Flux core cold metal transition electric arc additive reinforced particle monitoring method based on time sequence characteristics

Abstract

The invention discloses a flux-cored cold metal transition electric arc additive reinforced particle monitoring method based on time sequence characteristics, which comprises the following steps of: placing a particle reinforced metal matrix composite material into a cold metal transition electric arc additive material system for cold metal transition electric arc additive material processing; the image acquisition unit acquires optical images in the additive machining process in real time to construct a time sequence image data set; performing image segmentation on each optical image of the time sequence image data set to obtain wire, molten drop and molten pool areas in the images; performing feature extraction on the wire material, the molten drop and the molten pool area; and the extracted features are recombined, and the time sequence features are input into a regression prediction model based on a bidirectional long-short-term memory network and a time attention mechanism, so that the content prediction of the reinforced particles in the metal-based composite material formed by the additive material is obtained. The method is suitable for reinforced particle content monitoring of additive manufacturing metal matrix composite technologies with periodic properties such as pulse composite cold metal transition additive and alternating current pulse cold metal transition additive.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing quality monitoring, and in particular to a method for monitoring additive manufacturing enhanced particles based on a time sequence characteristic flux-cored cold metal transition arc. Background Art

[0002] Particle Reinforced Metal Matrix Composites (PRMMCs) are advanced materials that are manufactured by adding ceramic particles or other hard particles to a metal matrix to improve the strength, hardness, wear resistance and high-temperature performance of the material. The use of additive manufacturing methods to manufacture particle reinforced metal matrix composites has attracted great attention because it can enhance the mechanical, thermal and tribological properties of components by reinforcing particles in specific areas. Flux-cored cold metal transfer arc additive manufacturing, as an important branch of wire arc additive manufacturing, can achieve a composite manufacturing mode of metal wire and non-metallic powder by changing the composition of the wire. Given the reliability and repeatability issues of particle reinforced metal matrix composites manufactured by flux-cored cold metal transfer arc additive manufacturing, the development of reinforcement particle monitoring methods and their integration with flux-cored cold metal transfer arc additive technology has become a priority task.

[0003] The main challenge in manufacturing particle reinforced metal matrix composites is to obtain a high content of reinforcing particles in their original form, but to avoid agglomeration and excessive dissolution of the reinforcing particles. The volume fraction of reinforcing particles in the formed cross section is one of the core indicators affecting the performance of composite materials. Particle content prediction is the basis for judging the quality of reinforced metal matrix composites and accurately controlling the additive manufacturing process, such as the attached Figure 2 As shown in the figure, the cross section of the additively manufactured particulates is shown, where the particle content is the ratio of the total particle area to the cross-sectional area of the particulates. Ideally, the reinforcing particles inside the flux-cored wire can be wrapped by the molten droplets and follow the molten droplets into the molten pool. However, before entering the molten pool, some reinforcing particles are blown away by the shielding gas along with the molten droplets without having the opportunity to encounter the molten pool due to the excessive and uneven electromagnetic force at the initial arc stage associated with the tubular electrode. The principle is shown in the attached figure. Figure 1 As shown. Meanwhile, particles within the open wire tube sheath may fall off due to gravity and be blown away by the flowing shielding gas and electromagnetic force before the next droplet forms. Although the wire tube sheath can prevent the reinforcing particles, which are highly sensitive to thermal degradation, from direct contact with the high-temperature arc, it cannot prevent the particles from dissolving within the high-temperature molten pool. Therefore, after entering the molten pool, the reinforcing particles may also dissolve due to the high temperature, which is highly related to the existence time of the molten pool.

[0004] The above physical processes demonstrate that, unlike other additive methods that prioritize stable processing, cold metal transfer arc additive manufacturing, which can be divided into multiple stages, including droplet growth, droplet transfer, and short circuiting, involves a complex, dynamic series of cyclical processes, including arc initiation, wire melting, droplet transfer, arc cessation, and electrode expansion and contraction. These processes may affect the content of reinforcement particles in the resulting metal matrix composite. Existing monitoring methods applicable to other additive technologies that prioritize stable processing have difficulty predicting the content of reinforcement particles. Summary of the Invention

[0005] 1. Technical problems to be solved: In response to the above technical problems, the present invention provides a method for monitoring enhanced particles in flux-cored cold metal transition arc additive manufacturing based on time series characteristics, which solves the problems existing in the current methods for monitoring enhanced particles in flux-cored cold metal transition arc additive manufacturing.

[0006] 2. Technical solution: The method for monitoring particles of flux-cored cold metal transfer arc additive material enhancement based on time sequence characteristics is characterized by comprising the following steps: Step 1: Build a monitoring system; the monitoring system includes a cold metal transfer arc additive system and an image acquisition unit; Step 2: Image acquisition: placing the particle-reinforced metal matrix composite material into a cold metal transfer arc additive system for cold metal transfer arc additive processing, and using the optical image acquisition module of the image acquisition unit to acquire optical images of the additive processing process in real time to construct a time-series image dataset; Step 3: Time series feature extraction: perform image segmentation on each optical image in the collected time series image dataset to segment the wire, droplet, and molten pool regions in the image; perform feature extraction on the extracted wire, droplet, and molten pool regions; and reorganize the extracted features to obtain N×T dimensional features, where N is the sum of the number of wire features N1, the number of droplet features N2, and the number of molten pool features N3, and T is the number of consecutive frames; Step 4: Input the reorganized N×T dimensional time series features into the regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism to obtain the prediction of the reinforcing particle content in the additively formed metal matrix composite material.

[0007] Furthermore, the cold metal transfer arc additive system includes an arc generating device, a wire feeding device, a welding gun and its displacement device, and a particle-reinforced metal matrix composite material; the wire feeding device provides the welding gun with a tubular flux-cored welding wire with reinforced particles, and the welding gun, driven by its displacement device, welds the particle-reinforced metal matrix composite material through a wire feeding / withdrawal movement, and the arc generating device controls the current and voltage to enable the welding gun to generate arcs of different frequencies and periods; the image acquisition unit includes an optical image acquisition module and a laser light source module; the image acquisition module acquires images of the welding process through a camera, and the laser light source module is used to prevent the image acquisition module from being overexposed due to the arc.

[0008] Furthermore, the reinforcing particles in the tubular flux-cored welding wire with reinforcing particles include tungsten carbide or titanium carbide; and the sealing metal tube in the tubular flux-cored welding wire with reinforcing particles includes an iron-based, cobalt-based, nickel-based or aluminum-based material.

[0009] Furthermore, in the tubular flux-cored welding wire with reinforced particles, 2% to 10% of nickel powder and ferrosilicon are mixed with the reinforced particles, and the mixed powder is filled into a sealed metal tube at a filling rate of 40%.

[0010] Furthermore, the optical image acquisition module moves synchronously with the welding gun to achieve synchronous acquisition of images of the additive process, which include information of the wire, molten droplet and molten pool area; the acquisition cycle of the optical image acquisition module is: when the wire feeding / retraction movement frequency in cold metal transition additive manufacturing is k, the acquisition frequency of the optical image acquisition module is k / p, where p is an integer greater than or equal to 3.

[0011] Furthermore, in step three: the image segmentation method is a binary threshold segmentation or an image segmentation algorithm; in the N×T dimensional features, N is 13, which is specifically: the wire features include the cross-section of the wire and the length of the wire; the molten droplet features include the molten droplet width, molten droplet length, molten droplet area, molten droplet circumference and molten droplet inclination angle; the molten pool features include the molten pool width, molten pool length, molten pool area, molten pool circumference, the horizontal coordinate of the molten pool center of mass and the vertical coordinate of the molten pool center of mass; T is np, where n is the number of image sampling cycles, n≥5.

[0012] Furthermore, step three also includes: if the image segmentation result lacks a molten droplet area, the molten droplet feature extracted during the feature extraction process is 0; if the image segmentation result has multiple molten droplet areas, the molten droplet feature extracted during the feature extraction process is the maximum value of the multiple molten droplet features.

[0013] Furthermore, the regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism in step four; the model outputs the content of enhanced particles in single-channel forming through the N×T dimensional temporal features of the input image; the specific model includes an input layer, a temporal feature encoding module, a dynamic weight prediction module and a fully connected network output layer; wherein the temporal feature encoding module extracts the forward propagation state and the backward propagation state of the input N×T dimensional temporal features through the forward LSTM unit and the backward LSTM unit, and then performs cascade splicing to generate a feature vector H that integrates the information before and after the temporal sequence; the dynamic weight prediction module adaptively weights the encoded temporal state through the temporal attention mechanism, calculates the weight coefficient of the state vector of each time step based on the fully connected network and the Softmax function, and obtains the fused weighted feature by weighted summation of the time step state vector according to the weight coefficient, and finally inputs the fused weighted feature into the fully connected network to output the predicted value of the content of enhanced particles in the metal matrix composite material in additive manufacturing.

[0014] 3.Beneficial effects: (1) Existing methods for predicting additive manufacturing quality by directly inputting a single melt pool image are mainly targeted at additive technologies that expect stable processing processes, such as laser cladding, plasma cladding, and powder bed laser melting. Since cold metal transfer arc additive manufacturing has a series of periodically changing dynamic physical processes, such as arc starting, wire melting, droplet transfer, arc stopping, and electrode expansion and contraction, existing technologies are not applicable to cold metal transfer arc additive manufacturing. In the method disclosed in this method, the enhanced particle monitoring method for flux-cored cold metal transfer arc additive manufacturing based on time series features is used to segment each optical image of the collected time series image data set to segment the wire, droplet, and melt pool areas in the image; and feature extraction is performed on the extracted wire, droplet, and melt pool areas; and the extracted features are recombined to obtain N×T dimensional features. This can realize the time series feature extraction of multiple stages such as droplet growth, droplet transfer, and short circuit in cold metal transfer arc additive manufacturing, and can more accurately judge the additive manufacturing quality.

[0015] (2) In the method for monitoring the enhanced particles of the additive manufacturing process of the cold metal transition arc with a core based on the time series characteristics disclosed in this method, the enhanced particle content is used as the quality indicator of the additively formed metal matrix composite material. A regression prediction model based on a bidirectional long short-term memory network and a temporal attention mechanism is adopted to effectively capture the long-term correlation of the multi-stage time series characteristics and the dynamic changes of the key time nodes in the cold metal transition arc additive manufacturing process. The temporal attention mechanism accurately focuses on the data of the key process feature time periods such as the molten pool fluctuation and the molten droplet transient through adaptive weight distribution, thereby solving the problem of insufficient sensitivity of the traditional single time series model to local features.

[0016] (3) This monitoring only requires adding an image acquisition unit, a time series feature extraction unit, and an enhanced particle prediction unit to the original cold metal transition arc additive system to achieve real-time prediction of the enhanced particle content. The method has high integration and fast processing speed. It is suitable for monitoring the enhanced particle content of other additive manufacturing metal-based composite materials with periodic properties, such as pulse composite cold metal transition additive and AC pulse cold metal transition additive, and is easy to implement and promote. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a diagram showing the monitoring principle of the cold metal transition additive process involved in this method; Figure 2 is a schematic diagram of a forming cross section during the additive manufacturing process involved in the present method; Figure 3 Schematic diagram of the hardware device of the monitoring system of the present invention; Figure 4 is an overall flow chart of the monitoring method of the present invention; Figure 5 is the result of segmentation of the optical image collected in the specific embodiment; Figure 6 These are the data and images collected during the implementation process of the specific embodiment. DETAILED DESCRIPTION

[0018] The present invention will be described in detail below with reference to the accompanying drawings.

[0019] As attached Figure 4 As shown, the method for monitoring particles of flux-cored cold metal transition arc additive material enhancement based on time sequence characteristics is characterized by comprising the following steps: Step 1: Build a monitoring system; the monitoring system includes a cold metal transfer arc additive system and an image acquisition unit.

[0020] Step 2: Capture images; place the particle-reinforced metal matrix composite material into a cold metal transfer arc additive system for cold metal transfer arc additive processing, and use the optical image acquisition module of the image acquisition unit to capture optical images of the additive processing process in real time to construct a time-series image data set.

[0021] Step 3: Time series feature extraction; perform image segmentation on each optical image of the collected time series image data set to segment the wire, molten droplet and molten pool areas in the image; perform feature extraction on the extracted wire, molten droplet and molten pool areas; and reorganize the extracted features to obtain N×T dimensional features, where N is the sum of the number of wire features N1, the number of molten droplet features N2 and the number of molten pool features N3, and T is the number of consecutive frames.

[0022] Step 4: Input the reorganized N×T dimensional time series features into the regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism to obtain the prediction of the reinforcing particle content in the additively formed metal matrix composite material.

[0023] Furthermore, the cold metal transfer arc additive system includes an arc generating device, a wire feeding device, a welding gun and its displacement device, and a particle-reinforced metal matrix composite material; the wire feeding device provides the welding gun with a tubular flux-cored welding wire with reinforced particles, and the welding gun, driven by its displacement device, welds the particle-reinforced metal matrix composite material through a wire feeding / withdrawal movement, and the arc generating device controls the current and voltage to enable the welding gun to generate arcs of different frequencies and periods; the image acquisition unit includes an optical image acquisition module and a laser light source module; the image acquisition module acquires images of the welding process through a camera, and the laser light source module is used to prevent the image acquisition module from being overexposed due to the arc.

[0024] Furthermore, the reinforcing particles in the tubular flux-cored welding wire with reinforcing particles include tungsten carbide or titanium carbide; and the sealing metal tube in the tubular flux-cored welding wire with reinforcing particles includes an iron-based, cobalt-based, nickel-based or aluminum-based material.

[0025] Furthermore, in the tubular flux-cored welding wire with reinforced particles, 2% to 10% of nickel powder and ferrosilicon are mixed with the reinforced particles, and the mixed powder is filled into a sealed metal tube at a filling rate of 40%.

[0026] Furthermore, the optical image acquisition module moves synchronously with the welding gun to achieve synchronous acquisition of images of the additive process, which include information of the wire, molten droplet and molten pool area; the acquisition cycle of the optical image acquisition module is: when the wire feeding / retraction movement frequency in cold metal transition additive manufacturing is k, the acquisition frequency of the optical image acquisition module is k / p, where p is an integer greater than or equal to 3.

[0027] Furthermore, in step three: the image segmentation method is a binary threshold segmentation or an image segmentation algorithm; in the N×T dimensional features, N is 13, which is specifically: the wire features include the cross-section of the wire and the length of the wire; the molten droplet features include the molten droplet width, molten droplet length, molten droplet area, molten droplet circumference and molten droplet inclination angle; the molten pool features include the molten pool width, molten pool length, molten pool area, molten pool circumference, the horizontal coordinate of the molten pool center of mass and the vertical coordinate of the molten pool center of mass; T is np, where n is the number of image sampling cycles, n≥5.

[0028] Furthermore, step three also includes: if the image segmentation result lacks a molten droplet area, the molten droplet feature extracted during the feature extraction process is 0; if the image segmentation result has multiple molten droplet areas, the molten droplet feature extracted during the feature extraction process is the maximum value of the multiple molten droplet features.

[0029] Furthermore, the regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism in step four; the model outputs the content of enhanced particles in single-channel forming through the N×T dimensional temporal features of the input image; the specific model includes an input layer, a temporal feature encoding module, a dynamic weight prediction module and a fully connected network output layer; wherein the temporal feature encoding module extracts the forward propagation state and the backward propagation state of the input N×T dimensional temporal features through the forward LSTM unit and the backward LSTM unit, and then performs cascade splicing to generate a feature vector H that integrates the information before and after the temporal sequence; the dynamic weight prediction module adaptively weights the encoded temporal state through the temporal attention mechanism, calculates the weight coefficient of the state vector of each time step based on the fully connected network and the Softmax function, and obtains the fused weighted feature by weighted summation of the time step state vector according to the weight coefficient, and finally inputs the fused weighted feature into the fully connected network to output the predicted value of the content of enhanced particles in the metal matrix composite material in additive manufacturing. Specific embodiment: In this embodiment, a flux-cored cold metal transfer arc additive manufacturing method of a tungsten carbide particle reinforced iron-based metal-based composite is performed on the surface of a 316L steel substrate, and the tungsten carbide particle content of the formed tungsten carbide particle reinforced 316L metal-based composite is predicted.

[0031] As attached Figure 3 As shown, this embodiment uses the ArcMan S1 Adv arc additive manufacturing system produced by Nanjing Enigma Industrial Automation Technology Co., Ltd. This system includes an arc generator 1, an arc generator 2, a wire feeder 2, a shielding gas 3, a displacement device 4, and a welding gun 5. An industrial optical camera 7 and a computer 8 are added to this system. The industrial optical camera 7 has an additional built-in laser light source and is integrated with the welding gun 5 and moves with it. When the trigger of the displacement device 4 is set to 1, the industrial optical camera 7 is simultaneously triggered to begin capturing optical images.

[0032] As attached Figure 4 As shown in the figure, it is the overall block diagram of the monitoring system. The monitoring system includes a cold metal transfer arc additive system, an image acquisition unit, a timing feature extraction unit, and an enhanced particle prediction unit; wherein the timing feature extraction unit includes an optical image segmentation module, a feature extraction module, and a timing feature reconstruction module.

[0033] The specific monitoring process includes: S1, a tubular flux-cored wire with tungsten carbide reinforcement particles was selected. The tubular flux-cored wire was filled into a sealed H08 steel metal tube at a 40% filling ratio with a mixture of 8% nickel powder, 2% ferrosilicon, and 90% tungsten carbide particles. Cold metal transfer arc additive manufacturing (CMT) was performed on a 316L stainless steel substrate. The wire feed speed and traverse speed were set to 3.9 m / min and 0.5 m / min, respectively. The feed / retraction motion frequency of the CMT arc additive manufacturing was k = 20 ms. The optical image acquisition module of the image acquisition unit captured optical images in real time, with the acquisition frequency set to k / p = 5 ms (p = 4).

[0034] S2, the arc additive manufacturing system is started, and its operation and monitoring process are as shown in the attached Figure 6 As shown, the cold metal transfer arc additive manufacturing cycle can be divided into three phases: droplet growth, droplet transfer, and short circuit. During the droplet growth phase, the current rapidly increases, igniting a bell-shaped arc between the electrode and substrate. This heats and melts the wire tip, forming a droplet (point 2 in the left figure). As the arc burns, the droplet gradually grows (point 3 in the left figure). During the droplet transfer phase, the current rapidly decreases (point 4) to prevent the droplet from spherical transfer, while the wire advances toward the molten pool (point 5 in the left figure). The droplet's root contracts and elongates under the influence of gravity, electromagnetic contraction forces, and other factors. The wire propels the droplet until it contacts the molten pool (point 6 in the left figure), at which point the short circuit phase begins. The current and voltage immediately decrease to near zero (point 7 in the left figure), and the wire retracts, separating the droplet and transferring it into the molten pool (point 8 in the left figure). The wire is then retracted to complete the metal transfer phase (point 9 in the left figure) and prepare for the next droplet transfer. During this step, the image acquisition unit captures four time-series images at a 5 ms frequency, or four images per cycle.

[0035] S3, the optical image segmentation module in the time series feature extraction unit performs image segmentation on the collected optical image through the pre-trained YoloV8 model to extract the wire, droplet and molten pool areas respectively, such as Figure 5 As shown in the figure, the feature extraction module extracts features from the segmented wire, droplet, and melt pool. The 13 extracted features include wire length, wire cross-sectional area, droplet width, droplet length, droplet area, droplet perimeter, droplet inclination angle, melt pool width, melt pool length, melt pool area, melt pool perimeter, and the horizontal and vertical coordinates of the melt pool center of mass. The temporal feature reconstruction module reconstructs the features extracted from 20 consecutive frames (T = np, n = 5) into a 13 × 20 dimension.

[0036] S4, the 13×20 dimensional reorganized time series features obtained by the time series feature reorganization module are input into the long short-term memory network regression prediction model to predict the content of reinforcing particles in the metal matrix composite material after additive manufacturing.

[0037] Although the present invention has been disclosed above in terms of preferred embodiments, they are not intended to limit the present invention. Anyone skilled in the art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined by the claims of this application.

Claims

1. A method for monitoring particles in flux-cored cold metal transfer arc additive manufacturing based on time series characteristics, characterized by: The following steps are involved: Step 1: Build a monitoring system; the monitoring system includes a cold metal transfer arc additive system and an image acquisition unit; Step 2: Image acquisition: placing the particle-reinforced metal matrix composite material into a cold metal transfer arc additive system for cold metal transfer arc additive processing, and using the optical image acquisition module of the image acquisition unit to acquire optical images of the additive processing process in real time to construct a time-series image dataset; Step 3: Time series feature extraction; Image segmentation is performed on each optical image in the collected time-series image dataset to segment the wire, droplet, and melt pool regions in the image. Feature extraction is performed on the extracted wire, droplet, and melt pool regions. The extracted features are then recombined to obtain N×T dimensional features, where N is the sum of the number of wire features N1, the number of droplet features N2, and the number of melt pool features N3, and T is the number of consecutive frames. Step 4: Input the reorganized N×T dimensional time series features into the regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism to obtain the prediction of the reinforcing particle content in the additively formed metal matrix composite material.

2. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 1, characterized in that: The cold metal transfer arc additive system includes an arc generating device, a wire feeding device, a welding gun and its displacement device, and a particle-reinforced metal matrix composite material; the wire feeding device provides the welding gun with a tubular flux-cored welding wire with reinforced particles, and the welding gun, driven by its displacement device, welds the particle-reinforced metal matrix composite material through a wire feeding / withdrawal motion. The arc generating device controls the current and voltage to enable the welding gun to generate arcs of different frequencies and periods; the image acquisition unit includes an optical image acquisition module and a laser light source module; the image acquisition module acquires images of the welding process through a camera, and the laser light source module is used to prevent the image acquisition module from being overexposed due to the arc.

3. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 2, characterized in that: The reinforcing particles in the tubular flux-cored welding wire with reinforcing particles include tungsten carbide or titanium carbide; the sealing metal tube in the tubular flux-cored welding wire with reinforcing particles includes an iron-based, cobalt-based, nickel-based or aluminum-based material.

4. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 2, characterized in that: In the tubular flux-cored welding wire with reinforced particles, 2% to 10% of nickel powder and ferrosilicon are mixed with the reinforced particles, and the mixed powder is filled into a sealed metal tube at a filling rate of 40%.

5. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 1, characterized in that: The optical image acquisition module moves synchronously with the welding gun to achieve synchronous acquisition of images of the additive process, which include information about the wire, molten droplet, and molten pool area. The acquisition cycle of the optical image acquisition module is: when the wire feeding / retraction movement frequency in cold metal transition additive manufacturing is k, the acquisition frequency of the optical image acquisition module is k / p, where p is an integer greater than or equal to 3.

6. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 4, characterized in that: In step 3, the image segmentation method is a binary threshold segmentation or an image segmentation algorithm; in the N×T dimensional features, N is 13, which are specifically: the wire features include the cross section and length of the wire; the droplet features include the droplet width, droplet length, droplet area, droplet perimeter, and droplet inclination angle; the molten pool features include the molten pool width, molten pool length, molten pool area, molten pool perimeter, molten pool centroid abscissa, and molten pool centroid ordinate; T is np, where n is the number of image sampling periods, n≥5.

7. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 6, characterized in that: Step three also includes: if the image segmentation result lacks a molten droplet area, the molten droplet feature extracted during the feature extraction process is 0; if the image segmentation result has multiple molten droplet areas, the molten droplet feature extracted during the feature extraction process is the maximum value of the multiple molten droplet features.

8. The method for monitoring particles in a flux-cored cold metal transfer arc additive process based on time series characteristics according to claim 1, characterized in that: The regression prediction model based on the bidirectional long short-term memory network and the temporal attention mechanism in step 4; the model outputs the content of enhanced particles in single-pass forming through the N×T dimensional temporal features of the input image; the specific model includes an input layer, a temporal feature encoding module, a dynamic weight prediction module and a fully connected network output layer; the temporal feature encoding module extracts the forward propagation state and the backward propagation state of the input N×T dimensional temporal features through the forward LSTM unit and the backward LSTM unit, and then performs cascade splicing to generate a feature vector H that integrates the information before and after the temporal sequence; the dynamic weight prediction module adaptively weights the encoded temporal state through the temporal attention mechanism, calculates the weight coefficient of the state vector of each time step based on the fully connected network and the Softmax function, and obtains the fused weighted feature by weighted summation of the time step state vector according to the weight coefficient. Finally, the fused weighted feature is input into the fully connected network to output the predicted value of the content of enhanced particles in the metal matrix composite material in additive manufacturing.

Images