Intelligent control method for melting forming process of laser powder bed

Abstract

The invention discloses an intelligent control method for a laser powder bed melting forming process, relates to the field of a deep learning technology, a process control technology and a laser powder bed melting forming technology, and provides an intelligent control method similar to cascade control. An inner ring control system of the control system adopts a reinforcement learning intelligent agent as a controller for the temperature of a molten pool, and the intelligent agent learns and generates a proper control strategy in an environment based on a digital twinborn structure, so that the control system can better adapt to stronger nonlinearity of laser powder bed melting forming. And the outer ring is constructed by a process parameter optimization agent based on imitation learning, and the agent can simulate an expert mode of process parameter optimization, so that the forming process is enabled to develop towards an expected direction. In addition, the invention further designs a set of process anomaly detection and disposal system, and safe operation of the system is guaranteed.

Description

Technical Field

[0001] The present invention relates to the fields of deep learning technology, process control technology and laser powder bed fusion forming technology, and in particular to an intelligent control method for a laser powder bed fusion forming process. Background Art

[0002] Laser Powder Bed Fusion (LPBF) uses a layer-by-layer stacking method to build three-dimensional objects, converting complex 3D forming into 2D stacking forming, and greatly reducing the complexity of production by reducing dimensions. However, due to the temperature variation characteristics of material properties and the uncertainty of certain factors in the LPBF working environment, the LPBF forming results are unstable and have low repeatability. Therefore, it is necessary to implement closed-loop control of the LPBF process to reduce the impact of these factors. However, the LPBF process control method currently used basically controls one or two LPBF process states, so it is difficult to cope with more complex situations in the LPBF process. This is because the response speed of the LPBF process state to each component in the process parameters varies greatly, and these quantities have a strong coupling effect on the LPBF process state. Therefore, the difficulty of controller design is greatly increased.

[0003] However, the research progress in deep learning and reinforcement learning methods in recent years has provided new ideas for studying new methods for LPBF process control. Deep learning is a machine learning method that can automatically learn features and rules from a large amount of data to achieve various complex tasks. Reinforcement learning is another branch of machine learning, which allows intelligent agents to explore the best strategy to solve problems through trial and error. To this end, based on deep learning and reinforcement learning methods, we propose an intelligent control method for laser powder bed fusion forming process. Summary of the invention

[0004] The main purpose of the present invention is to provide an intelligent control method for the laser powder bed fusion forming process. The method proposed in the present invention innovatively combines process parameter optimization with process control, which can effectively improve the shape control and performance control capabilities of LPBF.

[0005] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for intelligent control of laser powder bed melting forming process, which adopts a reinforcement learning agent to realize intelligent control of LPBF forming process, characterized in that: the LPBF forming process control system designed by this method is composed of a molten pool temperature control loop and a process parameter optimization loop. The molten pool temperature control loop is the inner loop of the control system, and its main function is to maintain the LPBF forming process stably in the expected state; while the process parameter optimization loop is the outer loop of the control system, and its main function is to optimize the process parameters in time according to the current forming results so that the forming process develops to the required results. Two reinforcement learning agents are used as controllers in each of these two loops. In order to enable these two agents to learn the optimized action generation strategy, the present invention constructs a learning environment for the agent based on the digital twin of the LPBF forming process.

[0006] Preferably, the present invention uses an LPBF forming process detection system consisting of a coaxial molten pool detection device and a 3D scanner to provide the control system with necessary forming process data; The coaxial molten pool detection device is characterized in that a high-speed camera and a colorimetric pyrometer are used as basic detection equipment, and the temperature and morphology of the molten pool can be measured unobstructed and uninterruptedly through the coaxial observation optical path. The device is also characterized in that an FPGA computing acceleration card is used for molten pool image processing and control quantity calculation, thereby ensuring the real-time performance of the image processing algorithm and the control algorithm; Preferably, the so-called coaxial molten pool detection refers to constructing an optical path device with the same optical axis as the high-energy laser, so that the pyrometer and the high-speed near-infrared camera can always maintain a top-view observation of the molten pool state in the laser action area, which can not only ensure that the viewing angle always follows the molten pool, but also ensure that it is not blocked during observation. The molten pool image uses a threshold-based method to extract the molten pool area, and then calculates its size, image moment, etc. In addition, due to the high sampling rate of the molten pool detection system, the present invention uses FPGA to process the molten pool image and calculate the molten pool temperature control amount. The molten pool processing algorithm in the FPGA can extract the molten pool area in real time and accurately through image denoising, threshold processing and morphological operations. And the model structure of the molten pool temperature control agent is integrated into the FPGA after appropriate optimization, so that the molten pool control method based on reinforcement learning can adjust the laser power according to the molten pool state in real time, ensuring the real-time performance of the control system.

[0007] The 3D scanner is a morphology detection device that uses a straight line laser as structured light and a monochrome industrial camera as a detection element. According to claim 1, the 3D detection device of the present invention is characterized in that the point cloud data is matched with the molten pool data by using a method of converting the point cloud into the forming height, so as to facilitate the establishment of a forming process state data set. The point cloud conversion method refers to a method of converting the original point cloud data into the forming height through point cloud registration, grid model construction, point cloud data resampling and rearrangement.

[0008] Preferably, the LPBF forming process digital twin of the present invention is characterized in that the twin is composed of a forming process state prediction model and a forming result prediction model, and the output end of the state model is connected to the input end of the result model. The forming process state prediction model is a time series prediction model based on the PatchTST model, which is responsible for predicting the changing trend of the process state over a period of time based on the given process parameters and the current forming state. The forming result prediction model is a prediction model built on the GPT2 model, which can combine the input process parameters and the forming state to predict the forming result in the next stage.

[0009] Preferably, the present invention adopts a system structure similar to cascade control, wherein the process parameter optimization method is a method of constructing the control loop outside the system, which is used to control the overall development trend of the forming process. The process parameter optimization in the process refers to optimizing the process parameters according to the current forming results in the LPBF forming process in order to achieve the expected forming morphology, thereby ensuring that the forming process develops in the expected direction. The process parameter optimization method in the present invention is characterized in that the method adopts an imitation learning method to enable the parameter optimization agent to learn and imitate the expert's process parameter design mode, so that it can give an optimized process parameter adjustment plan according to the current forming results.

[0010] Preferably, the so-called outer control loop adopts the LPBF staged process parameter optimization method based on imitation learning; the so-called inner control loop adopts the LPBF process control method based on reinforcement learning. The two control loops use different control frequencies. The outer layer optimizes the process parameters for at least one forming layer; and the control frequency of the inner layer is the same as the sampling frequency of the process detection system, which is generally 1~100kHz.

[0011] Preferably, the present invention is also characterized by designing an environment for training reinforcement learning agents, which is based on the LPBF process digital twin and reward function, and it also clarifies the action space and observation space of the agent. During the agent training, the environment acts as a virtual control object of the agent, and evaluates different control output strategies given by the agent according to the reward function, guiding it to learn the ideal control output strategy.

[0012] Preferably, the molten pool temperature control method constructed by the present invention is the inner control loop of the cascade control system, which is used to maintain the stability of the forming process. The molten pool temperature control method is characterized in that the method adopts a reinforcement learning agent as the controller of the molten pool temperature, which can greatly improve the adaptability and robustness of the control system. The molten pool control method is also characterized in that the agent controls the molten pool temperature through laser power, and the control quantity generation strategy is learned by the agent from the environment constructed by the digital twin described in claim 3. The agent training strategy adopted in the learning is a method based on proximal strategy optimization.

[0013] Preferably, the present invention constructs an LPBF phased process parameter optimizer based on imitation learning. The so-called LPBF phased process parameter optimization refers to a method of adjusting the processing path and process parameters of the next layer or several layers by optimizing the forming target of the next stage according to the previous forming results. This method makes the forming morphology and mechanical properties gradually approach the design indicators by timely optimizing the process parameters. The so-called imitation learning is to introduce a discriminator containing an expert mode as a reward function for agent training on the basis of reinforcement learning, so as to narrow the search range of the agent strategy and improve the training efficiency. The so-called discriminator refers to a deep learning model composed of a deep convolutional network that can learn the expert mode when setting LPBF process parameters from an expert data set. The trained discriminator can determine whether the process parameters generated by the agent conform to the expert mode. The process parameter optimizer also includes a generator for generating process parameters, which learns the generation mode of process parameters according to the feedback of the discriminator, and uses the strategy difference as the loss function during the training process. Then the generated training process is to find the appropriate model weight to minimize the loss value.

[0014] Preferably, a deep learning-based forming process anomaly detection system is developed to detect abnormal states of the forming process, and to stop the operation in time when an abnormality occurs to prevent equipment damage. The anomaly detection system uses the state interval of the forming process given by the digital twin as a reference, and determines whether there is an abnormality by calculating the cross entropy between the reference value and the measured value. Then the anomaly classification model comprehensively determines the type of anomaly based on the molten pool temperature and molten pool image moment measured by the detection system. After receiving the signal of the abnormal type, the control system executes the corresponding disposal plan. The disposal plan includes but is not limited to alarming through sound and light signals, generating alarm logs, shutting down the laser, stopping the galvanometer, shutting down the scraper, etc.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In the present invention, an intelligent control method similar to cascade control is proposed, which is composed of two control loops, an inner loop and an outer loop. The inner loop control system of this control system adopts a reinforcement learning agent as the controller of the molten pool temperature. The agent learns to generate appropriate control strategies in an environment based on digital twin construction, so that the control system can better adapt to the strong nonlinearity of laser powder bed fusion forming. The outer loop is constructed by a process parameter optimization agent based on imitation learning, which can imitate the expert mode of process parameter optimization, thereby ensuring that the forming process develops in the desired direction. In addition, the present invention also designs a process anomaly detection and disposal system to ensure the safe operation of the system.

[0016] 2. In the present invention, a digital twin that characterizes the mapping relationship between LPBF process parameters and process states can be established based on LPBF process data as a basis for planning control quantities. On the other hand, an intelligent agent based on reinforcement learning can also train control strategies for controlling the LPBF process in a virtual environment established by the model. The method proposed in the present invention innovatively combines process parameter optimization with process control, which can effectively improve the shape and controllability of LPBF. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of the LPBF intelligent control system in the present invention; Figure 2 It is a schematic diagram of the overall structure of the coaxial detection device and the topography scanning device in the present invention; Figure 3 Schematic diagram of the prediction results of the digital twin of the LPBF process in the present invention; Figure 4 Schematic diagram of the operation of the process parameter optimizer and the LPBF process controller based on reinforcement learning in the present invention; Figure 5 This is a schematic diagram of the operation of a forming process anomaly detection system based on deep learning in the present invention. DETAILED DESCRIPTION

[0018] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0019] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0020] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0021] Please refer to Figure 1-5 As shown, the present invention is an intelligent control method for laser powder bed melting forming process, and its specific implementation steps are: 1. Establish the coaxial observation system and topography scanning device of LPBF process (1) Design the required transmittance and reflectivity of the dichroic mirror in the coaxial optical path and the passband window of the narrow-band filter according to the laser wavelength, the band of the molten pool thermal radiation, the detection band of the camera, and the detection band of the pyrometer.

[0022] (2) Design and manufacture the coaxial optical path structure, install the dichroic mirror, reflector, filter, pyrometer, and camera. During the installation process, fine-tune the orientation of the pyrometer and camera so that the center of their field of view coincides with the guide laser point.

[0023] (3) Set the camera’s window parameters, sampling rate, and exposure parameters, and then use the threshold optimization method to obtain the optimal threshold for segmenting the molten pool area.

[0024] (4) Install the topography scanner on the powder scraper, adjust the projection direction of the line laser, and adjust the direction and orientation of the camera field of view. Set the camera capture parameters, such as window opening, exposure time, capture frame rate, etc. Calibrate the camera's intrinsic and extrinsic parameter matrices and light plane parameters.

[0025] (5) End 2. Constructing and training the LPBF process digital twin can be completed by referring to the following steps: (1) Collect LPBF process data and process experiment data, and construct a data set.

[0026] (2) Let the input vector of the process model be the process parameters, and design the input layer of the process state model according to the dimension of the process parameters. Let the output of the process model be a vector containing process states such as the melt pool temperature and melt pool size, and design the output layer of the state model according to the dimension of the output vector.

[0027] (3) The input layer of the result model is designed according to the process parameters and the process state output by the process model, and the output layer of the result model is designed according to the required prediction of the forming result type (such as forming layer height, density, surface roughness, etc.). The hidden layers of the process model and the result model are designed to ensure a high representation capability of the model.

[0028] (4) Set appropriate training parameters such as learning rate and number of training cycles, use the Adam optimization algorithm to train the model, and save the model with the smallest loss value.

[0029] (5) End The types of process parameters used as input parameters in the model can be selected according to the size of the data set. If the existing data is small, only the types that have a significant impact on the process state and forming results of LPBF can be selected to train a model with a relatively single applicable scenario. Then, as the data accumulates, the types of process parameters used as input can be gradually increased, and the new model can be trained to make it applicable to more scenarios. In this way, the generalization of the model can be gradually improved.

[0030] 3. Establish an LPBF process anomaly detector (1) Design an LPBF experiment to make a predetermined form of anomaly occur in the experiment without damaging the equipment, and construct an anomaly data set using process data recorded by a pyrometer and camera.

[0031] (2) The same process parameters are input into the digital twin and its output is recorded. The cross entropy between the twin output and the measured process data is calculated. The measured data, cross entropy, and corresponding anomaly types are then combined to form a process anomaly data set.

[0032] (3) Fine-tune the pre-trained process model using the anomaly dataset, taking the anomaly type as the fitting target during training. Set the learning rate and number of training cycles of the Adam optimization method to make the training process converge.

[0033] (4) Save the model with the smallest loss value.

[0034] (5) End 4. Establish an LPBF phased process parameter optimizer based on imitation learning Assume that the workpiece to be formed has been divided into layers in the computer-aided manufacturing (CAM) software, and every N layers is a forming stage. The topography is scanned once each forming stage is completed.

[0035] (1) Collect process data including process experiment results and construct an LPBF process expert dataset.

[0036] (2) Construct a discriminator based on a deep convolutional network to learn the expert mode in the expert dataset. The discriminator consists of an input layer, three linear layers, two ReLU layers, and a Sigmoid layer. The ReLU layer is sandwiched between the three Linear layers, and the Sigmoid layer is used as the output layer.

[0037] (3) The Adam algorithm is used to train the discriminator and the learning rate and number of training cycles of the optimization method are optimized to make the training process converge.

[0038] (4) Construct a deep learning network as a process parameter optimizer. During the training process, the forming results in the data set are used as forming requirements, and the optimizer generates optimized process parameters based on these requirements. Then, the optimizer model updates the process parameter generation mode based on the discriminator's judgment results on this set of parameters until the discriminator cannot distinguish the difference between the input process parameter mode and the expert mode.

[0039] (5) Evaluate the actual effect of the process parameters given by the trained optimizer through actual experiments, and fine-tune the optimizer model based on the actual forming experimental results (6) End.

[0040] 5. Establishing LPBF process controller based on reinforcement learning (1) Based on the LPBF process model in the digital twin, the OpenAI Gym specification is used to establish the observation state and reward value of the process model, and to construct a training environment for training the LPBF process control agent.

[0041] (2) Construct a control agent whose actor network and critic network are composed of two deep learning networks respectively. Initialize the actor network and critic network, as well as their optimizers and loss functions.

[0042] (3) Initialize an experience replay buffer to store data generated by the interaction between the agent and the environment.

[0043] (4) According to the initial strategy Output an action Give the environment. The experience data (state , reward R, action etc.) into the buffer area.

[0044] (5) Randomly sample a set of data from the cache, and the critic network calculates the value of the state , and use the Bellman equation to calculate the advantage of an action ; Use the actor network to calculate the probability of each action and compare it with the old probability stored in the buffer to get the importance sampling ratio , calculate the clipping loss value.

[0045] (6) The Adam algorithm is used to update the parameters of the actor network and the critic network. The actor network then outputs new actions to the environment and stores the new states, rewards, and actions in the buffer. Return to step 4 to start a new round of training. The training is terminated when the scheduled training cycle is completed or the reward value is greater than a given threshold.

[0046] (7) Evaluate the actual stability and control performance of the trained agent through actual experiments. If these performances do not meet the requirements, modify the training hyperparameters and training termination conditions, and then retrain the agent until the actual performance of the agent meets the requirements.

[0047] (8) The reinforcement learning-based melt pool temperature control algorithm and the trained intelligent model are transplanted into FPGA programs and burned into the FPGA. Then, the FPGA is installed on the industrial computer as the main control device for the melt pool temperature, and the effectiveness and performance of the FPGA control system are verified in the actual forming experiment.

[0048] (7) End.

[0049] 6. Attach the above modules to Figure 1 The structure shown completes the construction of the LPBF intelligent control system.

[0050] 7. End.

[0051] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. An intelligent control method for laser powder bed fusion forming process, which uses a reinforcement learning agent to realize the intelligent control of the LPBF forming process, and is characterized by: The LPBF forming process control system designed by this method consists of a molten pool temperature control loop and a process parameter optimization loop. The molten pool temperature control loop is the inner loop of the control system, and its main function is to stably maintain the LPBF forming process in the expected state; while the process parameter optimization loop is the outer loop of the control system, and its main function is to timely optimize the process parameters according to the current forming results so that the forming process develops towards the required results. Two reinforcement learning agents are used as controllers in each of these two loops. In order to enable these two agents to learn the optimized action generation strategy, the present invention constructs an agent learning environment based on the LPBF forming process digital twin.

2. The intelligent control method for laser powder bed fusion forming process according to claim 1 is characterized in that: The present invention adopts an LPBF forming process detection system consisting of a coaxial molten pool detection device and a 3D scanner to provide the control system with necessary forming process data; the coaxial molten pool detection device is characterized in that a high-speed camera and a colorimetric pyrometer are used as basic detection equipment, and the molten pool temperature and morphology can be measured unobstructed and uninterruptedly through the coaxial observation optical path. The device is also characterized in that an FPGA computing acceleration card is used for molten pool image processing and control quantity calculation, thereby ensuring the real-time performance of the image processing algorithm and the control algorithm; the 3D scanner is a morphology detection device that uses a straight line laser as structured light and a monochrome industrial camera as a detection element. According to claim 1, the 3D detection device of the present invention is characterized in that the point cloud data is matched with the molten pool data by using the method of converting the point cloud into the forming height, so as to facilitate the establishment of the forming process state data set. The point cloud conversion method refers to a method of converting the original point cloud data into the forming height by point cloud registration, grid model construction, point cloud data resampling and rearrangement.

3. The intelligent control method for laser powder bed fusion forming process according to claim 2 is characterized in that: The so-called coaxial molten pool detection refers to constructing an optical path device with the same optical axis as the high-energy laser, so that the pyrometer and the high-speed near-infrared camera can always maintain a top-view observation of the molten pool state in the laser action area, which can not only ensure that the viewing angle always follows the molten pool, but also ensure that the observation is not blocked. The molten pool image uses a threshold-based method to extract the molten pool area, and then calculates its size, image moment, etc. In addition, due to the high sampling rate of the molten pool detection system, the present invention uses FPGA to process the molten pool image and calculate the molten pool temperature control amount. The molten pool processing algorithm in the FPGA can extract the molten pool area in real time and accurately through image denoising, threshold processing and morphological operations. And the model structure of the molten pool temperature control agent is integrated into the FPGA after appropriate optimization, so that the molten pool control method based on reinforcement learning can adjust the laser power according to the molten pool state in real time, ensuring the real-time performance of the control system.

4. The intelligent control method for laser powder bed fusion forming process according to claim 2 is characterized in that: The digital twin of the LPBF forming process in the present invention is characterized in that the twin consists of a forming process state prediction model and a forming result prediction model, and the output end of the state model is connected to the input end of the result model. The forming process state prediction model is a time series prediction model based on the PatchTST model, which is responsible for predicting the changing trend of the process state over a period of time based on the given process parameters and the current forming state. The forming result prediction model is a prediction model built on the GPT2 model, which can combine the input process parameters and the forming state to predict the forming result in the next stage.

5. The intelligent control method for laser powder bed fusion forming process according to claim 3 is characterized by: The present invention adopts a system structure similar to cascade control (as shown in Figure 1), which is composed of an outer control loop and an inner control loop. The so-called outer control loop adopts a LPBF process process parameter optimization method based on imitation learning; the so-called inner control loop adopts a LPBF process control method based on reinforcement learning. The two control loops use different control frequencies. The outer layer optimizes the process parameters for at least one forming layer in the forming process; and the control frequency of the inner layer is the same as the sampling frequency of the process detection system, generally 1~100kHz. Among them, the outer control loop of the system adopts the process parameter optimization method to control the overall development trend of the forming process; the inner control loop of the system adopts the molten pool temperature control method based on reinforcement learning to maintain the stability of the forming process. The process parameter optimization in the process refers to optimizing the process parameters according to the current forming results in the LPBF forming process in order to achieve the expected forming morphology, so as to ensure that the forming process develops in the expected direction. The process parameter optimization method of the present invention is characterized in that the method adopts an imitation learning method to enable the parameter optimization agent to learn and imitate the expert's process parameter design mode, so that it can give an optimized process parameter adjustment plan according to the current forming results.

6. The intelligent control method for laser powder bed fusion forming process according to claim 1 is characterized by: The present invention is also characterized by designing an environment for training reinforcement learning agents, which is based on the LPBF process digital twin and reward function, and it also clarifies the action space and observation space of the agent. During the agent training, the environment acts as a virtual control object of the agent, and evaluates the different control output strategies given by the agent according to the reward function, guiding it to learn the ideal control output strategy.

7. The intelligent control method for laser powder bed fusion forming process according to claim 4 is characterized in that: The molten pool temperature control method constructed by the present invention is the inner control loop of the cascade control system, which is used to maintain the stability of the forming process. The molten pool temperature control method is characterized in that the method adopts a reinforcement learning agent as the controller of the molten pool temperature, which can greatly improve the adaptability and robustness of the control system. The molten pool control method is also characterized in that the agent controls the molten pool temperature through laser power, and the control quantity generation strategy is learned by the agent from the environment constructed by the digital twin described in claim 3. The agent training strategy adopted in the learning is a method based on proximal strategy optimization.

8. The intelligent control method for laser powder bed fusion forming process according to claim 4 is characterized in that: The present invention constructs an LPBF phased process parameter optimizer based on imitation learning. The so-called LPBF phased process parameter optimization refers to a method for adjusting the processing path and process parameters of the next layer or several layers by optimizing the forming target of the next stage according to the previous forming results. The method makes the forming morphology and mechanical properties gradually approach the design indicators by timely optimizing the process parameters. The so-called imitation learning is to introduce a discriminator containing an expert mode as a reward function for agent training on the basis of reinforcement learning, so as to narrow the search range of the agent strategy and improve the training efficiency. The so-called discriminator refers to a deep learning model composed of a deep convolutional network that can learn the expert mode when setting the LPBF process parameters from an expert data set. The trained discriminator can determine whether the process parameters generated by the agent conform to the expert mode. The process parameter optimizer also includes a generator for generating process parameters, which learns the generation mode of the process parameters according to the feedback of the discriminator, and uses the strategy difference as the loss function during the training process. Then the generated training process is to find a suitable model weight to minimize the loss value.

9. The intelligent control method for laser powder bed fusion forming process according to claim 4 is characterized in that: A forming process anomaly detection system based on deep learning has been developed to detect abnormal states of the forming process and stop the operation in time when an abnormality occurs to prevent equipment damage. The anomaly detection system uses the state interval of the forming process given by the digital twin as a benchmark, and determines whether there is an abnormality by calculating the cross entropy between the benchmark value and the measured value. Then the anomaly classification model comprehensively determines the type of anomaly based on the molten pool temperature and the molten pool image moment measured by the detection system. After receiving the signal of the abnormal type, the control system executes the corresponding disposal plan. The disposal plan includes but is not limited to alarming through sound and light signals, generating alarm logs, shutting down the laser, stopping the galvanometer, stopping the scraper, etc.

Images