Pulse modulation plasma jet generation method and application thereof

Abstract

The invention discloses a pulse modulation plasma jet generation method and application thereof, and particularly relates to the field of microelectronics, and the method comprises the following steps: I, constructing a multidisciplinary association database, modulating a high-voltage pulse power supply, and optimizing pulse waveform parameters at the same time; iI, pulse voltage is applied through a high-voltage electrode to generate plasma jet, a coupling simulation model is established, and the evolution process of the plasma jet is simulated; according to the method, the requirement of manual operation can be reduced, subjective deviation and information loss are avoided, the robustness of resisting environmental interference is improved, and the method can adapt to the high-speed and high-precision process requirement; the testing of invalid parameter combinations can be avoided, the searching capability is improved, the experiment or simulation frequency is reduced, the anti-interference capability of the jet system on working condition fluctuation is improved, and the maintenance frequency is reduced.

Description

Technical Field

[0001] The present invention relates to the field of microelectronics, and particularly to a method for generating a pulsed modulated plasma jet and its application. Background Art

[0002] Since the plasma jet technology emerged in the 1990s, it has gradually moved from laboratory research to industrial applications. In the field of microelectronics, it is used for nanoscale etching and thin film deposition; in biomedicine, low-temperature plasma jets can selectively induce apoptosis of cancer cells; plasma catalysis has become one of the key technologies for waste gas treatment. However, with the complication of application scenarios, the jet characteristics of traditional jet generation methods are affected by strong multi-physical field coupling, the parameter optimization driven by experience has low efficiency and poor generalization, and the continuous discharge mode leads to significant energy waste. Although the pulsed modulation technology can improve energy efficiency, the waveform design and dynamic regulation lack theoretical guidance.

[0003] The existing methods for generating pulsed modulated plasma jets and their applications have a high demand for manual operation, are prone to subjective deviation and information loss, and have low robustness against environmental interference, and cannot meet the requirements of high-speed and high-precision processes; in addition, the existing methods for generating pulsed modulated plasma jets and their applications have tests of invalid parameter combinations, reducing the search ability, requiring more experiments or simulations, reducing the anti-interference ability of the jet system to working condition fluctuations, and having a high maintenance frequency; for this reason, we propose a method for generating a pulsed modulated plasma jet and its application. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects existing in the prior art, and to propose a method for generating a pulsed modulated plasma jet and its application.

[0005] In the first aspect of the implementation of the present invention, first, a method for generating a pulsed modulated plasma jet and its application is proposed, and the method includes:

[0006] Ⅰ. Construct a multidisciplinary association database, modulate a high-voltage pulsed power supply, and optimize the pulsed waveform parameters at the same time;

[0007] Ⅱ. Apply a pulsed voltage through a high-voltage electrode to generate a plasma jet, establish a coupled simulation model, and simulate the evolution process of the plasma jet;

[0008] Ⅲ. Collect plasma jet data, and predict the jet morphology in real time. According to the jet performance index, perform multi-parameter collaborative optimization on the plasma jet;

[0009] Ⅳ. Collect jet images and emission spectra through a high-speed camera and a spectrometer, and then dynamically adjust the pulse frequency and gas flow according to the real-time data and the parameter optimization results.

[0010] As a further solution of the present invention, the construction of the multidisciplinary association database described in step I:

[0011] S1.1: Collect data groups such as literature, experimental data, and simulation results in the fields of plasma physics, gas dynamics, and electromagnetics. Screen out duplicate data collected from different data sources. At the same time, use the box plot method to identify and delete abnormal data in each group of data, and fill in the missing values in each group of data by interpolation method;

[0012] S1.2: Convert the processed data groups into text format, perform term recognition through NLP technology, extract entities and relationships in each group of text data, then set the core concepts and their hierarchical relationships of plasma physics, gas dynamics, and electromagnetics, and use the identified entities in each group as nodes and relationships as edges. Connect each pair of nodes through the corresponding edges according to the recognition results;

[0013] S1.3: Use rule reasoning to mine the cross relationships between entities in plasma physics, gas dynamics, and electromagnetics, and establish mapping rules between corresponding interdisciplinary variables to construct a multidisciplinary association graph. Then store the multidisciplinary association graph through the Neo4j library and design corresponding data indexes to establish a multidisciplinary association database.

[0014] As a further solution of the present invention, the specific steps of optimizing the pulse waveform parameters described in step I are as follows:

[0015] S2.1: Through the relationship reasoning of the multidisciplinary association graph in the multidisciplinary association database, obtain the pulse waveform parameter combinations and their parameter ranges of the plasma breakdown voltage of the jet generator to set the parameter search space. The parameter combinations include rise time, pulse width, and repetition frequency, and set the objective function based on each gas type, gas pressure, and pulse waveform;

[0016] S2.2: Set the population size and the number of iterations, randomly initialize the positions of each individual in the search space, initialize the pheromone concentration of each parameter combination in the search space to the same value, and calculate the inverse of the breakdown voltage under the current parameter combination of each individual at the same time, and use it as its heuristic factor;

[0017] S2.3: Calculate the selection probability of each pulse waveform parameter according to the pheromone concentration and the heuristic factor, and make a random selection based on the selection probability to gradually construct a complete parameter path in the search space to construct a complete parameter combination. Calculate the breakdown voltage value corresponding to the current path through the objective function, and update the pheromone concentration of the corresponding path according to the breakdown voltage values corresponding to each path;

[0018] S2.4: Repeat path construction and pheromone concentration update until the change value of the pheromone concentration of each path converges to a preset threshold value or reaches the preset number of iterations. Then, traverse the constructed groups of paths and select the parameter combination corresponding to the path with the highest pheromone concentration as the optimal pulse parameter combination.

[0019] As a further solution of the present invention, the specific steps of collecting plasma jet data and predicting the jet morphology described in step III are as follows:

[0020] S3.1: Collect and preprocess each group of data of the pulse waveforms and scalar parameters of historical plasma jets, then normalize each group of processed data to the interval [-1, 1]. Divide each group of processed data into a training set, a validation set, and a test set. Then, design and construct a prediction model with a hybrid input structure. Then, collect the actual images of historical plasma jets, scale the actual images to a fixed size, and convert them into grayscale images;

[0021] S3.2: Input the training set into the prediction model. The prediction model performs forward propagation on the training set. The pulse waveform data in the training set passes through multiple layers of 1D convolution and pooling in the one-dimensional convolution branch of the prediction model to output a global average pooling feature vector. The scalar parameter data is mapped to a high-dimensional space through the fully connected layer in the prediction model, and the mapped data is concatenated with the feature vector to generate a fused feature. The fused feature is upsampled to the target image size through multiple layers of transposed convolution, and then a predicted image is generated through the output layer;

[0022] S3.3: Combine the MSE function and the SSIM function to calculate the loss value between the predicted image and the actual image. Then, perform backpropagation of the calculated loss value starting from the output layer of the prediction model, calculate the gradient of the loss value with respect to the model parameters, and then use the Adam optimizer to update the parameters;

[0023] S3.4: After each round of training, input the validation set into the prediction model and calculate its corresponding loss value. If the validation loss does not decrease continuously for multiple times, stop training. Otherwise, retrain and validate the prediction model again. After training, evaluate the peak signal-to-noise ratio and structural similarity of the prediction model through the test set, and deploy the trained prediction model to the monitoring platform;

[0024] S3.5: Input the collected real-time plasma jet data into the trained prediction model, obtain the jet morphology prediction image through the forward propagation of the prediction model, and simultaneously collect the actual jet morphology image in real time and locate the prediction error area for subsequent training adjustment.

[0025] As a further solution of the present invention, the specific steps of multi-parameter collaborative optimization of the plasma jet described in step III are as follows:

[0026] S4.1: Determine the adjustable parameters of the jet performance according to the multidisciplinary association database, including pulse voltage, air pressure, gas flow rate, and pulse frequency, and set the value ranges of each parameter to construct the corresponding parameter space. Then, perform a weighted combination of the multi-parameter indicators, and generate a corresponding single-objective function after normalization of the combination result;

[0027] S4.2: Generate multiple sample points within the parameter space through Latin hypercube sampling, and run experiments or simulations on each sample point through the coupled simulation model. At the same time, calculate the fitness value of each sample point using the single-objective function, and match the calculated fitness value with the corresponding sample point to construct the initial data set;

[0028] S4.3: Construct a Gaussian process surrogate model, and the Gaussian process surrogate model obtains the mean, standard deviation, and length scale corresponding to the initial data set by maximizing the marginal likelihood function. Then, select the improvement probability as the acquisition function, and calculate the improvement probability of each data in the initial data set through the mean and standard deviation predicted by the Gaussian process;

[0029] S4.4: Select the sample point pair with the highest improvement probability as the next evaluation point, use the coupled simulation model to simulate the parameter combination corresponding to the sample point with the highest improvement probability, and calculate its corresponding fitness value. Then, add the new data to the data set, refit the Gaussian process, and repeat multiple iterations of updating until the improvement amplitude of the fitness values of multiple rounds of new sample points is less than the preset threshold, then stop the iteration, and output the parameter combination with the highest fitness value.

[0030] In the second aspect of the implementation of the present invention, a method for applying a pulsed modulated plasma jet is proposed, including:

[0031] (1) Input the requirements, automatically match the historical data through the multidisciplinary association database, select argon as the working gas, set the pulse voltage range to 8 - 12 kV, and set the electrode spacing range to 2 - 5 mm, and generate the initial parameter combination;

[0032] (2) Set the target machining accuracy and the energy efficiency of the plasma jet, and predict the improvement probability of different parameter combinations through the Gaussian process model, and preferentially test the parameters with high improvement probability;

[0033] (3) Start the high-voltage pulse power supply according to the optimized parameters to generate a stepped pulse waveform with a pulse width of 200 ns. Then, argon is ionized by the high-voltage pulse to form a stable jet, and is accelerated to 150 m / s through a Venturi nozzle;

[0034] (4) The jet etches the material to be processed through physical sputtering and chemical reactions, while monitoring the jet temperature in real time to avoid thermal damage to the material. A high-speed camera with a frame rate of 10 kHz is used to collect images of the jet morphology, and a spectrometer monitors the concentration of active particles;

[0035] (5) According to the current pulse parameters and jet images, the predicted etching depth error is output. If the predicted error is greater than 5%, the pulse frequency is adjusted within the range of ±10 Hz, and the gas flow rate is adjusted within the range of ±1 m / s;

[0036] (6) The actual processing parameters are input into the COMSOL multi-physics model to simulate the jet-material interaction. If the processing result does not meet the standard, the knowledge graph database is updated, and the parameter optimization process is restarted.

[0037] Advantages of the present invention:

[0038] The present invention proposes a method for generating a pulsed modulated plasma jet and its application. By designing and constructing a prediction model with a hybrid input structure, collecting actual images of historical plasma jets, scaling the actual images to a fixed size, and converting them into grayscale images, the training set is input into the prediction model. The prediction model performs forward propagation on the training set. The pulse waveform data in the training set passes through multiple 1D convolutions and pooling in the one-dimensional convolution branch of the prediction model to output a global average pooling feature vector. The scalar parameter data is mapped to a high-dimensional space through the fully connected layer in the prediction model, and the mapped data is concatenated with the feature vector to generate a fused feature. The fused feature is upsampled to the target image size through multiple transposed convolutions, and then a prediction image is generated through the output layer. The loss value between the prediction image and the actual image is calculated by combining the MSE function and the SSIM function. Then, the calculated loss value is backpropagated starting from the output layer of the prediction model, and the gradient of the loss value with respect to the model parameters is calculated. Then, the Adam optimizer is used to update the parameters. After each round of training, the validation set is input into the prediction model, and its corresponding loss value is calculated. If the validation loss does not decrease continuously for multiple times, the training is stopped; otherwise, the prediction model is retrained and validated again. After the training is completed, the peak signal-to-noise ratio and structural similarity of the prediction model are evaluated through the test set, and the trained prediction model is deployed to the monitoring platform. The collected real-time plasma jet data is input into the trained prediction model, and a jet morphology prediction image is obtained through the forward propagation of the prediction model. At the same time, the actual jet morphology image is collected in real time, and the prediction error region is located for subsequent training adjustment. It can reduce the need for manual operation, avoid subjective bias and information loss, improve the robustness against environmental interference, and can adapt to the requirements of high-speed and high-precision processes.

[0039] The present invention provides a method for generating a pulsed-modulated plasma jet and its application. Adjustable parameters for jet performance are determined according to a multidisciplinary correlation database, including pulsed voltage, gas pressure, gas flow rate, and pulse frequency, and the value ranges of each parameter are set to construct a corresponding parameter space. Then, multi-parameter indicators are weighted and combined, and the combined result is normalized to generate a corresponding single-objective function. Multiple sample points are generated within the parameter space through Latin hypercube sampling, and experiments or simulations are run on each sample point through a coupled simulation model. At the same time, the fitness value of each sample point is calculated using the single-objective function, and the calculated fitness value is matched with the corresponding sample point to construct an initial dataset. A Gaussian process surrogate model is constructed, and the Gaussian process surrogate model obtains the mean, standard deviation, and length scale corresponding to the initial dataset by maximizing the marginal likelihood function. Then, the improvement probability is selected as the acquisition function, and the improvement probability of each data in the initial dataset is calculated through the mean and standard deviation predicted by the Gaussian process. The sample point pair with the highest improvement probability is selected as the next evaluation point, and the parameter combination corresponding to the sample point with the highest improvement probability is simulated using the coupled simulation model, and its corresponding fitness value is calculated. Then, the new data is added to the dataset, and the Gaussian process is refitted. The iteration is repeated multiple times until the improvement amplitude of the fitness value of multiple rounds of new sample points is less than a preset threshold, at which point the iteration stops, and the parameter combination with the highest fitness value is output. This can avoid testing invalid parameter combinations, improve the search ability, reduce the number of experiments or simulations, enhance the anti-interference ability of the jet system to operating condition fluctuations, and reduce the maintenance frequency. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The present invention will be further described below with reference to the accompanying drawings.

[0041] Figure 1 It is a flowchart of a method for generating a pulsed-modulated plasma jet provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] 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.

[0043] Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] An embodiment of the present invention provides a method for generating a pulsed-modulated plasma jet and its application. Refer to Figure 1 , Figure 1The figure is a flowchart of a method for generating a pulsed modulated plasma jet provided by an embodiment of the present invention. The method includes the following steps:

[0045] Construct a multidisciplinary association database, modulate a high-voltage pulsed power supply, and optimize the pulse waveform parameters.

[0046] Specifically, collect various data such as literature, experimental data, and simulation results in the fields of plasma physics, gas dynamics, and electromagnetics, screen out duplicate data collected from different data sources, and at the same time use the box plot method to identify and delete abnormal data in each group of data, fill in the missing values in each group of data by interpolation method, convert the processed data in each group into text format, perform term recognition through NLP technology, extract entities and relationships in each group of text data, then set the core concepts and their hierarchical relationships of plasma physics, gas dynamics, and electromagnetics, use the identified entities in each group as nodes and relationships as edges, connect each node pairwise through the corresponding edges according to the recognition results, use rule reasoning to mine the cross relationships between entities in plasma physics, gas dynamics, and electromagnetics, and establish the mapping rules between corresponding interdisciplinary variables to construct a multidisciplinary association graph. After that, store the multidisciplinary association graph through the Neo4j library and design the corresponding data index to establish a multidisciplinary association database.

[0047] Specifically, through the relationship reasoning of the multidisciplinary association graph in the multidisciplinary association database, obtain the pulse waveform parameter combinations and their parameter ranges of the plasma breakdown voltage of the jet generator to set the parameter search space. The parameter combinations include rise time, pulse width, and repetition frequency, and set the objective function based on each gas type, gas pressure, and pulse waveform. Set the population size and the number of iterations, randomly initialize the positions of each individual in the population in the search space, initialize the pheromone concentration of each parameter combination in the search space to the same value, and at the same time calculate the inverse of the breakdown voltage under the current parameter combination of each individual and use it as its heuristic factor. Calculate the selection probability of each pulse waveform parameter according to the pheromone concentration and the heuristic factor, and make a random selection based on the selection probability to gradually construct a complete parameter path in the search space to construct a complete parameter combination. Calculate the breakdown voltage value corresponding to the current path through the objective function, and at the same time update the pheromone concentration of the corresponding path according to the breakdown voltage values corresponding to each path. Repeat the path construction and pheromone concentration update until the change value of the pheromone concentration of each path converges to a preset threshold or reaches the preset number of iterations. After that, traverse the constructed paths and select the parameter combination corresponding to the path with the highest pheromone concentration as the optimal pulse parameter combination.

[0048] Apply a pulsed voltage through a high-voltage electrode to generate a plasma jet, establish a coupled simulation model, and simulate the evolution process of the plasma jet.

[0049] Collect plasma jet data, predict the jet morphology in real time, and perform multi-parameter collaborative optimization on the plasma jet according to the jet performance indicators.

[0050] Specifically, collect and preprocess each group of data of the pulse waveforms and scalar parameters of the historical plasma jets, then normalize each group of processed data to the interval [-1, 1], divide each group of processed data into a training set, a validation set, and a test set. After that, design and construct a prediction model with a hybrid input structure. Then collect the actual images of the historical plasma jets, scale the actual images to a fixed size, and convert them into grayscale images. Input the training set into the prediction model. The prediction model performs forward propagation on the training set. The pulse waveform data in the training set passes through multiple layers of 1D convolution and pooling in the one-dimensional convolution branch of the prediction model to output a global average pooling feature vector. The scalar parameter data is mapped to a high-dimensional space through the fully connected layer in the prediction model, and the mapped data is concatenated with the feature vector to generate a fused feature. The fused feature is upsampled to the target image size through multiple layers of transposed convolution, and then a predicted image is generated through the output layer. Calculate the loss value between the predicted image and the actual image by combining the MSE function and the SSIM function. Then perform backpropagation starting from the output layer of the prediction model for the calculated loss value, and calculate the gradient of the loss value with respect to the model parameters. Then use the Adam optimizer to update the parameters. After each round of training, input the validation set into the prediction model and calculate its corresponding loss value. If the validation loss does not decrease continuously for multiple times, stop training; otherwise, retrain and validate the prediction model again. After training, evaluate the peak signal-to-noise ratio and structural similarity of the prediction model through the test set, and deploy the trained prediction model to the monitoring platform. Input the collected real-time plasma jet data into the trained prediction model, and obtain the jet morphology prediction image through the forward propagation of the prediction model. At the same time, collect the actual jet morphology images in real time and locate the prediction error regions for subsequent training adjustment.

[0051] Specifically, the adjustable parameters of the jet performance are determined according to the multidisciplinary association database, including pulse voltage, air pressure, gas flow rate, and pulse frequency, and the value ranges of each parameter are set to construct the corresponding parameter space. Then, the multi-parameter indicators are weighted and combined, and the combined result is normalized to generate a corresponding single-objective function. Multiple sample points are generated in the parameter space through Latin hypercube sampling, and experiments or simulations are run on each sample point through a coupled simulation model. At the same time, the fitness value of each sample point is calculated using the single-objective function, and the calculated fitness value is matched with the corresponding sample point to construct an initial data set. A Gaussian process surrogate model is constructed, and the Gaussian process surrogate model obtains the mean, standard deviation, and length scale corresponding to the initial data set by maximizing the marginal likelihood function. Then, the improvement probability is selected as the acquisition function, and the improvement probability of each data in the initial data set is calculated through the mean and standard deviation predicted by the Gaussian process. The sample point pair with the highest improvement probability is selected as the next evaluation point, and the parameter combination corresponding to the sample point with the highest improvement probability is simulated using the coupled simulation model, and its corresponding fitness value is calculated. Then, the new data is added to the data set, and the Gaussian process is refitted. The iteration is repeated and updated multiple times until the improvement amplitude of the fitness value of multiple rounds of new sample points is less than the preset threshold, and the iteration is stopped, and the parameter combination with the highest fitness value is output.

[0052] The jet image and emission spectrum are collected by a high-speed camera and a spectrometer, and the pulse frequency and gas flow rate are dynamically adjusted according to the real-time data and the parameter optimization results.

[0053] Based on the same inventive concept, the embodiments of the present invention also provide an application of a method for generating a pulsed modulated plasma jet. It includes:

[0054] (1) Input the requirements, automatically match the historical data through the multidisciplinary association database, select argon as the working gas, set the pulse voltage range to 8 - 12 kV, and set the electrode spacing range to 2 - 5 mm, and generate an initial parameter combination;

[0055] (2) Set the target processing accuracy and the energy efficiency of the plasma jet, and predict the improvement probability of different parameter combinations through the Gaussian process model, and preferentially test the parameters with high improvement probability;

[0056] (3) Start the high-voltage pulse power supply according to the optimized parameters to generate a stepped pulse waveform with a pulse width of 200 ns. Then, argon is ionized by the high-voltage pulse to form a stable jet, which is accelerated to 150 m / s through a Venturi nozzle;

[0057] (4) The jet etches the material to be processed through physical sputtering and chemical reactions. At the same time, the jet temperature is monitored in real time to avoid thermal damage to the material. A high-speed camera with a frame rate of 10 kHz is used to collect the jet morphology image, and a spectrometer monitors the active particle concentration;

[0058] (5) Output the predicted etching depth error based on the current pulse parameters and jet image. If the predicted error is greater than 5%, adjust the pulse frequency within the range of ±10 Hz and adjust the gas flow rate within the range of ±1 m / s.

[0059] (6) Input the actual processing parameters into the COMSOL multi-physics model to simulate the jet-material interaction. If the processing result does not meet the standard, update the knowledge graph database and restart the parameter optimization process.

[0060] The above has described in detail an embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention shall still fall within the scope covered by the patent of the present invention.

Claims

1. A method for generating a pulse modulated plasma jet, characterized in that: The following steps are involved: Ⅰ. Build a multidisciplinary correlation database, modulate the high-voltage pulse power supply, and optimize the pulse waveform parameters; II. Generate plasma jet by applying pulse voltage through high-voltage electrodes, establish a coupled simulation model, and simulate the evolution process of plasma jet; III. Collect plasma jet data and predict jet morphology in real time, and perform multi-parameter collaborative optimization of plasma jet according to jet performance indicators; IV. Collect jet images and emission spectra through high-speed cameras and spectrometers, and then dynamically adjust the pulse frequency and gas flow rate based on real-time data and parameter optimization results.

2. The method for generating a pulse modulated plasma jet according to claim 1, characterized in that: Step Ⅰ: Construct a multidisciplinary relational database: S1.1: Collect literature, experimental data and simulation results in the fields of plasma physics, gas dynamics and electromagnetism, filter out duplicate data collected from different data sources, use the boxplot method to identify and delete abnormal data in each data set, and fill in missing values ​​in each data set by interpolation; S1.2: Convert each group of processed data into text format, perform term recognition through NLP technology, and extract entities and relationships in each group of text data. Then set the core concepts of plasma physics, gas dynamics, and electromagnetism and their hierarchical relationships, and identify each group of entities as nodes and relationships as edges. According to the recognition results, connect each node with the corresponding edges; S1.3: Use rule reasoning to mine the cross-relationships between entities in plasma physics, gas dynamics, and electromagnetism, and establish corresponding mapping rules between interdisciplinary variables to construct a multidisciplinary association map. Then, store the multidisciplinary association map through the Neo4j library, and design corresponding data indexes to establish a multidisciplinary association database.

3. The method for generating a pulse modulated plasma jet according to claim 2, characterized in that: The specific steps of optimizing the pulse waveform parameters in step I are as follows: S2.1: Through the relational reasoning of the multidisciplinary association graph in the multidisciplinary association database, the pulse waveform parameter combination and parameter range of the plasma breakdown voltage of the jet generator are obtained to set the parameter search space, and the parameter combination includes rise time, pulse width and repetition frequency, and the objective function is set based on each gas type, gas pressure and pulse waveform; S2.2: Set the population size and the number of iterations, and randomly initialize the position of each individual in the search space, and initialize the pheromone concentration of each parameter combination in the search space to the same value. At the same time, calculate the inverse ratio of the breakdown voltage of each individual under the current parameter combination and use it as its heuristic factor; S2.3: Calculate the selection probability of each pulse waveform parameter according to the pheromone concentration and the heuristic factor, and make random selections based on the selection probability to gradually construct a complete parameter path in the search space to construct a complete parameter combination, calculate the breakdown voltage value corresponding to the current path through the objective function, and update the pheromone concentration of the corresponding path according to the breakdown voltage value corresponding to each path; S2.4: Repeat the path construction and pheromone concentration update until the pheromone concentration change value of each path converges to the preset threshold value, or reaches the preset number of iterations, then traverse each group of constructed paths, and select the parameter combination corresponding to the path with the highest pheromone concentration as the optimal pulse parameter combination.

4. The method for generating a pulse modulated plasma jet according to claim 3, characterized in that: The specific steps of collecting plasma jet data and predicting the jet morphology in real time as described in step III are as follows: S3.1: Collect and preprocess the pulse waveform and scalar parameter data of the historical plasma jet, normalize each group of processed data to the interval [-1,1], divide each group of processed data into training set, validation set and test set, then design and construct a prediction model with a mixed input structure, collect the actual image of the historical plasma jet, scale the actual image to a fixed size, and convert it into a grayscale image; S3.2: The training set is input into the prediction model, which performs forward propagation on the training set. The pulse waveform data in the training set is subjected to multi-layer 1D convolution and pooling by the one-dimensional convolution branch in the prediction model, and a global average pooling feature vector is output. The scalar parameter data is mapped to a high-dimensional space through a fully connected layer in the prediction model, and the mapped data is concatenated with the feature vector to generate a fused feature. The fused feature is upsampled to the target image size through multi-layer transposed convolution, and a predicted image is generated through the output layer. S3.3: Combine the MSE function and the SSIM function to calculate the loss value of the predicted image and the actual image, then backpropagate the calculated loss value from the output layer of the prediction model, calculate the gradient of the loss value to the model parameters, and then use the Adam optimizer to update the parameters; S3.4: After each round of training, the validation set is input into the prediction model and its corresponding loss value is calculated. If the validation loss does not decrease for several consecutive times, the training is stopped. Otherwise, the prediction model is retrained and verified. After the training, the peak signal-to-noise ratio and structural similarity of the prediction model are evaluated through the test set, and the trained prediction model is deployed to the monitoring platform; S3.5: The collected real-time plasma jet data is input into the trained prediction model, and the jet morphology prediction image is obtained through the forward propagation of the prediction model. At the same time, the actual jet morphology image is collected in real time, and the prediction error area is located for subsequent training adjustments.

5. The method for generating a pulse modulated plasma jet according to claim 2, characterized in that: The specific steps of multi-parameter coordinated optimization of the plasma jet described in step III are as follows: S4.1: Determine the adjustable parameters of the jet performance according to the multidisciplinary correlation database, including pulse voltage, gas pressure, gas flow rate and pulse frequency, and set the value range of each parameter to construct the corresponding parameter space, then perform weighted combination of the multi-parameter indicators, and generate the corresponding single objective function after normalizing the combination result; S4.2: Generate multiple sample points in the parameter space through Latin hypercube sampling, and run experiments or simulations on each sample point through a coupled simulation model. At the same time, use a single objective function to calculate the fitness value of each sample point, match the calculated fitness value with the corresponding sample point, and construct an initial data set; S4.3: Construct a Gaussian process proxy model, and use the Gaussian process proxy model to obtain the corresponding mean, standard deviation, and length scale of the initial data set by maximizing the marginal likelihood function. Then, select the improved probability as the acquisition function, and calculate the improved probability of each data in the initial data set by using the mean and standard deviation predicted by the Gaussian process. S4.4: Select the sample point pair with the highest probability of improvement as the next evaluation point, use the coupled simulation model to simulate the corresponding parameter combination of the sample point with the highest probability of improvement, and calculate its corresponding fitness value, then add the new data to the data set, refit the Gaussian process, repeat the iterative update multiple times until the fitness value improvement of multiple rounds of new sample points is less than the preset threshold, stop the iteration, and output the parameter combination with the highest fitness value.

6. An application of a method for generating a pulse modulated plasma jet, used to implement a method for generating a pulse modulated plasma jet as claimed in any one of claims 1 to 5, characterized in that: include: (1) Input the requirements, automatically match historical data through the multidisciplinary association database, select argon as the working gas, set the pulse voltage range to 8-12 kV, set the electrode spacing range to 2-5 mm, and generate the initial parameter combination; (2) Setting the target machining accuracy and plasma jet energy efficiency, and predicting the improvement probability of different parameter combinations through the Gaussian process model, and giving priority to testing parameters with high improvement probability; (3) starting the high-voltage pulse power supply according to the optimized parameters to generate a step-type pulse waveform with a pulse width of 200 ns. After that, the argon gas is ionized by the high-voltage pulse to form a stable jet and is accelerated to 150 m / s through the Venturi nozzle; (4) The jet etches the processed material through physical sputtering and chemical reaction, and monitors the jet temperature in real time to avoid thermal damage to the material. A high-speed camera with a frame rate of 10 kHz is used to collect jet morphology images, and a spectrometer monitors the concentration of active particles. (5) Output the predicted etching depth error according to the current pulse parameters and the jet image. If the predicted error is greater than 5%, adjust the pulse frequency within the range of ±10 Hz and the gas flow rate within the range of ±1 m / s; (6) The actual processing parameters are input into the COMSOL multiphysics model to simulate the jet-material interaction. If the processing results do not meet the requirements, the knowledge graph database is updated and the parameter optimization process is restarted.

Images