Mechanism-guided data-driven prediction method for melt pool morphology evolution in selective laser melting process

Abstract

The present invention discloses a mechanism-guided data-driven method for predicting the evolution of the molten pool morphology in a selective laser melting process, comprising the following steps: 1. constructing a molten pool image dataset; 2. extracting morphological features such as the length, width, area, perimeter, and aspect ratio of the molten pool by processing the molten pool image; 3. constructing molten pool samples and labels; 4. calculating the theoretical temperature field of the molten pool surface using a mechanism analysis model; 5. calculating the angle between the laser movement direction and the shielding airflow direction; 6. constructing a mechanism-guided data-driven prediction model for the evolution of the molten pool morphology; 7. constructing a loss function; and 8. training the model. By integrating the melting mechanism and sensor data of the selective laser melting process, the present invention can not only accurately predict the evolution trend of the molten pool morphology in advance, but also enhance the prediction of the abnormal disturbance amount in the molten pool morphology evolution process caused by the historical scanning trajectory, thereby achieving accurate prediction of the molten pool morphology evolution.

Description

Technical Field

[0001] The present invention relates to the technical field of metal additive manufacturing process monitoring, and in particular to a mechanism-guided data-driven method for predicting the molten pool morphology evolution in a selective laser melting process. Background Art

[0002] Selective laser melting technology has unique advantages in manufacturing high-quality components with complex structures and internal features in the fields of biomedical engineering, automobiles, aerospace, etc. It is a very promising and environmentally friendly metal additive manufacturing technology. However, the manufactured parts often have defects such as pores, cracks and warping, which increases the manufacturing costs caused by trial and error and post-processing, and hinders its large-scale industrial application. The selective laser melting process involves complex heat transfer and fluid flow. The resulting molten pool is affected by surface tension, Marangoni force, recoil pressure and other factors. Its abnormal evolution will lead to the formation of defects, which in turn leads to microstructural inhomogeneity and significantly affects the mechanical properties and fatigue life of the formed parts. Therefore, in order to ensure the final quality of the parts, it is necessary to control the evolution of the molten pool morphology during the selective laser melting process. Accurately predicting the evolution of the molten pool morphology is an important basis for achieving melting state control and reducing process defects.

[0003] Currently, there are three approaches to modeling melt pool morphology evolution during selective laser melting (SLM): physics-based models, process data-based methods, and historical sensor data-based methods. Mainstream physics-based methods include numerical models based on finite elements (FEMs) and analytical models. Numerical models based on FEMs, due to their fine discretization in time and space, consume significant time and storage costs. Analytical models retain essential heat transfer characteristics while sacrificing some of the complex physical properties of the melt pool incorporated in the numerical model. They integrate material properties, process parameters, and scanning strategies in a mechanistically consistent manner to calculate the melting temperature field, enabling rapid calculation of melt pool morphology evolution across multiple tracks and even multiple layers at the part scale. Process data-based methods utilize machine learning models for faster predictions, but the integration of diverse process data may not align with the melting mechanism, resulting in low generalization capabilities. In actual manufacturing, process parameter fluctuations caused by machine errors, non-fixed material properties such as absorptivity, and random interference during melt pool motion can lead to unexpected changes in melt pool morphology. Therefore, methods based on physics models and process data struggle to accurately capture the transient evolution of melt pool morphology. There is inevitably noise or even erroneous data in the sensing data caused by environmental factors and sensor errors. Due to the lack of guidance and constraints from physical knowledge, the model may learn a false relationship between the historical melting state and the melt pool evolution, thereby reducing the performance of the prediction model. Summary of the Invention

[0004] In order to address the deficiencies of the above-mentioned prior art, the present invention proposes a mechanism-guided data-driven method for predicting the molten pool morphology evolution in the selective laser melting process, in order to achieve accurate prediction of the transient evolution of the molten pool morphology in the selective laser melting process across working conditions, thereby providing a basis for controlling the molten pool morphology evolution process and maintaining a good melting process, thereby reducing or even eliminating defects in the forming process, thereby improving the three-dimensional morphology accuracy, mechanical properties, and fatigue life of the formed parts, and ensuring the quality of the final product.

[0005] In order to achieve the above-mentioned object, the present invention adopts the following technical solutions:

[0006] The invention provides a mechanism-guided data-driven method for predicting the molten pool morphology evolution in a selective laser melting process, which comprises the following steps:

[0007] Step 1: Use a temperature-calibrated high-speed infrared camera to capture the melt pool image of the selective laser melting process in a coaxial manner and at a fixed frequency, and construct a melt pool image dataset ;in, represents the melt pool image at the i-th moment, is the number of melt pool images; L and H are the length and height of the melt pool image respectively;

[0008] Step 2: By taking the molten pool image at the i-th moment Processing is performed to extract the morphological features of the molten pool at the i-th moment , including: length ,width ,area ,perimeter , aspect ratio , thereby constructing a melt pool morphology feature dataset ;

[0009] Step 3: Momentary melt pool image To Momentary melt pool image As a sample , the morphological characteristics of the molten pool at the i-th moment To Morphological characteristics of the melt pool at each moment As a sample Tags ; Morphological characteristics of the melt pool at each moment To Morphological characteristics of the melt pool at each moment As a sample Tags Corresponding historical melt pool feature sequence ;

[0010] Step 4: Calculate the theoretical temperature field of the molten pool surface at the i-th moment based on material parameters, process parameters and scanning strategy ; thus obtaining the Time to The theoretical temperature field of the molten pool surface at time t, thus obtaining the label The corresponding theoretical temperature field sequence of the molten pool surface ;

[0011] Step 5: Calculate labels The angle between the corresponding laser movement direction and the protective airflow direction in the clockwise direction ;in, represents the angle between the laser moving direction and the shielding airflow direction in the clockwise direction at the i-th moment;

[0012] Step 6: Construct a prediction network for the molten pool morphology evolution, including: theoretical temperature field feature extraction module, improved attention mechanism module, mechanism and data feature fusion module, and 、 and Process and obtain Time to Moment-by-moment prediction of melt pool morphology feature sequences ;

[0013] Step 7: Use formula (13) to construct the loss function :

[0014] (13)

[0015] In formula (10), is the predicted value of the melt pool morphology feature sequence, is the difference value of the melt pool morphology feature sequence, is the predicted value of the sequence difference of the molten pool morphology characteristics, A parameter to determine the relative importance of the loss term;

[0016] Step 8: Use the gradient descent method to train the molten pool morphology feature prediction network, and calculate the total loss function loss to update the model parameters. When the total loss function loss converges, the trained molten pool morphology feature prediction model is obtained, which is used to predict the future molten pool morphology feature evolution process during the selective laser melting process.

[0017] The mechanism-guided data-driven method for predicting the molten pool morphology evolution in the selective laser melting process described in the present invention is also characterized in that step 2 includes the following steps:

[0018] Step 2.1: Liquidus temperature of a given powder material , extract the melt pool image at the i-th moment In, with The absolute value of the difference is less than the error The pixel points are then fitted into the i-th closed curve using the least squares method. ;

[0019] Step 2.2: Move two first straight lines parallel to the direction of laser movement from the center of the spot to both sides. When the two first straight lines intersect the i-th closed curve When there is only one intersection point, calculate the distance between the two first straight lines to obtain the i-th closed curve Width;

[0020] Move two second straight lines perpendicular to the moving direction of the laser from the center of the spot to both sides. When there is only one intersection point, calculate the distance between the two second straight lines to obtain the i-th closed curve length;

[0021] According to the instantaneous field of view of the high-speed infrared camera, the i-th closed curve After converting the length, width, area and perimeter, the length of the molten pool is obtained. ,width ,area and perimeter ;

[0022] The aspect ratio of the melt pool is obtained by calculating the ratio between the length and width of the melt pool. .

[0023] Furthermore, step 4 includes the following steps:

[0024] Step 4.1: Take the center of the laser spot at the i-th moment As the base point, create a rectangle with the length along the direction of laser movement at moment i and the width perpendicular to the direction of laser movement at moment i; among them, the center of the laser spot at moment i is The distances to the front and back edges of the rectangle are set to and , The distance to the left and right edges is set to ;

[0025] Step 4.2: Set the point set resolution to , extracted from the rectangle The coordinates of the points ,in, is the coordinate of the j-th point at the i-th moment in the working coordinate system of the selective laser melting equipment;

[0026] Step 4.3: Calculate the label using equations (1) to (3) The theoretical temperature of the jth point at the corresponding i-th moment :

[0027] (1)

[0028] (2)

[0029] (3)

[0030] In formula (1) to formula (3), To calculate the The number of historical laser scanning points used at the theoretical temperature at the moment; is the thermal diffusivity of the powder alloy, is the density of the powder material, is the initial temperature of the environment; For The specific heat capacity at the temperature, m is the coefficient of specific heat capacity changing with temperature; P is the laser power, is the absorption coefficient of the powder particles, is the beam radius, is the laser beam intensity distribution factor on the XY plane, is the center of the laser spot at the i-th moment The coordinates of the XY plane in the working coordinate system, is the time-integrated variable; is the time-integrated variable The time difference to the i-th moment; is the time-integrated variable The integrated parameter of laser properties at the i-th moment, h is the height of the body heat source;

[0031] Step 4.4: Calculate the label according to the process of step 4.3 Corresponding Middle point to points and point to The theoretical temperature at point i is obtained, thus obtaining the theoretical temperature field of the molten pool surface at moment i. .

[0032] Furthermore, step 5 includes the following steps:

[0033] Step 5.1: Calculate the label using formula (4) The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the corresponding moment i :

[0034] (4)

[0035] In formula (4), is the laser moving unit direction vector at the i-th moment, is the unit direction vector of the protective airflow at the i-th moment, is the function for calculating the clockwise angle;

[0036] Step 5.2: Follow the procedure in step 5.1 to calculate the Time to The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the moment and the Time to The angle between the laser moving direction at the moment and the protective airflow direction in the clockwise direction is obtained, so as to obtain the label Corresponding laser movement direction and protective airflow angle sequence .

[0037] Furthermore, step 6 includes the following steps:

[0038] Step 6.1: The theoretical temperature field feature extraction module includes: two convolutional pooling layers and one fully connected layer;

[0039] Will Input the theoretical temperature field feature extraction module for processing and obtain the label Corresponding mechanism characteristic sequence ,in, Indicates the The mechanistic characteristics of the moment;

[0040] Step 6.2: The improved attention mechanism module includes: a spatiotemporal difference extraction unit, a fully connected layer, and a Softmax function;

[0041] The spatiotemporal difference extraction unit uses equations (5) to (7) to calculate the Moment and Time difference , No. Laser spot center at the moment With the Laser spot center at the moment Distance in the X direction , Y direction distance , Euclidean distance , and form the The time-space difference sequence of the historical melt pool corresponding to the moment , thus obtaining the Time to The time-space difference sequence of the historical melt pool corresponding to the moment ;

[0042] (5)

[0043] (6)

[0044] (7)

[0045] In formula (6), For the The center of the laser spot at the moment Coordinates in the XY plane;

[0046] Will Input to the fully connected layer for processing to obtain the The spatiotemporal characteristics of moments , and continue to input it into the Softmax function for processing, and get the Time to The molten pool at the moment Attention weight of the moment melting pool ; thus obtaining the label The corresponding historical melt pool Time to Attention weight at the moment ;in, Indicates the The attention weight of the melting pool at the moment; Indicates the first moment corresponding to the The spatiotemporal characteristics of the moment, Indicates the first moment corresponding to the Attention weight of the melt pool feature at each moment;

[0047] Step 6.3: The mechanism and data feature fusion module includes: an encoder module and a decoder module;

[0048] Step 6.3.1: The encoder module is an encoder based on the encoding neural ODE-GRU structure, and uses formulas (8) to (9) to calculate the first The encoded input hidden state at time , thus obtaining the Time to The sequence of encoded hidden states at time ;

[0049] (8)

[0050] (9)

[0051] In formula (8) to formula (9), represents the first The pre-hidden state at the moment, For the The encoded input hidden state at time t, is the fully connected layer in the encoding neural ODE, To encode the parameters of the neural ODE, GRU stands for Gated Recurrent Unit; For the Encoded input features at time instant; Indicates the The mechanistic characteristics of the moment;

[0052] Step 6.3.2: Calculate the context vector at the i-th moment using formula (10) ; thus obtaining the label The corresponding time from moment i to moment The context vector sequence at each moment ;

[0053] (10)

[0054] In formula (10), is the hidden state of the encoded input at the kth moment, represents the attention weight of the melt pool feature at the kth moment corresponding to the i-th moment;

[0055] Step 6.3.3: The decoder is a decoder based on the decoding neural ODE-GRU, and uses formula (11) to formula (12) to calculate the first The decoded input hidden state at time , thus obtaining the Time to The decoded hidden state sequence at time ;

[0056] (11)

[0057] (12)

[0058] In formula (11) to formula (12), represents the first output of the decoded neural ODE The pre-hidden state at the moment, For the The decoded input hidden state at time t, To decode the fully connected layer in the neural ODE, is the parameter in the decoded neural ODE; For the Decoded input features at time t;

[0059] Will After passing through a fully connected layer, the output Time to Moment-by-moment prediction of melt pool morphology feature sequences .

[0060] The electronic device of the present invention includes a memory and a processor, and is characterized in that the memory is used to store a program that supports the processor to execute the method for predicting the molten pool morphology evolution in the selective laser melting process, and the processor is configured to execute the program stored in the memory.

[0061] The present invention provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, which is characterized in that when the computer program is run by a processor, the steps of the method for predicting the molten pool morphology evolution in the selective laser melting process are executed.

[0062] Compared with the prior art, the present invention has the following beneficial effects:

[0063] 1. The present invention integrates material properties, process parameters and scanning strategies through a mechanism analysis model, and combines machine learning methods to extract theoretical temperature field characteristics for the prediction of molten pool morphology evolution. It overcomes the compatibility problem between prediction speed and accuracy of existing numerical simulation technology, and greatly improves the efficiency of accurate modeling of molten pool morphology evolution.

[0064] 2. The present invention predicts the evolution of the molten pool morphology by fusing the sensor data features recorded in the historical scanning trajectory with the mechanism features, overcoming the problem in the existing technology that it is impossible to consider the abnormal disturbance of the historical scanning trajectory on the molten pool morphology evolution process, thereby improving the accuracy of the transient evolution modeling of the molten pool morphology.

[0065] 3 The present invention utilizes advanced technologies in the field of machine learning to extract theoretical temperature field features through a CNN model, and integrates the features of the angle between the laser movement direction and the protective airflow, as well as the historical molten pool features, into an encoder module and a decoder module based on ODE-GRU to predict the molten pool morphology evolution. It also establishes an attention mechanism that is more in line with the actual processing process, thereby improving the generalization of the molten pool morphology evolution prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a schematic diagram of the melt pool infrared camera monitoring;

[0067] Figure 2 This is the structural diagram of the theoretical temperature field feature extraction module on the molten pool surface;

[0068] Figure 3 This is the improved attention mechanism module diagram structure diagram;

[0069] Figure 4 It is the structural diagram of the mechanism and data feature fusion module. DETAILED DESCRIPTION

[0070] In this embodiment, a mechanism-guided data-driven method for predicting the evolution of the molten pool morphology in the selective laser melting process is a method that can integrate the melting process mechanism with the sensor data recorded in the historical scanning trajectory, thereby achieving accurate prediction of the transient evolution of the molten pool morphology. Specifically, this method is applied to Figure 1 In the scenario shown, a high-speed coaxial infrared camera is used to coaxially monitor the melt pool temperature during selective laser melting. The monitoring system primarily includes a laser, a scanning galvanometer system, a linear polarizer, a beam splitter, a 1:1 imaging lens, and an infrared bandpass filter. This monitoring system optically aligns the camera with the laser axis to obtain a still image of the melt pool temperature field. Specifically, the following steps are involved:

[0071] Step 1: Use a temperature-calibrated high-speed infrared camera to capture the melt pool image of the selective laser melting process in a coaxial manner and at a fixed frequency, and construct a melt pool image dataset ;in, represents the melt pool image at the i-th moment, is the number of melt pool image samples; L and H are the length and height of the image respectively; in this embodiment, L and H are both 350;

[0072] In this embodiment, the image data set used is collected by an infrared camera through coaxial monitoring. The camera model is IRC 912 infrared camera with a sampling frequency of 3 kHz.

[0073] In the specific example, the BLT-A160 equipment was used to implement the additive manufacturing process. The metal powder material used in the experiment was SS316L, and nitrogen was used as the shielding gas. A serpentine scanning strategy was adopted during the construction process. The scanning spacing, layer thickness, and laser spot diameter were 120 µm, 30 µm, and 100 µm, respectively. Throughout the experiment, a nominal constant scanning speed of 700 mm / s was maintained, and parts with three different geometric shapes, namely cubes, triangular prisms, and cylinders, were constructed under different laser powers. A bandpass filter of 1350 nm to 1600 nm was selected to accurately capture the molten pool area. Figure 1 Schematic diagram of the infrared camera monitoring process used in this example.

[0074] Step 2: Process the melt pool image to extract the melt pool's length, width, area, perimeter, aspect ratio and other morphological features;

[0075] Step 2.1: Liquidus temperature of a given powder material , extract melt pool image Zhongyu The absolute value of the difference is less than the error The pixel points are extracted and the least square method is used to fit the extracted point set into a closed curve. In this embodiment, is 1723 K, is 5K.

[0076] Step 2.2: Move two first straight lines parallel to the direction of laser movement from the center of the spot to both sides. When the two first straight lines intersect the i-th closed curve When there is only one intersection point, calculate the distance between the two first straight lines to obtain the i-th closed curve Width;

[0077] Move two second straight lines perpendicular to the moving direction of the laser from the center of the spot to both sides. When there is only one intersection point, calculate the distance between the two second straight lines to obtain the i-th closed curve length;

[0078] According to the instantaneous field of view of the high-speed infrared camera, the i-th closed curve After converting the length, width, area and perimeter, the length of the molten pool is obtained. ,width ,area and perimeter ;

[0079] The aspect ratio of the melt pool is obtained by calculating the ratio between the length and width of the melt pool. .

[0080] Step 3: Momentary melt pool image To Moment melt pool image As a sample , the morphological characteristics of the molten pool at the i-th moment To Morphological characteristics of the melt pool at each moment As a sample Tags ; Morphological characteristics of the melt pool at each moment To Morphological characteristics of the melt pool at each moment As a sample Tags Corresponding historical melt pool feature sequence ; In this embodiment, q is 41 and Q is 61.

[0081] Step 4: Calculate the theoretical temperature field of the molten pool surface at the i-th moment based on material parameters, process parameters and scanning strategy ; thus obtaining the Time to The theoretical temperature field of the molten pool surface at time t, thus obtaining the label The corresponding theoretical temperature field sequence of the molten pool surface ;

[0082] Step 4.1: Take the center of the laser spot at moment i As the base point, create a rectangle with the length along the direction of laser movement at moment i and the width perpendicular to the direction of laser movement at moment i; the center of the laser spot is The distances to the front and back edges of the rectangle are set to and , the distance to the left and right edges is set to In this embodiment, 0.1 mm, 0.6mm, is 0.25 mm.

[0083] Step 4.2: Set the point set resolution to , extracted from the rectangle The coordinates of the points In this embodiment, 0.05 mm / point, is 140; among them, It is the coordinate of the j-th point at the i-th moment in the working coordinate system of the selective laser melting equipment. The working coordinate system is determined by the machine manufacturer. First, an origin is determined, and two mutually perpendicular X-axis and Y-axis are determined on the horizontal plane parallel to the substrate, as well as a Z-axis perpendicular to the substrate and upward.

[0084] Step 4.3: Calculate the label using equations (1) to (3) The corresponding theoretical temperature of the jth point at the i-th moment :

[0085] (1)

[0086] (2)

[0087] (3)

[0088] In formula (1) to formula (3), To calculate the The number of historical laser scanning points used at the theoretical temperature at the moment, in this embodiment, is 100; is the thermal diffusivity of the powder alloy, is the density of the powder material, is the initial temperature of the environment; For The specific heat capacity at the temperature, m is the coefficient of the specific heat capacity changing with temperature. In this embodiment, 5.58e-6 m 2 / s, 7400 kg / m 3 , is 300 K, is 280 J / (kg•K), m is 0.89×10 -3 ; P is the laser power, is the absorption coefficient of the powder particles, is the beam radius, is the laser beam intensity distribution factor on the XY plane. In this embodiment, is 0.6, 0.05 mm, 1 is the center of the laser spot at the i-th moment The coordinates in the XY plane, is the time integral variable; in formula (2), is the integration variable The time difference to the i-th moment; in formula (3), is the time-integrated variable The integrated parameter of the laser properties at the i-th moment is h, which is the height of the body heat source. In this embodiment, h is 0.04 mm.

[0089] Step 4.4: Calculate the label according to the process of step 4.3 The corresponding i-th moment From the first point to the The theoretical temperature of the point point to The theoretical temperature at point i is obtained, thus obtaining the theoretical temperature field of the molten pool surface at moment i. ;

[0090] Step 4.5: Follow the process from step 4.1 to step 4.4 to calculate the Time to The theoretical temperature field of the molten pool surface at the moment Time to The theoretical temperature field of the molten pool surface at the moment, thus obtaining the label The corresponding theoretical temperature field sequence of the molten pool surface .

[0091] Step 5: Calculate labels The corresponding angle between the laser moving direction and the shielding airflow direction in the clockwise direction;

[0092] Step 5.1: Calculate the label using formula (4) The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the corresponding moment i :

[0093] (4)

[0094] In formula (4), is the laser moving unit direction vector at the i-th moment, is the unit direction vector of the protective airflow at the i-th moment, is the function for calculating the clockwise angle;

[0095] Step 5.2: Follow the procedure in step 5.1 to calculate the Time to The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the moment and the Time to The angle between the laser moving direction at the moment and the protective airflow direction in the clockwise direction is obtained, so as to obtain the label Corresponding laser movement direction and protective airflow angle sequence .

[0096] Step 6: Construct a prediction network for the molten pool morphology evolution, including: theoretical temperature field feature extraction module, improved attention mechanism module, mechanism and data feature fusion module, and 、 and Process and obtain Time to Moment-by-moment prediction of melt pool morphology feature sequences .

[0097] Step 6.1: The theoretical temperature field feature extraction module is as follows Figure 2 As shown, it includes: two convolutional pooling layers and one fully connected layer. In this embodiment, the convolution kernel size is 3×3, and the number of kernels is 16 and 32;

[0098] Will Input the theoretical temperature field feature extraction module for processing and obtain the label Corresponding mechanism characteristic sequence ,in, Indicates the The mechanism characteristics of the moment.

[0099] Step 6.2: The improved attention mechanism module is as follows Figure 3 As shown, it includes: a spatiotemporal difference extraction unit, a fully connected layer and a Softmax function. In this embodiment, the number of neurons in the fully connected layer is 128;

[0100] The spatiotemporal difference extraction unit uses equations (5) to (7) to calculate the Moment and Time difference , No. Laser spot center at the moment With the Laser spot center at the moment Distance in the X direction , Y direction distance , Euclidean distance , and form the The time-space difference sequence of the historical melt pool corresponding to the moment , thus obtaining the Time to The time-space difference sequence of the historical melt pool corresponding to the moment ;

[0101] (5)

[0102] (6)

[0103] (7)

[0104] In formula (6), For the The center of the laser spot at the moment Coordinates in the XY plane.

[0105] Will Input to the fully connected layer for processing to obtain the The spatiotemporal characteristics of moments , and continue to input it into the Softmax function for processing, and get the Time to The molten pool at the moment Attention weight of the moment melting pool ; thus obtaining the label The corresponding historical melt pool Time to Attention weight at the moment ;in, Indicates the The attention weight of the melting pool at the moment; Indicates the first moment corresponding to the The spatiotemporal characteristics of the moment, Indicates the first moment corresponding to the Attention weight of melt pool features at a given moment.

[0106] Step 6.3: The mechanism and data feature fusion module are as follows Figure 4 As shown, it includes: an encoder module and a decoder module;

[0107] Step 6.3.1: The encoder is based on the neural ODE-GRU structure. Using formula (8) to formula (9), the first The encoded input hidden state at time , thus obtaining the Time to The sequence of encoded hidden states at time ;

[0108] (8)

[0109] (9)

[0110] In formula (8) to formula (9), represents the first step of encoding the neural network ordinary differential equation encoding the neural ODE output The pre-hidden state at the moment, For the The encoded input hidden state at time t, is the fully connected layer in the encoding neural ODE, To encode the parameters of the neural ODE, GRU stands for Gated Recurrent Unit; For the Encoded input features at time instant; Indicates the The mechanistic characteristics of the moment;

[0111] The first Time to Melt pool characteristic sequence at time , No. Time to The sequence of laser movement direction and protective airflow angle at each moment Hedi Time to Mechanistic characteristic sequence of moments After splicing, the fusion coding features are obtained , and input into the encoder module for processing.

[0112] Step 6.3.2: Calculate the context vector at the i-th moment using formula (10) ; thus obtaining the label The corresponding time from moment i to moment The context vector sequence at each moment ;

[0113] (10)

[0114] In formula (10), is the hidden state of the encoded input at the kth moment, represents the attention weight of the melt pool feature at the kth moment corresponding to the i-th moment;

[0115] Step 6.3.3: The decoder is a neural ODE-GRU based decoder, and uses formula (11) to formula (12) to calculate the first The decoded input hidden state at time , thus obtaining the Time to The decoded hidden state sequence at time ;

[0116] (11)

[0117] (12)

[0118] In formula (11) to formula (12), Denotes the first output of the decoded neural network ordinary differential equation decoded neural ODE The pre-hidden state at the moment, For the The decoded input hidden state at time t, To decode the fully connected layer in the neural ODE, is the parameter in the decoded neural ODE; For the Decoded input features at time t;

[0119] The context vector sequence , Laser moving direction and protective airflow angle sequence and mechanism characteristic sequence After splicing, the fusion features are obtained , and input into the decoder module to get the Time to The decoder hidden state sequence at time ;

[0120] The decoder hidden state sequence After passing through a fully connected layer, the output Time to Moment-by-moment prediction of melt pool morphology feature sequences .

[0121] Step 7: Use formula (13) to construct the loss function :

[0122] (13)

[0123] In formula (10), the first term is the mean square error of the melt pool morphology characteristics, and the second term is the mean square error of the first-order difference of the melt pool morphology characteristics, which is used to improve the prediction accuracy of the melt pool morphology mutation. is the predicted value of the melt pool morphology feature sequence, is the difference value of the melt pool morphology feature sequence, is the predicted value of the first-order difference of the melt pool morphology feature sequence, To determine the parameters of the relative importance of the loss term, in this example, is 0.8.

[0124] Step 8: Use the gradient descent method to train the molten pool morphology feature prediction model, and calculate the total loss function loss to update the model parameters. When the total loss function loss converges, the trained molten pool morphology feature prediction model is obtained, which is used to predict the future molten pool morphology feature evolution process during the selective laser melting process.

[0125] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0126] In this embodiment, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are executed.

Claims

1. A mechanism-guided data-driven method for predicting the evolution of the melt pool morphology in the selective laser melting process, characterized in that: The following steps are involved: Step 1: Use a temperature-calibrated high-speed infrared camera to capture the melt pool image of the selective laser melting process in a coaxial manner and at a fixed frequency, and construct a melt pool image dataset ;in, represents the melt pool image at the i-th moment, is the number of melt pool images; L and H are the length and height of the melt pool image respectively; Step 2: By taking the molten pool image at the i-th moment Processing is performed to extract the morphological features of the molten pool at the i-th moment , including: length ,width ,area ,perimeter , aspect ratio , thereby constructing a melt pool morphology feature dataset ; Step 3: Moment melt pool image To Moment melt pool image As a sample , the morphological characteristics of the molten pool at the i-th moment To Morphological characteristics of the melt pool at each moment As a sample Tags ; Morphological characteristics of the melt pool at each moment To Morphological characteristics of the melt pool at each moment As a historical melt pool characteristic sequence ; Step 4: Calculate the theoretical temperature field of the molten pool surface at the i-th moment based on material parameters, process parameters and scanning strategy ; thus obtaining the Time to The theoretical temperature field of the molten pool surface at time t, thus obtaining the label The corresponding theoretical temperature field sequence of the molten pool surface ; Step 5: Calculate labels The angle between the corresponding laser movement direction and the protective airflow direction in the clockwise direction ;in, represents the angle between the laser moving direction and the shielding airflow direction in the clockwise direction at the i-th moment; Step 6: Construct a prediction network for the molten pool morphology evolution, including: theoretical temperature field feature extraction module, improved attention mechanism module, mechanism and data feature fusion module, and 、 and Process it and get Time to Moment-by-moment prediction of melt pool morphology feature sequences ; Step 7: Use formula (13) to construct the total loss function : (13) In formula (10), is the predicted value of the melt pool morphology feature sequence, is the difference value of the melt pool morphology feature sequence, is the predicted value of the sequence difference of the molten pool morphology characteristics, is the parameter of importance; Step 8: Use the gradient descent method to train the molten pool morphology feature prediction network, and calculate the total loss function loss to update the model parameters. When the total loss function loss converges, the trained molten pool morphology feature prediction model is obtained, which is used to predict the future molten pool morphology feature evolution process during the selective laser melting process.

2. The mechanism-guided data-driven method for predicting the molten pool morphology evolution in the selective laser melting process according to claim 1 is characterized in that: Step 2 includes the following steps: Step 2.1: Liquidus temperature of a given powder material , extract the melt pool image at the i-th moment In, with The absolute value of the difference is less than the error The pixel points are then fitted into the i-th closed curve using the least squares method. ; Step 2.2: Move two first straight lines parallel to the direction of laser movement from the center of the spot to both sides. When the two first straight lines intersect the i-th closed curve When there is only one intersection point, calculate the distance between the two first straight lines to obtain the i-th closed curve width; Move two second straight lines perpendicular to the moving direction of the laser from the center of the spot to both sides. When there is only one intersection point, calculate the distance between the two second straight lines to obtain the i-th closed curve length; According to the instantaneous field of view of the high-speed infrared camera, the i-th closed curve After converting the length, width, area and perimeter, the length of the molten pool is obtained. ,width ,area and perimeter ; The aspect ratio of the melt pool is obtained by calculating the ratio between the length and width of the melt pool. .

3. The mechanism-guided data-driven method for predicting the molten pool morphology evolution in the selective laser melting process according to claim 2 is characterized in that: Step 4 includes the following steps: Step 4.1: Take the center of the laser spot at the i-th moment As the base point, create a rectangle with the length along the direction of laser movement at moment i and the width perpendicular to the direction of laser movement at moment i; among them, the center of the laser spot at moment i is The distances to the front and back edges of the rectangle are set to and , The distance to the left and right edges is set to ; Step 4.2: Set the point set resolution to , extracted from the rectangle The coordinates of the points ,in, is the coordinate of the j-th point at the i-th moment in the working coordinate system of the selective laser melting equipment; Step 4.3: Calculate the label using equations (1) to (3) The theoretical temperature of the jth point at the corresponding i-th moment : (1) (2) (3) In formula (1) to formula (3), To calculate the The number of historical laser scanning points used at the theoretical temperature at the moment; is the thermal diffusivity of the powder alloy, is the density of the powder material, is the initial temperature of the environment; For The specific heat capacity at the temperature, m is the coefficient of specific heat capacity changing with temperature; P is the laser power, is the absorption coefficient of the powder particles, is the beam radius, is the laser beam intensity distribution factor on the XY plane, is the center of the laser spot at the i-th moment The coordinates of the XY plane in the working coordinate system, is the time-integrated variable; is the time-integrated variable The time difference to the i-th moment; is the time-integrated variable The integrated parameter of laser properties at the i-th moment, h is the height of the body heat source; Step 4.4: Calculate the label according to the process of step 4.3 Corresponding Middle point to points and point to The theoretical temperature at point i is obtained, thus obtaining the theoretical temperature field of the molten pool surface at moment i. .

4. The mechanism-guided data-driven method for predicting the evolution of the molten pool morphology in the selective laser melting process according to claim 3 is characterized in that: Step 5 includes the following steps: Step 5.1: Calculate the label using formula (4) The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the corresponding moment i : (4) In formula (4), is the laser moving unit direction vector at the i-th moment, is the unit direction vector of the protective airflow at the i-th moment, is the function for calculating the clockwise angle; Step 5.2: Follow the procedure in step 5.1 to calculate the Time to The angle between the laser moving direction and the protective airflow direction in the clockwise direction at the moment and the Time to The angle between the laser moving direction at the moment and the protective airflow direction in the clockwise direction is obtained, so as to obtain the label Corresponding laser movement direction and protective airflow angle sequence .

5. The mechanism-guided data-driven method for predicting the molten pool morphology evolution in the selective laser melting process according to claim 4 is characterized in that: Step 6 includes the following steps: Step 6.1: The theoretical temperature field feature extraction module includes: two convolutional pooling layers and one fully connected layer; Will Input the theoretical temperature field feature extraction module for processing and obtain the label Corresponding mechanism characteristic sequence ,in, Indicates the The mechanistic characteristics of the moment; Step 6.2: The improved attention mechanism module includes: a spatiotemporal difference extraction unit, a fully connected layer, and a Softmax function; The spatiotemporal difference extraction unit uses equations (5) to (7) to calculate the Moment and Time difference , No. Laser spot center at the moment With the Laser spot center at the moment Distance in the X direction , Y direction distance , Euclidean distance , and form the The time-space difference sequence of the historical melt pool corresponding to the moment , thus obtaining the Time to The time-space difference sequence of the historical melt pool corresponding to the moment ; (5) (6) (7) In formula (6), For the The center of the laser spot at the moment Coordinates in the XY plane; Will Input to the fully connected layer for processing to obtain the Spatiotemporal characteristics of moments , and continue to input it into the Softmax function for processing, and get the Time to The molten pool at the moment Attention weight of the moment melt pool ; thus obtaining the label The corresponding historical melt pool Time to Attention weight at the moment ;in, Indicates the The attention weight of the melting pool at the moment; Indicates the first moment corresponding to the The spatiotemporal characteristics of the moment, Indicates the first moment corresponding to the Attention weight of the melt pool feature at each moment; Step 6.3: The mechanism and data feature fusion module includes: an encoder module and a decoder module; Step 6.3.1: The encoder module is an encoder based on the encoding neural ODE-GRU structure, and uses formulas (8) to (9) to calculate the first The encoded input hidden state at time , thus obtaining the Time to The sequence of encoded hidden states at time ; (8) (9) In formula (8) to formula (9), represents the first The pre-hidden state at the moment, For the The encoded input hidden state at time t, is the fully connected layer in the encoding neural ODE, To encode the parameters of the neural ODE, GRU stands for Gated Recurrent Unit; For the Encoded input features at time instant; Indicates the The mechanistic characteristics of the moment; Step 6.3.2: Calculate the context vector at the i-th moment using formula (10) ; thus obtaining the label The corresponding time from moment i to moment The context vector sequence at each moment ; (10) In formula (10), is the hidden state of the encoded input at the kth moment, represents the attention weight of the melt pool feature at the kth moment corresponding to the i-th moment; Step 6.3.3: The decoder is a decoder based on the decoding neural ODE-GRU, and uses formula (11) to formula (12) to calculate the first The decoded input hidden state at time , thus obtaining the Time to The decoded hidden state sequence at time ; (11) (12) In formula (11) to formula (12), represents the first output of the decoded neural ODE The pre-hidden state at the moment, For the The decoded input hidden state at time t, To decode the fully connected layer in the neural ODE, is the parameter in the decoded neural ODE; For the Decoded input features at time t; Will After passing through a fully connected layer, the output Time to Moment-by-moment prediction of melt pool morphology feature sequences .

6. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the method for predicting the evolution of the molten pool morphology in the selective laser melting process as described in any one of claims 1 to 5, and the processor is configured to execute the program stored in the memory.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the morphology evolution of the molten pool in the selective laser melting process according to any one of claims 1 to 5 are executed.

Images