Real-time defect detection and in-situ repair method and device in additive manufacturing process

Abstract

The invention relates to the technical field of additive manufacturing, and particularly discloses a real-time defect detection and in-situ repair method and device in the additive manufacturing process, and the method comprises the steps that a multi-mode sensing system is integrated on additive manufacturing equipment, the multi-mode sensing system comprises a high-resolution CCD camera, a laser scanner and a thermal infrared imager, and data acquisition is synchronized; a hierarchical scanning strategy is adopted, two-dimensional surface images, three-dimensional point cloud data and thermal field distribution information are collected before and after each layer is machined, and the data synchronization precision is smaller than 0.1 ms; according to the invention, a multi-modal sensing system and an advanced algorithm are integrated, so that real-time detection and accurate repair of defects are successfully realized; a high-resolution CCD camera, a laser scanner, a thermal infrared imager and other sensors are adopted, a cross-scale RANSAC algorithm and a depth feature fusion network are combined, surface and internal defects are accurately detected, and laser remelting parameters are adjusted in real time for in-situ repair.

Description

Technical Field

[0001] The present invention belongs to the technical field of additive manufacturing, and particularly relates to a method and device for real-time defect detection and in-situ repair during the additive manufacturing process. Background Art

[0002] Additive manufacturing technology is an advanced intelligent digital manufacturing technology developed in the late 1980s of the 20th century. It is a manufacturing method that takes product model data as the basis and accumulates materials layer by layer into a solid. Its core is layer-by-layer stacking, which transforms the manufacturing of three-dimensional parts into the manufacturing of a series of two-dimensional thin slices. Using computer modeling, the model is discretized and then solidified by superposition in the processing system.

[0003] However, additive manufacturing is the melting and stacking of powder by high-energy beam rays, and at the same time lacks traditional forging processes. Pores, cracks and other defects will inevitably appear in additive manufacturing products during the manufacturing process. These defects directly affect the mechanical properties and service life of additive manufacturing components. Traditional defect detection in the additive manufacturing process usually relies on later manual inspection or off-line inspection, and it is difficult to achieve real-time monitoring and rapid repair. In addition, defect repair mainly relies on manual intervention or complex post-processing techniques and lacks the ability of in-situ repair. To solve these problems, it is necessary to conduct on-line inspection of the part state during the additive manufacturing process and add repair means. The current additive manufacturing defect detection methods mainly rely on single data sources such as visual images, thermal imaging or laser scanning, or discover defects through post-processing, and it is difficult to achieve real-time and accurate defect detection. For defect repair, traditional repair methods often rely on manual operations, resulting in unstable repair effects and long repair process cycles.

[0004] Therefore, it is necessary to propose a method and device for real-time defect detection and in-situ repair during the additive manufacturing process to solve the problems existing in the prior art that it is difficult to perform real-time defect detection and repair during the additive manufacturing process.

[0005] The above information disclosed in this background art is only used to increase the understanding of the background art of the present invention. Therefore, it may include prior art that is not known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and device for real-time defect detection and in-situ repair during the additive manufacturing process to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] A method for real-time defect detection and in-situ repair during the additive manufacturing process includes:

[0009] Integrate a multimodal sensing system on an additive manufacturing device. The multimodal sensing system includes a high-resolution CCD camera, a laser scanner, and an infrared thermal imager, and synchronize data acquisition;

[0010] Adopt a hierarchical scanning strategy to collect two-dimensional surface images, three-dimensional point cloud data, and thermal field distribution information before and after each layer of processing, and the data synchronization accuracy is less than 0.1 ms;

[0011] Perform real-time defect detection based on an improved cross-scale RANSAC algorithm and a deep feature fusion network. The cross-scale RANSAC algorithm uses a dynamic scale sampling strategy and geometric consistency constraints for defect evaluation;

[0012] Process the two-dimensional surface image through a dual-branch CNN, combine PointNet++ to process the three-dimensional point cloud data, and achieve multi-source data fusion detection through a feature cross-attention mechanism;

[0013] For the detected defect area, dynamically adjust the laser remelting parameters according to the defect type and severity for in-situ repair;

[0014] Adopt CUDA parallel computing and TensorRT acceleration to achieve real-time data processing, and the end-to-end delay is less than 200 ms.

[0015] Preferably, the multimodal sensing system includes a high-resolution CCD camera, a laser scanner, and an infrared thermal imager, and synchronize data acquisition, including:

[0016] Realize the time alignment of the CCD camera, laser scanner, and infrared thermal imager based on the timestamp synchronization algorithm;

[0017] Adopt an adaptive exposure control algorithm to dynamically adjust the exposure parameters of the CCD camera;

[0018] Fuse multi-view laser scanning data through a point cloud registration algorithm to optimize the three-dimensional reconstruction accuracy.

[0019] Preferably, the analysis steps of the improved cross-scale RANSAC algorithm include:

[0020] Introduce a scale adaptive function S(i), and dynamically adjust the sampling scale through the curvature or local geometric features of the points:

[0021]

[0022] In the formula, α is an adjustment parameter, and C(i) is the curvature value of point i;

[0023] Dynamically adjust the maximum number of iterations and the distance threshold, and the number of iterations ranges from 100 to 1000 times;

[0024] Introduce geometric consistency constraints, including curvature consistency and normal vector consistency;

[0025] Adopt GPU parallel acceleration technology, and the processing speed reaches 5 layers per minute.

[0026] Preferably, the analysis steps of the depth feature fusion network include:

[0027] Use the ResNet-34 network structure combined with a multi-scale feature pyramid to process two-dimensional surface images;

[0028] Use the lightweight PointNet++ network structure embedded with a non-local attention mechanism to process three-dimensional point cloud data;

[0029] Feature fusion is achieved through a cross-attention mechanism, and the weight coefficients are dynamically adjusted according to different features.

[0030] Preferably, the in-situ repair by dynamically adjusting the laser remelting parameters according to the defect type and severity includes:

[0031] Based on historical defect repair data, establish a defect feature - repair parameter mapping database;

[0032] According to the three-dimensional size and position of the defect, use the Bayesian optimization algorithm to search for the optimal combination of laser power, scanning speed, and remelting times;

[0033]

[0034] Where EI(x) is the parameter point of the maximum expected improvement, f best is the current optimal repair effect, σ(x) is the standard deviation, μ(x) is the predicted mean, and Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively;

[0035] Adopt a reinforcement learning algorithm to continuously optimize the parameter matching strategy, and predict the repair parameter drift trend of multi-layer continuous defects based on the LSTM time series prediction model.

[0036] Preferably, the method further includes an embedded processing unit based on NVIDIA Jetson AGX Orin; the embedded processing unit adopts a memory pool optimization technology and a multi-task parallel processing framework to optimize the data processing efficiency.

[0037] Preferably, the defect detection includes the following aspects:

[0038] Surface defect detection: including the identification of spheroidization, splash residue, and surface cracks, with a detection accuracy of 20μm;

[0039] Internal defect prediction: predict the risks of lack of fusion, porosity, and interlayer separation through abnormal thermal fields;

[0040] Introduce an anomaly detection method based on a thermal model:

[0041] ΔT = |T a - T d |

[0042] Wherein, T a and T d are the actual thermal field temperature and the expected thermal field temperature respectively;

[0043] Severity classification of defects: Combine the size, depth and position of the defects for three-level classification.

[0044] Preferably, the method further includes a quality traceability system:

[0045] Record the detection data, repair parameters and process parameters of each layer, and generate a three-dimensional defect distribution map and a quality analysis report;

[0046] Construct a three-dimensional relationship network of process parameters - defect types - repair effects based on a knowledge graph construction algorithm, and provide process reverse traceability and parameter optimization suggestions.

[0047] A device for real-time detection and in-situ repair of defects in the additive manufacturing process, comprising:

[0048] A multi-modal sensing module, integrating a CCD camera, a laser scanner and an infrared thermal imager, for data acquisition;

[0049] A real-time processing module, equipped with an improved cross-scale RANSAC algorithm and a deep feature fusion network, for defect detection;

[0050] A laser remelting module, equipped with a dynamic parameter adjustment system, for defect repair;

[0051] A control center, for realizing synchronous communication of each module through Gigabit Ethernet;

[0052] A human-computer interaction interface, for real-time displaying the three-dimensional defect distribution and the repair process, and providing an operation control interface.

[0053] Preferably, the laser remelting module is further used for:

[0054] Combining the A* algorithm and B-spline curve fitting to generate a jitter-free scanning path;

[0055] Adopting an energy density equalization algorithm to eliminate the edge heat accumulation effect in the repair area;

[0056] Establishing an optical path offset compensation model based on Gaussian process regression, with a positioning error compensation accuracy of ±2μm;

[0057] Gaussian process regression compensation formula:

[0058] D(r) = k(r, R)K(R, R) -1 Y

[0059] Where D(r) is the predicted compensation value, k(r, R) is the covariance function between the test point and the training point, K(R, R) is the covariance matrix of the training data points, and Y is the training label.

[0060] Compared with the prior art, the beneficial effects of the present invention are:

[0061] Through the integration of a multi-modal sensing system and advanced algorithms, the present invention has successfully achieved real-time detection and precise repair of defects; by using sensors such as high-resolution CCD cameras, laser scanners, and infrared thermal imagers, combined with cross-scale RANSAC algorithms and deep feature fusion networks, surface and internal defects are accurately detected, and in-situ repair is carried out by real-time adjustment of laser remelting parameters; at the same time, the feature cross-attention mechanism is used to optimize multi-source data fusion, combined with CUDA parallel computing and TensorRT acceleration technology, to achieve efficient data processing and real-time feedback, ensuring the efficiency of defect detection and repair. Through Bayesian optimization and reinforcement learning algorithms, the repair parameters are automatically adjusted to further improve the stability and accuracy of the repair. The entire system also realizes quality traceability and process optimization, records the detection and repair data of each layer, supports the construction of a three-dimensional relationship network between process parameters and defect types, and provides support for subsequent process optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 is a flowchart of the method for real-time defect detection and in-situ repair during the additive manufacturing process of the present invention;

[0063] Figure 2 is a framework diagram of the device for real-time defect detection and in-situ repair during the additive manufacturing process of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] Example 1:

[0066] Please refer to Figure 1 As shown, a method for real-time defect detection and in-situ repair during the additive manufacturing process includes:

[0067] Integrate a multimodal sensing system on an additive manufacturing device. The multimodal sensing system includes a high-resolution CCD camera, a laser scanner, and an infrared thermal imager, and synchronize data acquisition.

[0068] Based on the timestamp synchronization algorithm, achieve time alignment of the CCD camera, laser scanner, and infrared thermal imager.

[0069] Adopt an adaptive exposure control algorithm to dynamically adjust the exposure parameters of the CCD camera.

[0070] Fuse multi-view laser scanning data through a point cloud registration algorithm to optimize the accuracy of three-dimensional reconstruction.

[0071] Furthermore, integrate a multimodal sensing system on an additive manufacturing device. Through the synchronous data acquisition and time alignment of the high-resolution CCD camera, laser scanner, and infrared thermal imager, monitor the defects in the manufacturing process in real time and improve the detection accuracy.

[0072] Adopt a layer-by-layer scanning strategy. Before and after each layer of processing, collect two-dimensional surface images, three-dimensional point cloud data, and thermal field distribution information respectively. The data synchronization accuracy is less than 0.1ms.

[0073] Based on an improved cross-scale RANSAC algorithm and a deep feature fusion network, perform real-time defect detection. The defect detection includes the following aspects:

[0074] Surface defect detection: including the identification of balling, spatter residue, and surface cracks. The detection accuracy reaches 20μm.

[0075] Internal defect prediction: Predict the risks of lack of fusion, porosity, and interlayer separation through thermal field anomalies.

[0076] Defect severity classification: Combine the size, depth, and location of the defects for three-level classification.

[0077] The analysis steps of the deep feature fusion network include:

[0078] Adopt a ResNet-34 network structure combined with a multi-scale feature pyramid to process two-dimensional surface images.

[0079] Adopt a lightweight PointNet++ network structure embedded with a non-local attention mechanism to process three-dimensional point cloud data.

[0080] Feature fusion is achieved through a cross-attention mechanism, and the weight coefficients are dynamically adjusted according to different features.

[0081] Furthermore, by using a ResNet-34 network that combines a multi-scale feature pyramid to process two-dimensional images and a PointNet++ network that embeds a non-local attention mechanism to process three-dimensional point cloud data, surface defects (such as spheroidization, splash residue, cracks) can be accurately identified and the risk of internal defects (such as lack of fusion, porosity, interlayer separation) can be predicted. Feature fusion adopts a cross-attention mechanism to dynamically adjust the weight coefficients, thereby improving the detection accuracy and the three-level classification ability of the defect severity, and effectively enhancing the quality control and defect warning capabilities in the additive manufacturing process.

[0082] The cross-scale RANSAC algorithm uses a dynamic scale sampling strategy and geometric consistency constraints for defect assessment;

[0083] The analysis steps of the improved cross-scale RANSAC algorithm include:

[0084] Introduce a scale adaptive function to dynamically adjust the sampling scale through the curvature of points or local geometric features;

[0085] Dynamically adjust the maximum number of iterations and the distance threshold, and the range of the number of iterations is from 100 to 1000 times;

[0086] Introduce geometric consistency constraint conditions, including curvature consistency and normal vector consistency;

[0087] Adopt GPU parallel acceleration technology, and the processing speed reaches 5 layers per minute.

[0088] Furthermore, the algorithm adopts a probability-guided sampling strategy, preferentially selects the point cloud in the region of curvature mutation as the initial seed, enhancing the ability to identify the defect region. At the same time, dynamically adjust the maximum number of iterations and the distance threshold to optimize the robustness and flexibility of the algorithm. By introducing geometric constraints such as curvature consistency and normal vector consistency, the accuracy of the evaluation results is further improved. In addition, the application of GPU parallel acceleration technology significantly increases the processing speed, reaching 5 layers per minute, greatly improving the efficiency of real-time defect detection.

[0089] Process two-dimensional surface images through a dual-branch CNN, combine PointNet++ to process three-dimensional point cloud data, and achieve multi-source data fusion detection through a feature cross-attention mechanism;

[0090] For the detected defect region, dynamically adjust the laser remelting parameters for in-situ repair according to the defect type and severity;

[0091] The dynamic parameter adjustment process includes:

[0092] Based on historical defect repair data, establish a defect feature - repair parameter mapping database;

[0093] According to the three-dimensional size and position of the defect, the Bayesian optimization algorithm is used to search for the optimal combination of laser power, scanning speed, and remelting times;

[0094] The reinforcement learning algorithm is used to continuously optimize the parameter matching strategy, and the repair parameter drift trend of multi-layer continuous defects is predicted based on the LSTM time series prediction model.

[0095] Furthermore, the Bayesian optimization algorithm is used to search for the optimal combination of laser power, scanning speed, and remelting times, ensuring the best match of the repair effect. At the same time, the application of the reinforcement learning algorithm continuously optimizes the parameter strategy, improving the intelligence and adaptability of the repair process. In addition, the LSTM time series prediction model effectively predicts the repair parameter drift trend of multi-layer continuous defects, further improving the repair accuracy and stability, and enhancing the repair efficiency and quality control ability in the additive manufacturing process.

[0096] CUDA parallel computing and TensorRT acceleration are used to achieve real-time data processing, with an end-to-end delay of less than 200 ms.

[0097] The method also includes an embedded processing unit based on NVIDIA Jetson AGX Orin;

[0098] The memory pool optimization technology and multi-task parallel processing framework are used to optimize the data processing efficiency.

[0099] The method also includes a quality traceability system, which is used for:

[0100] Recording the detection data, repair parameters, and process parameters of each layer, and generating a three-dimensional defect distribution map and a quality analysis report;

[0101] Constructing a three-dimensional relationship network of process parameters - defect types - repair effects based on the knowledge graph construction algorithm, and providing process reverse traceability and parameter optimization suggestions.

[0102] Furthermore, the method uses an embedded processing unit based on NVIDIA Jetson AGX Orin, combined with the memory pool optimization technology and multi-task parallel processing framework, significantly improving the data processing efficiency and real-time performance. At the same time, the introduction of the quality traceability system effectively records the detection data, repair parameters, and process parameters of each layer, generating a detailed three-dimensional defect distribution map and a quality analysis report, ensuring the traceability of the process.

[0103] Example 2:

[0104] Please refer to Figure 2 As shown, a device for real-time detection and in-situ repair of defects in the additive manufacturing process includes:

[0105] Multimodal sensing module, integrating a CCD camera, a laser scanner, and an infrared thermal imager, for data acquisition;

[0106] Real-time processing module, equipped with an improved cross-scale RANSAC algorithm and a depth feature fusion network, for defect detection;

[0107] Laser remelting module, equipped with a dynamic parameter adjustment system, for defect repair;

[0108] Control center, for realizing synchronous communication among modules through Gigabit Ethernet;

[0109] Human-machine interaction interface, for real-time displaying of three-dimensional defect distribution and repair process, and providing an operation control interface.

[0110] The laser remelting module is also used to generate a jitter-free scanning path by combining the A* algorithm and B-spline curve fitting;

[0111] An energy density equalization algorithm is adopted to eliminate the edge heat accumulation effect in the repair area;

[0112] Based on Gaussian process regression, an optical path offset compensation model is established, and the positioning error compensation accuracy reaches ±2μm.

[0113] Application example: Real-time defect detection and in-situ repair during additive manufacturing

[0114] I. Application background

[0115] Additive manufacturing technology, as a star technology in modern advanced manufacturing, is shining brightly in many fields such as aerospace, automotive, and medical devices. However, at the same time, this technology also faces an important challenge: during the manufacturing process, due to the layer-by-layer stacking of materials, the rapid change of temperature, and the complex stress distribution, defects such as pores, cracks, balling phenomena, and splash residues often occur. These defects not only affect the aesthetics of the product but also pose a serious threat to its mechanical properties and service life. To address this challenge, we combine an innovative method and device for real-time defect detection and in-situ repair during additive manufacturing, aiming to achieve instant monitoring and repair during the manufacturing process to ensure product quality.

[0116] II. Technical solution

[0117] On the additive manufacturing equipment, we integrate a multimodal sensing system including a high-resolution CCD camera, a laser scanner, and an infrared thermal imager. These sensors can synchronously collect two-dimensional surface images, three-dimensional point cloud data, and thermal field distribution information during the manufacturing process, and the data synchronization accuracy is less than 0.1ms. Such high-precision data acquisition provides a solid foundation for subsequent defect detection.

[0118] We adopted an improved cross-scale RANSAC algorithm and a deep feature fusion network for real-time defect detection. The cross-scale RANSAC algorithm can dynamically adjust the sampling scale according to the curvature of points or local geometric features, and conduct defect evaluation by combining geometric consistency constraints (including curvature consistency and normal vector consistency). Meanwhile, the deep feature fusion network realizes the feature fusion detection of two-dimensional surface images and three-dimensional point cloud data by combining the ResNet-34 network structure of the multi-scale feature pyramid and the lightweight PointNet++ network structure embedded with the non-local attention mechanism.

[0119] Once a defect is detected, we immediately dynamically adjust the laser remelting parameters for in-situ repair according to the type and severity of the defect. This process is based on a defect feature-repair parameter mapping database established from historical defect repair data, and uses the Bayesian optimization algorithm to search for the optimal combination of laser power, scanning speed, and remelting times. At the same time, we also use the reinforcement learning algorithm to continuously optimize the parameter matching strategy, and predict the repair parameter drift trend of multi-layer continuous defects based on the LSTM time series prediction model.

[0120] To ensure real-time performance, we adopted the NVIDIA Jetson AGX Orin embedded processing unit, and combined memory pool optimization technology and a multi-task parallel processing framework to optimize data processing efficiency. In addition, the application of CUDA parallel computing and TensorRT acceleration technology also makes real-time data processing possible, with an end-to-end latency of less than 200 ms.

[0121] We also introduced a quality traceability system to record the detection data, repair parameters, and process parameters of each layer, and generate a three-dimensional defect distribution map and a quality analysis report. Based on the knowledge graph construction algorithm, we associated a three-dimensional relationship network of process parameters-defect types-repair effects, providing strong support for subsequent process optimization.

[0122] III. Application Process

[0123] During the additive manufacturing process, the multi-modal sensing system continuously collects data, and conducts real-time defect detection through the improved cross-scale RANSAC algorithm and the deep feature fusion network. Once a defect is detected, the system immediately dynamically adjusts the laser remelting parameters for in-situ repair according to the defect characteristics and severity. At the same time, the quality traceability system records all relevant parameters and data, providing a basis for subsequent analysis.

[0124] IV. Application Effects

[0125] Through the application of this technical solution, we have successfully achieved the real-time detection and in-situ repair of defects during the additive manufacturing process. This not only significantly improves the manufacturing quality and production efficiency of products but also ensures the traceability of the process. The detailed three-dimensional defect distribution maps and quality analysis reports generated by the quality traceability system provide strong support for subsequent process optimization.

[0126] Example 3:

[0127] The embodiment of the present invention also provides a computer-readable storage medium. A program of the method for real-time detection and in-situ repair of defects during additive manufacturing as described in any one of the above is stored on the computer-readable storage medium. When the program is executed by a processor, it realizes each process of the above-described embodiment of the method for real-time detection and in-situ repair of defects and can achieve the same technical effects. To avoid repetition, it will not be elaborated here. Among them, the computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc, etc.

[0128] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0129] In the drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments of the present invention are involved. Other structures can refer to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0130] The flowcharts shown in the drawings are only illustrative examples and do not necessarily include all the contents and operations / steps, nor do they necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged. Therefore, the actual execution order may change according to the actual situation.

[0131] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time defect detection and in-situ repair during additive manufacturing, characterized in that, Including: Integrating a multi-modal sensing system on an additive manufacturing device, the multi-modal sensing system includes a high-resolution CCD camera, a laser scanner, and an infrared thermal imager, and synchronizes data acquisition; Adopting a layer-by-layer scanning strategy, acquiring two-dimensional surface images, three-dimensional point cloud data, and thermal field distribution information before and after processing each layer respectively, and the data synchronization accuracy is less than 0.1 ms; Performing real-time defect detection based on an improved cross-scale RANSAC algorithm and a deep feature fusion network, and the cross-scale RANSAC algorithm uses a dynamic scale sampling strategy and geometric consistency constraints for defect evaluation; Processing two-dimensional surface images through a dual-branch CNN, combining PointNet++ to process three-dimensional point cloud data, and realizing multi-source data fusion detection through a feature cross-attention mechanism; For the detected defect area, dynamically adjust the laser remelting parameters according to the defect type and severity for in-situ repair; Adopting CUDA parallel computing and TensorRT acceleration to achieve real-time data processing, and the end-to-end delay is less than 200 ms.

2. The real-time defect detection and in-situ repair method during additive manufacturing according to claim 1, wherein, The multi-modal sensing system includes a high-resolution CCD camera, a laser scanner, and an infrared thermal imager, and synchronizes data acquisition, including: Realizing time alignment of the CCD camera, laser scanner, and infrared thermal imager based on a timestamp synchronization algorithm; Adopting an adaptive exposure control algorithm to dynamically adjust the exposure parameters of the CCD camera; Fusing multi-view laser scanning data through a point cloud registration algorithm to optimize the three-dimensional reconstruction accuracy.

3. The real-time defect detection and in-situ repair method during additive manufacturing according to claim 2, wherein, The analysis steps of the improved cross-scale RANSAC algorithm include: Introducing a scale adaptive function S(i), and dynamically adjusting the sampling scale through the curvature or local geometric features of the points: In the formula, α is an adjustment parameter, and C(i) is the curvature value of point i; Dynamically adjusting the maximum number of iterations and the distance threshold, and the range of the number of iterations is 100 to 1000 times; Introducing geometric consistency constraint conditions, including curvature consistency and normal vector consistency; Adopting GPU parallel acceleration technology, and the processing speed reaches 5 layers per minute.

4. The real-time defect detection and in-situ repair method during additive manufacturing according to claim 3, characterized in that, The analysis steps of the deep feature fusion network include: Adopting a ResNet-34 network structure combined with a multi-scale feature pyramid to process two-dimensional surface images; Adopting a lightweight PointNet++ network structure embedded with a non-local attention mechanism to process three-dimensional point cloud data; Feature fusion is realized through a cross-attention mechanism, and the weight coefficients are dynamically adjusted according to different features.

5. A real-time defect detection and in-situ repair method during additive manufacturing according to claim 4, characterized in that, Dynamically adjusting the laser remelting parameters according to the defect type and severity for in-situ repair, including: Based on historical defect repair data, establishing a defect feature-repair parameter mapping database; According to the three-dimensional size and position of the defect, using the Bayesian optimization algorithm to search for the optimal combination of laser power, scanning speed, and remelting times; where EI(x) is the parameter point of the maximum expected improvement, f best is the current optimal repair effect, σ(x) is the standard deviation, μ(x) is the predicted mean, and Φ and φ are the cumulative distribution function and probability density function of the standard normal distribution, respectively; Adopting a reinforcement learning algorithm to continuously optimize the parameter matching strategy, and predicting the repair parameter drift trend of multi-layer continuous defects based on an LSTM time series prediction model.

6. The real-time defect detection and in-situ repair method during additive manufacturing according to claim 5, wherein, The method also includes an embedded processing unit based on NVIDIA Jetson AGX Orin; the embedded processing unit adopts a memory pool optimization technology and a multi-task parallel processing framework.

7. A method for real-time defect detection and in-situ repair during additive manufacturing according to claim 6, characterized in that The defect detection includes the following aspects: Surface defect detection: including the identification of spheroidization, splash residue and surface cracks, with a detection accuracy of 20μm; Internal defect prediction: predicting the risks of lack of fusion, porosity and interlayer separation through abnormal thermal fields; Introducing an anomaly detection method based on a thermal model: ΔT = |T a - T d | where, T a and T d are the actual thermal field and the expected thermal field temperatures, respectively; Defect severity classification: performing three-level classification by combining the size, depth and location of the defects.

8. A method for real-time defect detection and in-situ repair during additive manufacturing according to claim 7, characterized in that, The method further includes a quality traceability system, and the quality traceability system is used for: Recording the detection data, repair parameters and process parameters of each layer, and generating a three-dimensional defect distribution map and a quality analysis report; Constructing a three-dimensional relationship network of process parameters - defect types - repair effects based on a knowledge graph construction algorithm, and providing process reverse traceability and parameter optimization suggestions.

9. A device for real-time detection and in-situ repair of defects during additive manufacturing, characterized in that, Including: A multi-modal sensing module, integrating a CCD camera, a laser scanner and an infrared thermal imager for data acquisition; A real-time processing module, equipped with an improved cross-scale RANSAC algorithm and a deep feature fusion network for defect detection; A laser remelting module, equipped with a dynamic parameter adjustment system for defect repair; A control center for realizing synchronous communication among the modules through Gigabit Ethernet; A human-machine interaction interface for real-time displaying the three-dimensional defect distribution and the repair process, and providing an operation control interface.

10. The real-time defect detection and in-situ repair device during the additive manufacturing process according to claim 9, characterized in that, The laser remelting module is further used for: Combining the A* algorithm with B-spline curve fitting to generate a jitter-free scanning path; Adopting an energy density balancing algorithm to eliminate the edge heat accumulation effect in the repair area; Establishing an optical path offset compensation model based on Gaussian process regression, with a positioning error compensation accuracy of ±2μm; Gaussian process regression compensation formula: D(r) = k(r,R)K(R,R) -1 Y Where D(r) is the predicted compensation value, k(r,R) is the covariance function between the test point and the training point, K(R,R) is the covariance matrix of the training data points, and Y is the training label.

Images