LPBF hole defect prevention and control method and system based on bimodal fusion digital twinning
By using dual-modal fusion digital twin technology, we can achieve proactive prediction and process intervention of LPBF hole defects. This solves the problems of lagging quality control, fragmented prediction models, and difficulties in human-computer interaction in LPBF technology, improves process optimization efficiency and model credibility, and reduces defect incidence and optimization costs.
Patent Information
- Application Number
- CN202511798306.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-03
AI Technical Summary
LPBF technology suffers from problems in printing and process control, such as lagging quality control, fragmented predictive model functions, difficulties in human-machine interaction, and low efficiency in process optimization, making it difficult to predict and correct defects.
A dual-modal fusion digital twin approach is adopted, which acquires multi-source heterogeneous data through online status management and data acquisition modules. By utilizing the multi-source data perception and fusion module, dual-modal collaborative adaptive model, anomaly localization model and simulation module built into the digital twin system, proactive prediction and process intervention of hole defects are achieved. Combined with Score-CAM technology, visualization heat map generation and virtual entity simulation are performed to construct an adaptive intelligent closed-loop control system.
It enables proactive prediction and prevention of LPBF hole defects, reduces the defect incidence rate, improves model credibility and executability, shortens process optimization cycle, reduces costs and improves on-site response speed.
Smart Images
Figure CN121598787A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of additive manufacturing, and in particular to a method and system for preventing and controlling LPBF hole defects based on dual-modal fusion digital twin. Background Technology
[0002] Laser Powder Bed Fusion (LPBF), as one of the core technologies of additive manufacturing, has significant advantages such as manufacturing complex structures and saving material costs, and is widely used in high-value-added fields such as aerospace and medical devices. However, LPBF technology involves complex dynamic processes such as rapid heating, molten pool formation and solidification, which leads to several challenges in printing and process control: First, the quality control system lags behind the defect formation process. Existing systems mostly adopt a post-incident detection mode, relying on offline methods such as X-ray computed tomography (XCT) or metallographic analysis. By the time process deviations are detected, defects have already formed irreversibly. This passive correction mode cannot proactively predict and intervene in highly dynamic and transient defects such as keyholes.
[0003] Second, the predictive models suffer from fragmented functionality and insufficient reliability. Existing monitoring systems are mostly single-modal systems, focusing either solely on predicting printouts or independently optimizing parameters, lacking the ability to effectively integrate multi-source heterogeneous data (equipment commands, sensor data, process parameters). A fidelity gap exists between simulation results and physical entity operations, leading to poor decision-making feasibility. Furthermore, the model prediction process is a "black box," lacking interpretability, making it difficult for operators to trust the model's decisions.
[0004] Third, there is a cognitive gap in human-computer interaction. The traditional two-dimensional screen's data abstraction display method makes it difficult to intuitively connect digital information with the physical manufacturing site. Operators find it difficult to quickly understand the causes of defects and make accurate judgments, which weakens the system's on-site responsiveness.
[0005] Process optimization is inefficient. Traditional trial-and-error methods or offline calibration of single parameters are inefficient under high-dimensional and multi-constraint conditions, making it difficult to obtain the globally optimal combination of process parameters, resulting in high parameter optimization costs and lengthy cycles. Summary of the Invention
[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an LPBF hole defect prevention and control method and system based on dual-modal fusion digital twin, so as to realize the fundamental transformation of manufacturing quality from passive correction to active predictive prevention, and fundamentally improve the stability of LPBF process and manufacturing quality.
[0007] To achieve the above objectives, the technical solution provided by this invention is as follows: A method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins includes: The status is determined by the online status management and data acquisition module, and multi-source heterogeneous data is obtained based on the status result. The multi-source heterogeneous data includes static parameters and dynamic thermal characteristic parameters. The static parameters are composed of process parameters, model geometric parameters and kinematic parameters, and the dynamic thermal characteristic parameters are high-frequency time-series thermal imaging data of the molten pool region. The multi-source heterogeneous data is input into the digital twin system. The multi-source data sensing and fusion module built into the digital twin system senses the received multi-source heterogeneous data. If the sensed multi-source heterogeneous data are all static parameters, the static parameters are directly input into the dual-modal cooperative adaptive model built into the digital twin system. If the sensed multi-source heterogeneous data include static parameters and dynamic thermal characteristic parameters, the static parameters and dynamic thermal characteristic parameters are fused and then input into the dual-modal cooperative adaptive model. The dual-modal collaborative adaptive model adaptively selects the prediction path based on the validity of the dynamic thermal characteristic parameters in the input multi-source heterogeneous data: when the dynamic thermal characteristic parameters are valid, the online prediction path that integrates static parameters and dynamic thermal characteristic parameters is activated to predict porosity and obtain the first porosity value; when the dynamic thermal characteristic parameters are missing, the model automatically switches to the offline simulation prediction path that relies only on static parameters to predict porosity and obtain the second porosity value. If the dual-modal collaborative adaptive model outputs the first porosity value, i.e., in online state management mode, the online state management and data acquisition module generates a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, and feeds it back to the laser powder bed melting equipment to realize active prevention and process intervention of pore defects; if the dual-modal collaborative adaptive model outputs the second porosity value, i.e., in offline simulation mode, closed-loop control and cyclic optimization of process parameters are performed.
[0008] Furthermore, both the online status management mode and the offline simulation mode include inputting dynamic thermal characteristic parameters into the anomaly localization model built into the digital twin system, using Score-CAM technology to generate a two-dimensional visual thermal map, and then mapping it to three-dimensional space to complete the stereoscopic projection of the pore defects; at the same time, based on geometric and kinematic parameters, the simulation module built into the digital twin system performs high-fidelity process simulation to generate a complete virtual entity of the printed part. In the online status management mode, the input is the dynamic thermal characteristic parameters acquired by the online status management and data acquisition module, while in the offline simulation mode, the input is the dynamic thermal characteristic parameters obtained by simulation.
[0009] Furthermore, the dynamic thermal characteristic parameters obtained from the simulation are simulated using a dynamic-static correlation model built into the digital twin system. This dynamic-static correlation model maps the relationship between static parameters and dynamic thermal characteristic parameters.
[0010] Furthermore, both the collected dynamic thermal characteristic parameters and the simulated dynamic thermal characteristic parameters include molten pool geometric morphology characteristic parameters and time-temperature related characteristic parameters; Among them, the geometric features of the molten pool include: molten pool area, molten pool length, molten pool width, molten pool eccentricity, molten pool perimeter, average molten pool temperature, and maximum molten pool temperature; the time-temperature related feature parameters include: the duration of a single pixel temperature exceeding 1200K, the duration of a temperature exceeding 1680K, and the duration of a temperature exceeding 2400K.
[0011] Furthermore, in offline simulation mode, the output of the dual-modal collaborative adaptive model is the second porosity value, the three-dimensional projection of pore defects, and the complete virtual entity. The process parameters are repeatedly adjusted and iteratively verified through the offline simulation and process optimization module built into the digital twin system, thereby completing the process optimization.
[0012] Furthermore, the online status management and data acquisition module is interconnected with the laser powder bed melting equipment to acquire multi-source heterogeneous data obtained from the control card and infrared thermal imager in the laser powder bed melting equipment in real time. The control card reads parameters such as laser power, scanning speed, scanning path coordinates, and powder layer thickness. Infrared thermal imagers capture high-frequency time-series thermal imaging data.
[0013] Furthermore, in both online status management mode and offline simulation mode, the digital twin system ensures a high degree of coordination between motion simulation and the physical printing process by parsing G-code, thereby reproducing the generation process of the printed object in virtual space.
[0014] Furthermore, to achieve the above objectives, the present invention also provides an LPBF hole defect prevention and control system based on dual-modal fusion digital twin, used to implement the above-mentioned LPBF hole defect prevention and control method based on dual-modal fusion digital twin, including an online status management and data acquisition module and a digital twin system; The digital twin system incorporates a multi-source data sensing and fusion module, a dual-modal collaborative adaptive model, an anomaly localization model, a dynamic-static correlation model, a simulation module, and an offline simulation and process optimization module. in, The online status management and data acquisition module is used to determine whether the equipment hardware is in an online or offline state and to acquire multi-source heterogeneous data based on the status result. It also includes generating a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, and feeding it back to the laser powder bed melting equipment. The multi-source data sensing and fusion module senses and fuses the input multi-source heterogeneous data; The dual-modal collaborative adaptive model adaptively selects the prediction path based on the effectiveness of dynamic thermal characteristic parameters in multi-source heterogeneous data, and predicts the corresponding porosity value through the selected prediction path. The anomaly localization model uses Score-CAM technology to generate a two-dimensional visualization heat map, which is then mapped to three-dimensional space to complete the stereoscopic projection of pore defects. The simulation module performs high-fidelity process simulation to generate a complete virtual entity of the printed part. The dynamic-static correlation model is used to obtain the dynamic thermal characteristic parameters of the simulation. The offline simulation and process optimization module, combined with the dual-modal collaborative adaptive model, outputs a second porosity value, a three-dimensional projection of pore defects, and a complete virtual entity. It repeatedly adjusts process parameters and performs iterative verification to complete process optimization.
[0015] Furthermore, the dual-modal collaborative adaptive model includes an input layer, a feature extraction layer, an adaptive fusion layer, and an output layer; in, The input layer is provided with a static parameter channel and a dynamic parameter channel, which are used to input static parameters and dynamic thermal characteristic parameters respectively; The feature extraction layer is equipped with a static feature encoder and a dynamic feature encoder to extract static features and dynamic thermal features, respectively. The adaptive fusion layer is equipped with a context-gated network to detect the validity of dynamic thermal feature parameters in real time. In online state management mode, static features and dynamic thermal features are weighted and fused through a cross-modal attention mechanism. In offline simulation mode, it automatically switches to the static feature bypass path and blocks the dynamic channel. The output layer outputs porosity via a prediction head.
[0016] Compared with existing technologies, the principles and advantages of this technical solution are as follows: 1. Achieve a paradigm shift in quality control: from "post-event detection and passive correction" to "pre-event prediction and proactive prevention," enabling proactive intervention in highly dynamic defects such as keyholes, reducing the defect incidence rate by more than 60%.
[0017] 2. Improve model credibility and executability: Introduce Score-CAM interpretability technology to transform black-box prediction into a visual heatmap, allowing operators to intuitively understand the causes of defects, improving the transparency of model decision-making and significantly enhancing human-machine trust.
[0018] 3. Significantly reduce process optimization costs: Offline virtual simulation mode replaces physical trial and error, shortening the cycle of a single process optimization from several weeks to several hours, reducing experimental costs by more than 70%, while supporting global optimal search in high-dimensional parameter space.
[0019] 4. Constructing an adaptive intelligent closed loop: The dual-modal collaborative adaptive model can automatically adjust the decision path based on data completeness without manual switching. Combined with an online continuous learning mechanism, the system's adaptability to new working conditions is continuously enhanced.
[0020] 5. Enhance on-site interaction efficiency: Digital twin technology enables seamless overlay of digital information in physical space, allowing operators to obtain a "what you see is what you get" monitoring experience in real-world scenarios, with response speed improved by more than 50%. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the services required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0022] Figure 1 This is a control principle diagram of the LPBF hole defect prevention and control system based on dual-modal fusion digital twin in an embodiment of the present invention. Detailed Implementation
[0023] The present invention will be further described below with reference to specific embodiments: like Figure 1 As shown in this embodiment, the LPBF hole defect prevention and control system based on dual-modal fusion digital twin includes an online status management and data acquisition module and a digital twin system.
[0024] The digital twin system incorporates a multi-source data perception and fusion module, a dual-modal collaborative adaptive model, an anomaly localization model, a dynamic-static correlation model, a simulation module, and an offline simulation and process optimization module.
[0025] The system includes: an online status management and data acquisition module to determine whether the equipment hardware is online or offline and acquire multi-source heterogeneous data based on the status result; a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, which is then fed back to the laser powder bed melting equipment; a multi-source data perception and fusion module to perceive and fuse the input multi-source heterogeneous data; a dual-modal collaborative adaptive model to adaptively select the prediction path based on the effectiveness of dynamic thermal characteristic parameters in the multi-source heterogeneous data, and predict the corresponding porosity value through the selected prediction path; an anomaly localization model to generate a two-dimensional visualized thermal map using Score-CAM technology, and then map it to three-dimensional space to complete the stereoscopic projection of pore defects; a simulation module to perform high-fidelity process simulation and generate a complete virtual entity of the printed part; a dynamic-static correlation model to obtain the dynamic thermal characteristic parameters of the simulation; and an offline simulation and process optimization module that combines the output of the dual-modal collaborative adaptive model (which outputs the second porosity value, the stereoscopic projection of pore defects, and the complete virtual entity) to repeatedly adjust process parameters and perform iterative verification to complete process optimization.
[0026] Specifically, in this embodiment, the dual-modal collaborative adaptive model includes an input layer, a feature extraction layer, an adaptive fusion layer, and an output layer.
[0027] The input layer has static parameter channels and dynamic parameter channels for inputting static parameters and dynamic thermal feature parameters, respectively; the feature extraction layer has static feature encoders and dynamic feature encoders for extracting static features and dynamic thermal features, respectively; the adaptive fusion layer has a context-gated network to detect the validity of dynamic thermal feature parameters in real time. In online state management mode, static features and dynamic thermal features are weighted and fused through a cross-modal attention mechanism, while in offline simulation mode, it automatically switches to the static feature bypass path and blocks the dynamic channel; the output layer outputs porosity through a prediction head.
[0028] like Figure 1 As shown, the working principle of this embodiment is as follows: S1. The online status management and data acquisition module determines the status (whether the hardware is online or offline) and acquires multi-source heterogeneous data based on the status result. The multi-source heterogeneous data includes static parameters and dynamic thermal characteristic parameters. The static parameters are composed of process parameters, model geometric parameters and kinematic parameters, and the dynamic thermal characteristic parameters are high-frequency time-series thermal imaging data of the molten pool area.
[0029] S2. Input multi-source heterogeneous data into the digital twin system. The multi-source data sensing and fusion module built into the digital twin system senses the received multi-source heterogeneous data. If the sensed multi-source heterogeneous data are all static parameters, the static parameters are directly input into the dual-modal cooperative adaptive model built into the digital twin system. If the sensed multi-source heterogeneous data include static parameters and dynamic thermal characteristic parameters, the static parameters and dynamic thermal characteristic parameters are fused and then input into the dual-modal cooperative adaptive model.
[0030] S3. The dual-modal collaborative adaptive model adaptively selects the prediction path based on the validity of the dynamic thermal characteristic parameters in the input multi-source heterogeneous data: when the dynamic thermal characteristic parameters are valid, the online prediction path that integrates static parameters and dynamic thermal characteristic parameters is activated to predict porosity and obtain the first porosity value; when the dynamic thermal characteristic parameters are missing, the model automatically switches to the offline simulation prediction path that relies only on static parameters to predict porosity and obtain the second porosity value.
[0031] If the dual-modal collaborative adaptive model outputs the first porosity value, i.e., in online state management mode, the online state management and data acquisition module generates a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, and feeds it back to the laser powder bed melting equipment to realize active prevention and process intervention of pore defects; if the dual-modal collaborative adaptive model outputs the second porosity value, i.e., in offline simulation mode, closed-loop control and cyclic optimization of process parameters are performed.
[0032] In both the online status management mode and the offline simulation mode, the dynamic thermal characteristic parameters are input into the anomaly localization model built into the digital twin system. Using Score-CAM technology, a two-dimensional visual thermal map is generated and then mapped to three-dimensional space to complete the stereoscopic projection of the pore defects. At the same time, based on geometric and kinematic parameters, the simulation module built into the digital twin system performs high-fidelity process simulation to generate a complete virtual entity of the printed part.
[0033] In the online status management mode, the input is the dynamic thermal characteristic parameters acquired by the online status management and data acquisition module, while in the offline simulation mode, the input is the dynamic thermal characteristic parameters obtained by simulation.
[0034] The dynamic thermal characteristic parameters obtained from the simulation are simulated using the dynamic-static correlation model built into the digital twin system. This dynamic-static correlation model maps the relationship between static parameters and dynamic thermal characteristic parameters.
[0035] In this embodiment, both the dynamic thermal characteristic parameters collected and the dynamic thermal characteristic parameters obtained by simulation include molten pool geometric morphology characteristic parameters and time-temperature related characteristic parameters; Among them, the geometric features of the molten pool include: molten pool area, molten pool length, molten pool width, molten pool eccentricity, molten pool perimeter, average molten pool temperature, and maximum molten pool temperature; the time-temperature related feature parameters include: the duration of a single pixel temperature exceeding 1200K, the duration of a temperature exceeding 1680K, and the duration of a temperature exceeding 2400K.
[0036] The online status management and data acquisition module is interconnected with the laser powder bed melting equipment to acquire multi-source heterogeneous data from the control card and infrared thermal imager in the laser powder bed melting equipment in real time. The control card reads laser power, scanning speed, scanning path coordinates, and powder layer thickness parameters; the infrared thermal imager captures high-frequency time-series thermal imaging data.
[0037] In offline simulation mode, the output of the dual-modal collaborative adaptive model is the second porosity value, the three-dimensional projection of pore defects, and the complete virtual entity. The process parameters are repeatedly adjusted and iteratively verified through the offline simulation and process optimization module built into the digital twin system, thereby completing the process optimization.
[0038] In both online status management mode and offline simulation mode, the digital twin system ensures a high degree of coordination between motion simulation and physical printing process by parsing G-code, thereby reproducing the generation process of the printed object in virtual space.
[0039] This embodiment constructs a complete intelligent closed-loop system: the system determines the optimal process baseline through efficient virtual optimization in offline simulation mode, and transforms this optimization result into precise control commands for the laser powder bed melting equipment in online status management mode. Simultaneously, the online status management and data acquisition module continuously monitors the system to ensure rapid response to any process disturbances. This virtual-real fusion and data-driven approach not only significantly accelerates the process development and quality verification cycle of LPBF, but also achieves high-quality, adaptive, and precise closed-loop control that prevents defects from the source.
[0040] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twin, characterized in that, include: The status is determined by the online status management and data acquisition module, and multi-source heterogeneous data is obtained based on the status result. The multi-source heterogeneous data includes static parameters and dynamic thermal characteristic parameters. The static parameters are composed of process parameters, model geometric parameters and kinematic parameters, and the dynamic thermal characteristic parameters are high-frequency time-series thermal imaging data of the molten pool region. The multi-source heterogeneous data is input into the digital twin system. The multi-source data sensing and fusion module built into the digital twin system senses the received multi-source heterogeneous data. If the sensed multi-source heterogeneous data are all static parameters, the static parameters are directly input into the dual-modal cooperative adaptive model built into the digital twin system. If the sensed multi-source heterogeneous data include static parameters and dynamic thermal characteristic parameters, the static parameters and dynamic thermal characteristic parameters are fused and then input into the dual-modal cooperative adaptive model. The dual-modal collaborative adaptive model adaptively selects the prediction path based on the validity of the dynamic thermal characteristic parameters in the input multi-source heterogeneous data: when the dynamic thermal characteristic parameters are valid, the online prediction path that integrates static parameters and dynamic thermal characteristic parameters is activated to predict porosity and obtain the first porosity value; when the dynamic thermal characteristic parameters are missing, the model automatically switches to the offline simulation prediction path that relies only on static parameters to predict porosity and obtain the second porosity value. If the dual-modal collaborative adaptive model outputs the first porosity value, i.e., in online state management mode, the online state management and data acquisition module generates a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, and feeds it back to the laser powder bed melting equipment to realize active prevention and process intervention of pore defects; if the dual-modal collaborative adaptive model outputs the second porosity value, i.e., in offline simulation mode, closed-loop control and cyclic optimization of process parameters are performed.
2. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 1, characterized in that, Both online status management mode and offline simulation mode include inputting dynamic thermal characteristic parameters into the anomaly localization model built into the digital twin system, using Score-CAM technology to generate a two-dimensional visual thermal map, and then mapping it to three-dimensional space to complete the stereoscopic projection of pore defects; at the same time, based on geometric and kinematic parameters, the simulation module built into the digital twin system performs high-fidelity process simulation to generate a complete virtual entity of the printed part. In the online status management mode, the input is the dynamic thermal characteristic parameters acquired by the online status management and data acquisition module, while in the offline simulation mode, the input is the dynamic thermal characteristic parameters obtained by simulation.
3. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 2, characterized in that, The dynamic thermal characteristic parameters obtained from the simulation are simulated using a dynamic-static correlation model built into the digital twin system. This dynamic-static correlation model maps the relationship between static parameters and dynamic thermal characteristic parameters.
4. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 2, characterized in that, Both the dynamic thermal characteristic parameters obtained from the acquisition and the dynamic thermal characteristic parameters obtained from the simulation include molten pool geometric morphology characteristic parameters and time-temperature related characteristic parameters. Among them, the geometric features of the molten pool include: molten pool area, molten pool length, molten pool width, molten pool eccentricity, molten pool perimeter, average molten pool temperature, and maximum molten pool temperature; the time-temperature related feature parameters include: the duration of a single pixel temperature exceeding 1200K, the duration of a temperature exceeding 1680K, and the duration of a temperature exceeding 2400K.
5. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 2, characterized in that, In offline simulation mode, the output of the dual-modal collaborative adaptive model is the second porosity value, the three-dimensional projection of pore defects, and the complete virtual entity. The process parameters are repeatedly adjusted and iteratively verified through the offline simulation and process optimization module built into the digital twin system, thereby completing the process optimization.
6. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 1, characterized in that, The online status management and data acquisition module is interconnected with the laser powder bed melting equipment and acquires multi-source heterogeneous data from the control card and infrared thermal imager in the laser powder bed melting equipment in real time. The control card reads parameters such as laser power, scanning speed, scanning path coordinates, and powder layer thickness. Infrared thermal imagers capture high-frequency time-series thermal imaging data.
7. The method for preventing and controlling LPBF hole defects based on dual-modal fusion digital twins according to claim 2, characterized in that, In both online status management mode and offline simulation mode, the digital twin system ensures a high degree of coordination between motion simulation and physical printing process by parsing G-code, thereby reproducing the generation process of the printed object in virtual space.
8. An LPBF hole defect prevention and control system based on dual-modal fusion digital twin, used to implement the LPBF hole defect prevention and control method based on dual-modal fusion digital twin as described in any one of claims 1-7, characterized in that, Includes online status management and data acquisition modules and a digital twin system; The digital twin system incorporates a multi-source data sensing and fusion module, a dual-modal collaborative adaptive model, an anomaly localization model, a dynamic-static correlation model, a simulation module, and an offline simulation and process optimization module. in, The online status management and data acquisition module is used to determine whether the equipment hardware is in an online or offline state and to acquire multi-source heterogeneous data based on the status result. It also includes generating a closed-loop control signal containing parameter adjustment range or shutdown command based on the comparison result of the predicted first porosity value and the preset quality threshold, and feeding it back to the laser powder bed melting equipment. The multi-source data sensing and fusion module senses and fuses the input multi-source heterogeneous data; The dual-modal collaborative adaptive model adaptively selects the prediction path based on the effectiveness of dynamic thermal characteristic parameters in multi-source heterogeneous data, and predicts the corresponding porosity value through the selected prediction path. The anomaly localization model uses Score-CAM technology to generate a two-dimensional visualization heat map, which is then mapped to three-dimensional space to complete the stereoscopic projection of pore defects. The simulation module performs high-fidelity process simulation to generate a complete virtual entity of the printed part. The dynamic-static correlation model is used to obtain the dynamic thermal characteristic parameters of the simulation. The offline simulation and process optimization module, combined with the dual-modal collaborative adaptive model, outputs a second porosity value, a three-dimensional projection of pore defects, and a complete virtual entity. It repeatedly adjusts process parameters and performs iterative verification to complete process optimization.
9. The LPBF hole defect prevention and control system based on dual-modal fusion digital twin according to claim 8, wherein the dual-modal cooperative adaptive model includes an input layer, a feature extraction layer, an adaptive fusion layer, and an output layer; in, The input layer is provided with a static parameter channel and a dynamic parameter channel, which are used to input static parameters and dynamic thermal characteristic parameters respectively; The feature extraction layer is equipped with a static feature encoder and a dynamic feature encoder to extract static features and dynamic thermal features, respectively. The adaptive fusion layer is equipped with a context-gated network to detect the validity of dynamic thermal feature parameters in real time. In online state management mode, static features and dynamic thermal features are weighted and fused through a cross-modal attention mechanism. In offline simulation mode, it automatically switches to the static feature bypass path and blocks the dynamic channel. The output layer outputs porosity via a prediction head.