradient backpropagation algorithm, and fuse the air flow path optimization matrix to generate a dynamic pressure drop compensation instruction.

5. The flue gas exhaust system for the forming chamber according to claim 4, characterized in that, The collaborative control module further includes: Define the exhaust fan speed, the differential pressure threshold of the filtering device, and the power threshold of the cooling unit as the joint action space, and use the multi-objective Pareto optimization algorithm to solve the global optimal parameter combination; Balance the conflict of local control instructions through an asynchronous policy update mechanism.

6. The flue gas exhaust system for the forming chamber according to claim 1, characterized in that, The abnormal handling module further includes: Use the density clustering algorithm to divide the risk level of the abnormal fluctuation, generate a hierarchical warning signal, and associate the trigger condition of the self-repair mechanism; When the particulate matter concentration exceeds the limit, activate the redundant filtering layer deployment strategy and reconstruct the priority weight of the air flow path optimization matrix.

7. The flue gas exhaust system for the forming chamber according to claim 6, characterized in that, The self-repair mechanism further includes: Use a bi-directional graph neural network to model the coupling relationship between the filtering device and the exhaust fan, extract the fault conduction feature from the forward path, and extract the redundant resource scheduling feature from the reverse path; Generate a repair plan through attention weight fusion, and update the risk assessment parameters of the dynamic filtering network.

8. The flue gas exhaust system for the forming chamber according to claim 6, characterized in that, The system further includes: Build a flue gas toxicity index prediction model. Based on historical process data and real-time gas component parameters, use a temporal convolutional network to fuse the generation law of thermal degradation products and dynamically adjust the chemical adsorption priority of the filter medium.

9. The flue gas exhaust system for the forming chamber according to claim 8, characterized in that, The flue gas toxicity index prediction model further includes: Use a generative adversarial network to simulate the gas mixing distribution under extreme conditions, generate a toxicity risk probability map, and embed a redundant filter layer deployment strategy.

10. The fume exhaust system of the forming silo according to claim 3, wherein, The treatment of the turbulent core region further includes: Use an adaptive vortex tracking algorithm to balance the measurement accuracy of high-speed airflows and low-speed stagnant zones, capture the non-linear characteristics of vortex evolution through a residual convolutional network, construct an airflow stability evaluation matrix, and fuse it into the weight allocation unit of the dynamic filtration network.

Images