Laser fusion wire additive manufacturing process control method and system with machine vision sensing

Abstract

The application discloses a laser fused filament additive manufacturing forming control method and system based on machine vision sensing, wherein during laser fused filament additive manufacturing, a CCD camera is coaxially installed with a laser head and collects a front surface image of a molten pool, the front surface image of the molten pool collected under a distance d from a light filament intersection to a substrate surface is labeled to obtain a data set; a convolutional neural network model is built, and the convolutional neural network model is trained and optimized by using the data set; during laser fused filament additive manufacturing, the front surface image of the molten pool collected by the CCD camera is taken as an input of the convolutional neural network model, a distance value from the light filament intersection to a surface of a stacking layer is calculated in real time, and a closed-loop controller adjusts a wire feeding speed according to a deviation size between a calculated value of the distance from the light filament intersection to the surface of the stacking layer and a set value and a deviation change trend, so that a metal wire filling amount on a stacking path is controlled, and the problem of on-line control of a stacking layer height during laser fused filament additive manufacturing is solved.

Description

Technical Field

[0001] This invention belongs to the field of laser additive manufacturing technology, specifically relating to a laser filament additive manufacturing forming control method and system based on machine vision sensing. Background Technology

[0002] Laser filament additive manufacturing uses a laser beam as the heat source and a metal wire as the filler material. It forms components by melting the metal wire layer by layer, achieving a balance between forming accuracy and efficiency, while offering lower manufacturing costs and higher material utilization. Compared to laser powder additive manufacturing, this technology avoids the risks of metal powder combustion and explosion. Therefore, laser filament additive manufacturing has attracted widespread attention in the industrial manufacturing sector.

[0003] In laser filament additive manufacturing, the forming quality of the deposited layers is affected by the surface condition of the previous layer, fluctuations in process parameters, and heat accumulation. The resulting deviation in layer height is a significant technological challenge in laser filament additive manufacturing, and this deviation accumulates layer by layer, severely reducing the stability of the component forming process and even preventing subsequent deposition. Therefore, effective measures must be taken to address the technical challenges of forming control in laser filament additive manufacturing.

[0004] Currently, there are two main methods to solve this technical problem: 1) Use methods such as optimizing the stacking path and applying air cooling or water cooling during the stacking process to reduce the impact of heat accumulation on the stacking layer height. However, these methods require a large number of process experiments or complex auxiliary equipment, which are costly and have limited control effects, and cannot completely eliminate the forming dimension deviation in the continuous stacking process; 2) Introduce subtractive manufacturing technology, and use CNC milling cutters to mill the stacking path for each stacked layer. However, this method will significantly increase the manufacturing cycle and reduce the material utilization rate, thus increasing costs.

[0005] Chinese invention patent application CN201780055277, entitled "System and Method for Z-Height Measurement and Adjustment in Additive Manufacturing," utilizes a nonlinear mathematical model to extract the stack height information from paraxial image information and inputs the stack height deviation into a motion controller to correct the stack height. However, the accuracy of this sensing method is easily affected by laser plasma and dust, and it requires high installation accuracy of the vision sensor. Chinese invention patent application CN202311492282, entitled "A Real-Time Quality Control Method for Full-Size Additive Manufacturing Components," uses a lateral CCD camera mounted perpendicular to the stacking path to monitor changes in stack height in real time, and achieves stack size control by adjusting laser power or travel speed in real time. However, this laterally mounted CCD camera is prone to spatial interference with the manufactured component, making it difficult to apply to the size control of components with complex geometries. Therefore, a new method is urgently needed to solve the problem of stack size control in laser filament additive manufacturing processes. Summary of the Invention

[0006] The purpose of this invention is to solve the problem of high instability of the deposited layer in the laser filament additive manufacturing process, and to provide a laser filament additive manufacturing forming control method and system based on machine vision sensing.

[0007] To achieve the above-mentioned objectives, the technical solution of this invention is as follows: A laser filament additive manufacturing forming control method based on machine vision sensing. During the laser filament additive manufacturing process, a CCD camera is coaxially mounted with a laser head and acquires images of the front of the molten pool. The front images of the molten pool acquired when the distance from the intersection of the filament (i.e., the intersection of the metal wire tip and the laser beam tip) to the substrate surface is d are labeled to obtain a dataset. A convolutional neural network model is built, and the dataset is used to train and optimize the convolutional neural network model. During the laser filament additive manufacturing process, the front image of the molten pool acquired by the CCD camera is used as the input to the convolutional neural network model. The distance from the filament intersection to the surface of the stacked layer is calculated in real time. The closed-loop controller adjusts the wire feeding speed based on the deviation between the calculated value and the set value of the distance from the filament intersection to the surface of the stacked layer, thereby controlling the amount of metal wire filling on the stacking path and achieving control of the stacked layer height in laser filament additive manufacturing. The method includes the following steps:

[0008] Step 1: Slice the 3D model of the metal component into layers to generate the stacking layer processing path within each layer;

[0009] Step 2: Adjust the position and orientation of the laser head so that the angle between the laser beam and the substrate surface is θ1, the angle between the metal wire and the substrate surface is θ2, and the distance from the intersection of the optical wires to the substrate surface is 0mm; mount the CCD camera on the laser head so that the field of view of the CCD camera is coaxial with the laser beam, and set the sampling frame rate of the CCD camera to f.

[0010] Step 3: Set the wire feeding speed to W s , the laser power to P0, and the traveling speed to T s , turn on the laser metal deposition additive manufacturing system and the CCD camera. The robot controls the laser metal deposition additive manufacturing system to move along the deposition path, so that the distance d from the light-wire intersection point to the substrate surface gradually increases from 0 to d max , where d max is set to 3 - 5 mm, and the CCD camera captures the frontal image of the molten pool in real time; the linear interpolation method is used to calculate the distance value d from the light-wire intersection point corresponding to each frame of the frontal image of the molten pool to the substrate surface n , where 0 < n ≤ N, N is the total number of frames of the frontal images of the molten pool captured during the deposition process, and the calculation method of d n is: d n = x * d max / S, S is the length that the laser metal deposition additive manufacturing system moves from the deposition starting position along the deposition path to the deposition ending position, x is the distance from the current calculation frame position to the deposition starting position, and the calculation method of x is: x = k / f × T s , where k is the number of frames experienced from the deposition starting position to the current calculation frame position, f is the sampling frame rate of the CCD camera, and d n is used to label each frame of the molten pool image to obtain a dataset, and the dataset is divided into a training set, a test set, and a validation set according to the ratio of α:β:(1 - α - β), where α + β < 1;

[0011] Step 4: Build a convolutional neural network model, including 1 input layer, m convolutional layers, m pooling layers, n fully connected layers, and a regression layer; use the dataset to train the convolutional neural network model and optimize the hyperparameters, and save and deploy the trained convolutional neural network model; System deployment is a commonly used term in the field of machine learning, that is, placing the trained convolutional neural network model on other devices or platforms for invocation.

[0012] Step 5: Prepare another substrate and fix it with a fixture, ensuring that the attitude of the laser head remains unchanged, θ 1、 θ2 remains unchanged, set the distance from the light-wire intersection point to the substrate surface to d s , 0 < d s < d max , set the wire feeding speed to W s , the laser power to P0, and the traveling speed to T s , turn on the laser metal deposition additive manufacturing system and the CCD camera. The CCD camera captures the frontal image of the molten pool in real time. At time t, the captured frontal image of the molten pool is input into the convolutional neural network model after cropping, and the distance d p (t) from the light-wire intersection point to the surface of the deposition layer is calculated in real time, and the average value d of the distance from the light-wire intersection point to the surface of the deposition layer is calculateda = (d p (t) + d p (t - 1) + … + d p (t - q)) / (q + 1), where d p (t - 1) is the distance value from the intersection point of the optical filaments at the previous moment to the surface of the deposition layer, t - 1 represents the previous calculation moment, d p (t - q) is the distance from the intersection point of the optical filaments calculated at the q-th previous moment to the surface of the deposition layer, where 2 < q < 5, and the closed-loop controller calculates the deviation e(t) between the average value of the distance from the intersection point of the optical filaments at the current moment to the surface of the deposition layer and the set value = d a - d s and the change trend Δe(t) of the deviation = e(t) - e(t - 1), where d s is the set value of the distance from the intersection point of the optical filaments to the surface of the deposition layer, e(t - 1) is the deviation of the distance from the intersection point of the optical filaments to the surface of the deposition layer at the previous moment, and the closed-loop controller calculates and outputs the wire feeding speed adjustment increment W a to control the filling amount of the wire material on the deposition path. After the first layer is deposited, the laser metal deposition additive manufacturing system is turned off, and the laser metal deposition additive manufacturing system is lifted by a set height h, where h is the thickness of the three-dimensional model's layer slicing;

[0013] Step Six: Continue to execute Step Five until the second layer, the third layer until the remaining layers are deposited, to achieve control of the deposition height of the metal component.

[0014] As a preferred method, the value range of θ1 in Step Two is set to 75° - 85° because when θ1 is close to 90° for incidence, the laser beam is prone to specular reflection when melting metals with high reflectivity, and the reflected laser beam is likely to damage the laser head and the optical fiber. When the value of θ1 is too small, the incident tilt angle of the laser beam is too large, resulting in a decrease in the energy utilization rate of the laser beam; the value range of the included angle θ2 is set to 30° - 45° because when θ1 is too large or too small, the absorption efficiency of the metal wire to the laser beam will decrease; the value range of the parameter f is set to 20Hz - 50Hz because when f is too small, the number of samples collected during the monitoring process is small, and it cannot effectively reflect the distance value from the intersection point of the optical filaments at the current moment to the surface of the deposition layer, affecting the monitoring accuracy. When f is too large, the computational amount of the convolutional neural network model becomes larger, and the consumption of computing resources increases significantly.

[0015] As a preferred approach, the parameter α in step three is set to a range of 0.6-0.75 because if α is too small, the training samples for the convolutional neural network model are few, resulting in poor performance; if α is too large, the convolutional neural network model is prone to overfitting, leading to a decrease in generalization ability. The parameter β is set to a range of 0.15-0.25 because if β is too small, the representativeness of the test set decreases, failing to effectively reflect the performance of the convolutional neural network model; if β is too large, the proportion of the training set decreases, affecting the performance of the convolutional neural network model.

[0016] As a preferred approach, the parameters m and n in step four are positive integers. The range of m is set to 3-7 because if m is too small, it will affect the ability of the convolutional neural network model to extract features from image information, resulting in a decrease in the performance of the convolutional neural network model. If m is too large, there will be too many training parameters for the model, significantly increasing the consumption of computing resources. The range of n is set to 2-4 because if n is too small, it will affect the ability of the convolutional neural network model to map the extracted features to high-dimensional features, resulting in a decrease in the performance of the convolutional neural network model. If n is too large, it will easily lead to overfitting of the convolutional neural network model.

[0017] As a preferred method, the convolutional neural network model training and hyperparameter optimization in step four involve training the convolutional neural network model using a training set, evaluating the deviation between the calculated and true values ​​using the root mean square loss function, updating the model weight coefficients using a gradient momentum descent optimizer, performing accuracy testing on the trained convolutional neural network model using a test set, and verifying the accuracy of the trained model using a validation set. The hyperparameters include the number of convolutional layers, learning rate, activation function, number of fully connected layers, and number of neurons. The system deployment of the convolutional neural network is implemented through the onnxruntime module.

[0018] The second objective of this invention is to provide a preferred embodiment comprising: a substrate 1, a wire feeder 2, a laser power supply 3, a computer 4, a CCD camera 5, a control box 6, and a laser welding torch 7; wherein the CCD camera 5 is mounted on the laser welding torch 7 via a clamp, and the CCD camera 5 is connected to the computer 4 via a USB data interface; the wire feeder 2 and the control box 6 are connected to the computer 4 via a data acquisition card; the CCD camera 5 acquires real-time images of the front of the molten pool; the computer 4, after cropping the front image of the molten pool, inputs it into a convolutional neural network model to calculate the distance from the intersection of the optical filaments to the surface of the deposited layer; the closed-loop controller in the computer 4 calculates the deviation between the calculated value and the set value of the distance from the intersection of the optical filaments (i.e., the intersection of the metal wire tip and the laser beam tip) to the surface of the deposited layer at the current moment, and performs feedback control based on the trend of the deviation; the closed-loop controller outputs an incremental adjustment of the wire feeding speed to change the wire feeding speed of the wire feeder 2, thereby changing the amount of metal wire filling on the deposited path, thus realizing online control of the deposited layer height;

[0019] The principle of this invention is as follows: Figure 2 As shown, in the laser filament additive manufacturing process, a CCD camera coaxially mounted with the laser beam is used to acquire images of the front of the molten pool. The distance from the intersection of the filaments to the substrate surface is used to label the images of the front of the molten pool to obtain a dataset. The dataset is used to train and optimize the constructed convolutional neural network model. In the laser filament additive manufacturing process, the front images of the molten pool acquired in real time by the CCD camera are input into the trained convolutional neural network model to calculate the distance from the intersection of the filaments to the surface of the stacked layer online. The closed-loop controller calculates the deviation between the calculated value and the set value of the distance from the intersection of the filaments to the surface of the stacked layer, and outputs the adjustment increment of the wire feeding speed to change the metal filling amount on the stacking path, thereby realizing real-time control of the stacked layer size.

[0020] The key advantages of this invention are: in the laser filament additive manufacturing process, a coaxial vision sensing method is used to acquire a frontal image of the molten pool. This image serves as input to a convolutional neural network model to calculate the distance from the filament intersection point to the surface of the deposited layer. Real-time feedback is provided on changes in the deposited layer height. The closed-loop controller outputs the adjustment increment of the wire feeding speed by calculating the deviation between the calculated value and the set value of the distance from the filament intersection point to the deposited layer surface, and the trend of this deviation, thereby achieving online control of the deposited layer size. Compared to traditional methods, this invention combines coaxial vision sensing with a convolutional neural network model to control the deposited layer height in laser filament additive manufacturing, solving the spatial interference problem between traditional vision sensors and additive manufacturing components, and achieving real-time control of the deposited layer height in the laser filament additive manufacturing process. Attached Figure Description

[0021] Figure 1 This is a schematic diagram showing the installation positions of the CCD camera, metal wire, and laser heat source of this invention;

[0022] Figure 2 This is a schematic diagram of the laser filament additive manufacturing process of the present invention and the distance from the intersection of the filaments to the surface of the deposited layer;

[0023] Figure 3 This is a schematic diagram of the laser filament additive manufacturing stacking layer height control system of the present invention.

[0024] 1 is the substrate, 2 is the wire feeder, 3 is the laser power supply, 4 is the computer, 5 is the CCD camera, 6 is the control box, and 7 is the laser welding torch. Detailed Implementation

[0025] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand the other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific implementation manners. All details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0026] Example 1

[0027] As Figure 1 shown, this embodiment provides a forming control method for laser wire additive manufacturing with machine vision sensing. During the laser wire additive manufacturing process, a CCD camera is coaxially installed with the laser head and captures the front image of the molten pool. Label operations are performed on the front image of the molten pool captured when the distance from the intersection point of the light and wire, that is, the intersection point of the front end of the metal wire and the front end of the laser beam, to the substrate surface is d, to obtain a data set; a convolutional neural network model is built, and the data set is used to train and optimize the convolutional neural network model; during the laser wire additive manufacturing process, the front image of the molten pool captured by the CCD camera is used as the input of the convolutional neural network model, and the distance value from the intersection point of the light and wire to the surface of the stacking layer is calculated in real time. The closed-loop controller adjusts the wire feeding speed according to the deviation magnitude and deviation change trend between the calculated value and the set value of the distance from the intersection point of the light and wire to the surface of the stacking layer, so as to control the filling amount of the metal wire on the stacking path and achieve the control of the stacking layer height in laser wire additive manufacturing, including the following steps:

[0028] Step 1: Perform layer slicing on the three-dimensional model of the metal component to generate the stacking layer processing path within each layer slice;

[0029] Step 2: Adjust the position and posture of the laser head so that the angle between the laser beam and the substrate surface is θ1, the angle between the metal wire and the substrate surface is θ2, and the distance from the intersection point of the light and wire to the substrate surface is 0 mm; install the CCD camera on the laser head so that the field of view of the CCD camera is coaxial with the laser beam, and set the sampling frame rate of the CCD camera to f;

[0030] Step 3: As Figure 2 shown, set the wire feeding speed to W s , the laser power to P0, the walking speed to T s , start the laser wire additive manufacturing system and the CCD camera, and the robot controls the laser wire additive manufacturing system to move along the stacking path, so that the distance d from the intersection point of the light and wire to the substrate surface gradually increases from 0 to d max , where d max is set to 3 - 5 mm, and the CCD camera captures the front image of the molten pool in real time; use the linear interpolation method to calculate the distance value d n from the intersection point of the light and wire corresponding to each frame of the front image of the molten pool to the substrate surface, where 0 < n ≤ N, N is the total number of frames of the front image of the molten pool captured during the stacking process, dn The calculation method of d is as follows: n d = x * d max / S, where S is the length that the laser - wire additive manufacturing system moves from the starting position of deposition along the deposition path to the ending position of deposition, x is the distance from the current calculation frame position to the starting position of deposition, and the calculation method of x is: x = k / f × T s , where k is the number of frames experienced from the starting position of deposition to the current calculation frame position, f is the sampling frame rate of the CCD camera. Using d n to perform label operations on each frame of the molten pool image to obtain a dataset. The dataset is divided into a training set, a test set, and a validation set according to the ratio of α:β:(1 - α - β), where α + β < 1;

[0031] Step 4: Build a convolutional neural network model, including 1 input layer, m convolutional layers, m pooling layers, n fully - connected layers, and a regression layer; use the dataset to train the convolutional neural network model and optimize hyperparameters, and save and deploy the trained convolutional neural network model; System deployment is a commonly used term in the field of deep learning, that is, putting the existing model on other devices or platforms for invocation.

[0032] Step 5: Prepare a new substrate again and fix it with a fixture to ensure that the posture of the laser head remains unchanged, θ 1、 θ2 remains unchanged. Set the distance from the light - wire intersection point to the substrate surface as d s , 0 < d s < d max , set the wire - feeding speed as W s , the laser power as P0, and the walking speed as T s , turn on the laser - wire additive manufacturing system and the CCD camera. The CCD camera collects the front - face image of the molten pool in real - time. At time t, the collected front - face image of the molten pool is input into the convolutional neural network model after cropping, and the distance d p (t) from the light - wire intersection point to the deposition layer surface is calculated in real - time. Calculate the average value d a = (d p (t)+d p (t - 1)+…+d p (t - q)) / (q + 1), where d p (t - 1) is the distance value from the light - wire intersection point to the deposition layer surface at the previous moment, t - 1 represents the previous calculation moment, and d p (t - q) is the distance from the light - wire intersection point to the deposition layer surface calculated at the q - th previous moment, where 2 < q < 5. The closed - loop controller calculates the deviation e(t) between the average value of the distance from the light - wire intersection point to the deposition layer surface at the current moment and the set value: e(t)=d a - d sand the trend of the deviation Δe(t)=e(t)-e(t-1), where d s Let e(t) be the set value of the distance from the intersection of the filaments to the surface of the stacked layer, and let e(t-1) be the deviation of the distance from the intersection of the filaments to the surface of the stacked layer at the previous moment. The closed-loop controller calculates the output wire feed speed adjustment increment W based on e(t) and Δe(t). a This controls the amount of metal wire filling along the stacking path. After the first layer is stacked, the laser filament additive manufacturing system is shut down, and the laser filament additive manufacturing system is raised to a set height h, where h is the thickness of the three-dimensional model layer slice.

[0033] Step Six: Continue with Step Five until the second and third layers are stacked, and so on until the remaining layers are stacked, thus controlling the stacking height of the metal components.

[0034] In step two, the included angle θ1 ranges from 75° to 85°, the included angle θ2 ranges from 30° to 45°, and the parameter f ranges from 20Hz to 50Hz. The value range of θ1 mentioned in step two is set to 75°-85° because when θ1 is close to 90° of incident angle, the laser beam is prone to specular reflection when melting metals with high reflectivity. The reflected laser beam is prone to damaging the laser head and optical fiber. If the value of θ1 is too small, the incident tilt angle of the laser beam is too large, resulting in a decrease in the energy utilization rate of the laser beam. The value range of the included angle θ2 is set to 30°-45° because when θ1 is too large or too small, the absorption efficiency of the metal wire to the laser beam will decrease. The value range of parameter f is set to 20Hz-50Hz because when f is too small, the number of samples collected during the monitoring process is small, which cannot effectively reflect the distance value from the intersection of the optical filaments to the surface of the stacked layer at the current moment, affecting the monitoring accuracy. If f is too large, the computational load of the convolutional neural network model increases, and the consumption of computing resources increases significantly.

[0035] In step three, the parameter α ranges from 0.6 to 0.75, and the parameter β ranges from 0.15 to 0.25. The range of 0.6-0.75 is chosen because if α is too small, the training samples for the convolutional neural network model are insufficient, resulting in poor model performance; if α is too large, the model is prone to overfitting, leading to decreased generalization ability. The range of β is set to 0.15-0.25 because if β is too small, the representativeness of the test set decreases, failing to effectively reflect the performance of the convolutional neural network model; if β is too large, the proportion of the training set decreases, affecting the performance of the convolutional neural network model.

[0036] In step four, the parameters m and n are positive integers, with m ranging from 3 to 7 and n ranging from 2 to 4. The range of m is set to 3-7 because if m is too small, it affects the convolutional neural network model's ability to extract features from image information, leading to a decrease in model performance; if m is too large, the model has too many training parameters, significantly increasing computational resource consumption. The range of n is set to 2-4 because if n is too small, it affects the convolutional neural network model's ability to map extracted features to high-dimensional features, also leading to a decrease in model performance; if n is too large, it can easily cause overfitting in the convolutional neural network model.

[0037] Step four, the training and hyperparameter optimization of the convolutional neural network model, involves training the model using a training set, evaluating the deviation between the calculated and true values ​​using the root mean square loss function, updating the model weight coefficients using a gradient momentum descent optimizer, testing the accuracy of the trained convolutional neural network model using a test set, and validating the accuracy of the trained model using a validation set. The hyperparameters include the number of convolutional layers, learning rate, activation function, number of fully connected layers, and number of neurons. The system deployment of the convolutional neural network is implemented through the onnxruntime module.

[0038] like Figure 3 As shown, this embodiment also provides a machine vision sensing-based laser filament additive manufacturing forming control system, including: a substrate 1, a wire feeder 2, a laser power supply 3, a computer 4, a CCD camera 5, a control box 6, and a laser welding torch 7; wherein, the CCD camera 5 is mounted on the laser welding torch 7 by a fixture, and the CCD camera 5 is connected to the computer 4 via a USB data interface, and the wire feeder 2 and the control box 6 are connected to the computer 4; the CCD camera 5 acquires real-time images of the front of the molten pool, and the computer 4 inputs the cropped images of the front of the molten pool into a convolutional neural network model to calculate the distance from the intersection of the filaments to the surface of the deposited layer, and the closed-loop controller in the computer 4 calculates the deviation between the calculated value and the set value of the distance from the intersection of the filaments (i.e., the intersection of the metal wire tip and the laser beam tip) to the surface of the deposited layer at the current moment, and performs feedback control based on the trend of the deviation. The closed-loop controller outputs the wire feeding speed adjustment increment to change the wire feeding speed of the wire feeder 2, thereby changing the filling amount of metal wire on the deposited path, thereby realizing online control of the deposited layer height.

[0039] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A laser filament additive manufacturing forming control method based on machine vision sensing, characterized in that, In laser filament additive manufacturing, a CCD camera is coaxially mounted with the laser head to capture images of the front of the molten pool. The distance from the intersection of the filaments (the point where the metal wire tip intersects with the laser beam tip) to the substrate surface is... d The process involves labeling the front images of the molten pool acquired in real time to obtain a dataset; building a convolutional neural network model; and training and optimizing the convolutional neural network model using the dataset. During laser fused wire additive manufacturing, the front images of the molten pool acquired by a CCD camera are used as input to the convolutional neural network model to calculate the distance from the filament intersection point to the surface of the deposited layer in real time. The closed-loop controller adjusts the wire feeding speed based on the deviation between the calculated and set values ​​of this distance, thereby controlling the amount of metal wire filling along the depositing path and achieving control over the height of the deposited layer in laser fused wire additive manufacturing. This includes the following steps: Step 1: Slice the 3D model of the metal component into layers to generate the stacking layer processing path within each layer; Step 2: Adjust the position and orientation of the laser head so that the angle between the laser beam and the substrate surface is... θ 1. The angle between the metal wire and the substrate surface is... θ 2. The distance from the intersection of the optical filaments to the substrate surface is 0mm; mount the CCD camera on the laser head, making the CCD camera's field of view coaxial with the laser beam, and set the CCD camera's sampling frame rate to [value missing]. f ; Step 3: Set the wire feeding speed to... W s The laser power is P 0, walking speed is T s The laser filament additive manufacturing system and CCD camera are activated. The robot controls the laser filament additive manufacturing system to move along the deposition path, adjusting the distance from the intersection of the filaments to the substrate surface. d Gradually increase from 0 to d max ,in d max The CCD camera is set to 3-5mm to capture real-time images of the front of the molten pool; linear interpolation is used to calculate the distance from the intersection of the photofilaments to the substrate surface for each frame of the molten pool front image. d n , where 0 < n ≤ N , N This represents the total number of frames of images of the front of the molten pool acquired during the deposition process. d n The calculation method is as follows: d n = x * d max / S , S This refers to the length that the laser filament additive manufacturing system moves from the initial position of the deposition path to the final position. x This is the distance from the current frame position to the start of the stacking. x The calculation method is as follows: x = k / f × T s ,in k This represents the number of frames elapsed from the initial position of the stack to the current computation frame position. f It is the sampling frame rate of the CCD camera, using d n Label each frame of the molten pool image to obtain a dataset, which is then sorted according to... α : β :(1- α - β The dataset is divided into training, testing, and validation sets, with the following proportions: α + β < 1; Step 4: Build a convolutional neural network model, containing one input layer. m One convolutional layer, m A pooling layer, n One fully connected layer and one regression layer; the convolutional neural network model is trained and its hyperparameters are optimized using a dataset; the trained convolutional neural network model is then saved and deployed to the system. Step 5: Prepare a new substrate and fix it with a clamp to ensure that the laser head's orientation remains unchanged. θ 1、 θ 2 remains unchanged, and the distance from the intersection of the optical filaments to the substrate surface is set to... d s , 0 < d s < d max Set the wire feeding speed to W s The laser power is P 0, walking speed is T s The laser filament additive manufacturing system and CCD camera are activated, and the CCD camera acquires real-time images of the front of the molten pool. t At any given moment, the captured frontal image of the molten pool is cropped and input into a convolutional neural network model to calculate the distance from the intersection of the photofilaments to the surface of the deposited layer in real time. d p ( t ), calculate the average distance from the intersection of the filaments to the surface of the stacked layer. d a =( d p ( t )+ d p ( t- 1)+…+ d p ( t- q )) / ( q+ 1), of which d p ( t- 1) is the distance from the intersection of the filaments to the surface of the deposited layer at the previous moment. t- 1 indicates the previous calculation time. d p ( tq ) is the first q The distance from the intersection of the filaments to the surface of the stacked layer is calculated at each time step, where 2 < q <5. The closed-loop controller calculates the deviation between the average distance from the intersection of the filaments to the surface of the deposited layer at the current moment and the set value. e ( t )= d a - d s and the trend of deviation Δ e ( t )= e ( t )- e ( t -1), where d s This is the set value for the distance from the intersection of the optical filaments to the surface of the deposited layer. e ( t -1) represents the distance deviation from the intersection of the filaments to the surface of the deposited layer at the previous moment, which the closed-loop controller determines based on... e ( t ) and Δ e ( t ) Calculate the output wire feed speed adjustment increment W a This controls the amount of metal wire filling along the stacking path. After the first layer is stacked, the laser filament additive manufacturing system is shut down and raised to a set height. h ,in h The thickness of slices for layering a 3D model; Step Six: Continue with Step Five until the second and third layers are stacked, and so on until the remaining layers are stacked, thus controlling the stacking height of the metal components.

2. The laser filament additive manufacturing forming control method with machine vision sensing according to claim 1, characterized in that, The included angle in step two θ The value of 1 ranges from 75° to 85°, and the included angle is... θ The value of 2 ranges from 30° to 45°. f The value range is 20Hz-50Hz.

3. The laser filament additive manufacturing forming control method with machine vision sensing according to claim 1, characterized in that, The parameters in step three α The value range is 0.6-0.

75. β The value range is 0.15-0.

25.

4. The laser filament additive manufacturing forming control method based on machine vision sensing according to claim 1, characterized in that, The parameters in step four m , n It is a positive integer. m The value range is 3-7. n The value range is 2-4.

5. The laser filament additive manufacturing forming control method using machine vision sensing according to claim 1, characterized in that, Step four, the training and hyperparameter optimization of the convolutional neural network model, involves training the model using a training set, evaluating the deviation between the calculated and true values ​​using the root mean square loss function, updating the model weight coefficients using a gradient momentum descent optimizer, testing the accuracy of the trained convolutional neural network model using a test set, and validating the accuracy of the trained model using a validation set. The hyperparameters include the number of convolutional layers, learning rate, activation function, number of fully connected layers, and number of neurons. The system deployment of the convolutional neural network is implemented through the onnxruntime module.

6. A machine vision-sensing laser filament additive manufacturing forming control system, used to implement the control method according to any one of claims 1 to 5, characterized in that, include: The components include a substrate (1), a wire feeder (2), a laser power supply (3), a computer (4), a CCD camera (5), a control box (6), and a laser welding torch (7). The CCD camera (5) is mounted on the laser welding torch (7) via a clamp and connected to the computer (4) via a USB data interface. The wire feeder (2) and the control box (6) are connected to the computer (4). The CCD camera (5) acquires real-time images of the front of the molten pool. The computer (4) inputs the cropped images of the front of the molten pool into a convolutional neural network model to calculate the distance from the intersection of the optical filaments to the surface of the deposited layer. The closed-loop controller in the computer (4) calculates the deviation between the calculated value and the set value of the distance from the intersection of the optical filaments (i.e., the intersection of the front end of the metal wire and the front end of the laser beam) to the surface of the deposited layer and performs feedback control. The closed-loop controller outputs the wire feeding speed adjustment increment to change the wire feeding speed of the wire feeder (2) and change the amount of metal wire filling on the deposited path, thereby realizing online control of the deposited layer height.

Images