Defect Identification and Early Warning System in Additive Manufacturing Process

Abstract

The present invention discloses a defect identification and early warning system in the additive manufacturing process. The system includes: a molten pool image data acquisition module, which forms a time series molten pool image data matrix; a molten pool image data preprocessing module, which forms a preprocessed molten pool image data matrix; a Harris hawk algorithm defect area positioning module, which uses an improved Harris hawk algorithm for global search and outputs a molten pool image data matrix of defect candidate areas; a fractal segmentation algorithm microcrack defect fine identification module, which obtains a molten pool image data matrix of the finely segmented microcrack defect areas; a microcrack defect feature extraction and quantitative evaluation module, which forms a quantitative evaluation result of microcrack defects; and a microcrack defect early warning module, which triggers a real-time early warning feedback mechanism. The present invention reduces the professionalism and technical threshold required for traditional modeling and improves the universality and convenience of 3D modeling.

Description

Technical Field

[0001] The present invention relates to the field of radar technology, and in particular to a defect identification and early warning system in the additive manufacturing process. Background Art

[0002] With the rapid development of additive manufacturing technology, metal 3D printing, as an important branch thereof, has been widely applied in the fields of aerospace, medical devices and high-end manufacturing. Compared with traditional subtractive manufacturing technology, metal additive manufacturing can achieve precise forming of complex structures and has significant advantages in terms of material utilization rate and design freedom. However, due to the high energy input and complex melt pool dynamics involved in additive manufacturing, various defects are likely to occur during the manufacturing process, such as microcracks, pores, and unfused regions. These defects directly affect the mechanical properties, fatigue life and service reliability of the finished parts. Therefore, how to identify and early warn these defects in real time during the additive manufacturing process has become an important challenge in current research and industrial applications.

[0003] At present, the mainstream additive manufacturing defect detection methods mainly include two categories: offline detection and online detection. Offline detection usually adopts X-ray tomography, metallographic analysis, and ultrasonic detection means to conduct high-precision internal structure analysis on the finished product, but there are two main problems: First, the detection cycle is long and cannot meet the requirements of real-time monitoring during the production process; Second, the detection cost is high and it is difficult to be widely applied in industrial production on a large scale. Therefore, the industry is gradually shifting towards online detection technology in order to identify and control defects in real time during the manufacturing process.

[0004] Online detection technology mainly relies on optical monitoring, thermal imaging analysis and data-driven modeling methods. Among them, optical monitoring technology uses a high-speed camera to capture the changes in the melt pool state and analyze the potential signs of defect generation, which can more intuitively reflect the dynamic characteristics of the melt pool during the manufacturing process. However, it has poor robustness in complex backgrounds and is easily interfered by environmental noise and the metal surface reflection characteristics, resulting in low recognition accuracy. In addition, thermal imaging analysis technology can monitor the melt pool temperature field distribution through an infrared thermal imager and infer the potential defect locations. However, due to the sensor accuracy and environmental temperature changes, there are limitations of high false positive rate and difficulty in accurately quantifying defects.

[0005] In recent years, data-driven intelligent analysis methods, such as deep learning and machine learning, have been widely studied in the field of additive manufacturing defect detection. Usually, a neural network model is trained based on a large amount of experimental data to achieve automatic defect identification. However, the main defects of such methods include: First, they are highly dependent on high-quality training data and it is difficult to generalize to the additive manufacturing process under different materials and process parameters; Second, the computational complexity is high and it is difficult to meet the real-time requirements; Third, there is a lack of in-depth understanding of the defect formation mechanism and it is difficult to conduct effective early warning and process optimization.

[0006] In summary, the existing defect detection technologies in the additive manufacturing process still have significant deficiencies in terms of real-time performance, accuracy, anti-interference ability, and process controllability, and it is difficult to meet the stringent requirements for product quality and consistency in the high-end manufacturing field. Therefore, there is an urgent need for a more efficient and accurate defect identification and early warning method to improve the reliability of the additive manufacturing process and product performance. Summary of the Invention

[0007] An object of the present invention is to propose a defect identification and early warning system in the additive manufacturing process. The present invention reduces the professionalism and technical threshold required for traditional modeling and improves the universality and convenience of 3D modeling.

[0008] A defect identification and early warning system in the additive manufacturing process according to an embodiment of the present invention includes:

[0009] A molten pool image data acquisition module, configured to obtain visible light molten pool image data and temperature field molten pool image data during the metal additive manufacturing process, and form a time series molten pool image data matrix;

[0010] A molten pool image data preprocessing module, configured to preprocess the time series molten pool image data matrix to form a preprocessed molten pool image data matrix;

[0011] A Harris hawk algorithm defect area location module, configured to perform a global search based on the preprocessed molten pool image data matrix using an improved Harris hawk algorithm to determine a microcrack defect candidate area, and output a defect candidate area molten pool image data matrix;

[0012] A fractal segmentation algorithm microcrack defect fine identification module, configured to calculate the local gray-scale difference feature of the microcrack defect candidate area based on the defect candidate area molten pool image data matrix, and perform local threshold segmentation on the microcrack defect candidate area to obtain a finely segmented microcrack defect area molten pool image data matrix;

[0013] A microcrack defect feature extraction and quantitative evaluation module, configured to extract the geometric features and morphological features of the microcrack based on the finely segmented microcrack defect area molten pool image data matrix to form a microcrack defect quantitative evaluation result;

[0014] A microcrack defect early warning module, configured to trigger a real-time early warning feedback mechanism based on the microcrack defect quantitative evaluation result.

[0015] A defect identification and early warning method in the additive manufacturing process, applied to a defect identification and early warning system in the additive manufacturing process, includes the following steps:

[0016] S1. During the metal additive manufacturing process, use an imaging device to monitor the molten pool in real time and obtain a continuous time-series molten pool image data matrix;

[0017] S2. Preprocess the time-series molten pool image data matrix to obtain a preprocessed molten pool image data matrix;

[0018] S3. Use the Harris hawks algorithm to perform a global search on the preprocessed molten pool image data matrix, automatically locate the candidate regions of microcrack defects in the molten pool image data matrix, and form a molten pool image data matrix of the defect candidate regions;

[0019] S4. Apply the fractal segmentation algorithm to analyze the fractal features of the microcrack defects in the molten pool image data of the defect candidate regions, and obtain a molten pool image data matrix of the finely segmented microcrack defect regions;

[0020] S5. Extract the geometric features and morphological features of the microcracks based on the molten pool image data matrix of the finely segmented microcrack defect regions to form a quantitative evaluation result of the microcrack defects;

[0021] S6. Based on the quantitative evaluation result of the microcrack defects, trigger a real-time early warning feedback mechanism.

[0022] Optionally, step S1 specifically includes the following steps:

[0023] S11. During the metal additive manufacturing process, use an imaging device to monitor the molten pool in real time. The imaging device includes a high-speed camera and an infrared thermal imager. The high-speed camera is used to obtain the visible light molten pool image data of the molten pool area, and the infrared thermal imager is used to collect the temperature field molten pool image data of the molten pool area, and synchronously store the obtained visible light molten pool image data and temperature field molten pool image data to form multi-modal molten pool image data;

[0024] S12. Perform data fusion on the collected multi-modal molten pool image data to construct a time-series molten pool image data matrix :

[0025]

[0026] Among them, is the time-series molten pool image data matrix, is the visible light molten pool image data matrix, is the temperature field molten pool image data matrix, represents the time-series frame index, represents the image space index.

[0027] Optionally, step S2 specifically includes the following steps:

[0028] S21. Perform image noise removal processing on the time - series molten pool image data matrix and use median filtering to denoise each frame of image data in the time - series molten pool image data matrix , eliminate the random noise interference caused by high - temperature gasification in the molten pool environment, and obtain the preliminarily denoised molten pool image data matrix ;

[0029] S22. Perform gray - scale normalization processing on the preliminarily denoised molten pool image data matrix to map the gray - scale values of each frame of molten pool image data in the molten pool image data matrix to a unified interval [0, 1], and obtain the gray - scale normalized molten pool image data matrix ;

[0030] S23. Perform contrast enhancement processing on each frame of the molten pool image in the gray - scale normalized molten pool image data matrix , enhance the gray - scale difference between the micro - crack region and the molten pool background region, highlight the characteristics of the crack region, and form the contrast - enhanced molten pool image data matrix ;

[0031] S24. Apply edge filtering processing to each frame of the molten pool image in the contrast - enhanced molten pool image data matrix . Use an edge - preserving Gaussian bilateral filter for filtering to enhance the clarity of the crack region boundary, and form the pre - processed molten pool image data matrix .

[0032] Optionally, the specific steps of S3 are as follows:

[0033] S31. During the metal additive manufacturing process, initialize the position matrix of the Harris hawks algorithm based on the pre - processed molten pool image data matrix , and the initial position of each individual in the hawk group in the molten pool image data matrix is represented as:

[0034]

[0035] where represents the initial position coordinates of the th individual in the hawk group in the molten pool image data matrix, is the total number of individuals in the hawk group, is the two - dimensional spatial coordinates of the kth individual in the hawk group, representing the specific horizontal and vertical coordinates of the individual in the molten pool image data matrix;

[0036] S32. Establish the fitness function of the hawk group individuals. The fitness function of the hawk group individuals is calculated based on the gray - scale difference between the current position of the hawk group individuals and the average gray - scale value of the molten pool image background in their neighborhood:

[0037]

[0038] Among them, is the fitness function value of the k-th individual in the eagle group at the t-th iteration, is for the th iteration and the th position coordinate of the individual in the eagle group, is a neighborhood window set with the position coordinate of the individual in the eagle group as the center, is the molten pool image data matrix after preprocessing at the position coordinate of the individual in the eagle group, is the neighborhood window and the total number of pixels within it, is the molten pool image data matrix after preprocessing within the neighborhood window at the pixel point ;

[0039] S33. Construct the memory weight of the individual in the eagle group according to the historical search information of the eagle group , and update the position of the individual in the eagle group:

[0040]

[0041] Among them, is for the th iteration and the th position coordinate of the individual in the eagle group, is the memory weight of the individual in the eagle group:

[0042]

[0043] Among them, is the fitness function value of the i-th individual in the eagle group at the t-th iteration;

[0044] The memory weight of the individual in the eagle group is determined according to the difference between the current fitness function value of the individual in the eagle group and the historical fitness function value of the individual in the eagle group, so that the position of the individual in the eagle group moves towards the region with a higher gray level difference;

[0045] S34. Construct an automatic judgment index for the search strategy of the eagle group :

[0046]

[0047] Among them, is the automatic judgment index for the search strategy of the eagle group, is the highest individual fitness function value of the eagle group, is the total number of individuals in the swarm, when the swarm search strategy automatically determines the index Less than the preset threshold When , the eagle group search automatically switches to the local refinement mode, otherwise it switches to the global exploration mode;

[0048] S35. In the global exploration mode, the individual positions of the eagle group are updated as follows:

[0049]

[0050] in, is the next iteration position of the eagle group individuals in the global exploration mode, is the current optimal individual position of the eagle group, represents the position of an individual eagle group randomly selected from the current search space at the t + 1th iteration, and is a random factor in the interval [0,1];

[0051] S36. In the local fine mode, the individual positions of the eagle group are constrained by the temperature field of the molten pool. When the gray value of the temperature field molten pool image data corresponding to the individual position of the eagle group Above the temperature abnormality threshold When , the individual positions of the eagle group are updated to:

[0052]

[0053] in, is the next iteration position of the eagle group individuals in the local exploration mode, is the gray value at the current position coordinate of the eagle group individual in the temperature field melt pool image data matrix, representing the temperature field intensity of the current position coordinate. is the abnormal temperature threshold of the molten pool, is the temperature field constraint coefficient;

[0054] S37. After the eagle group search is completed, select the optimal eagle group individual position of the final iteration As the center position of the microcrack defect candidate area, a fixed-size molten pool image area is intercepted with the center position of the microcrack defect candidate area as the center to form a defect candidate area molten pool image data matrix .

[0055] Optionally, the S4 specifically includes the following steps:

[0056] S41, based on the defect candidate area melt pool image data matrix Calculate the local grayscale difference characteristics of each defect candidate area in the defect candidate area molten pool image data matrix :

[0057]

[0058] Among them, represents the local gray-scale difference feature of the defect candidate region centered on the position coordinates of the defect candidate region is the gray-scale value at the position coordinates of the defect candidate region in the defect candidate region molten pool image data matrix represents the overall average gray-scale value of the defect candidate region is the gray-scale value at the position coordinates of the defect candidate region is the overall average gray-scale value of the defect candidate region are the number of rows and columns of the local window respectively;

[0059] S42. Determine the adaptive fractal scale for the local gray-scale difference feature The adaptive fractal scale is dynamically adjusted according to the local gray-scale difference feature:

[0060]

[0061] Among them, is the adaptive fractal scale of the defect candidate region at the position coordinates of the defect candidate region is the standard fractal box size set initially is the adaptive adjustment coefficient; is the adaptive adjustment coefficient;

[0062] S43. Calculate the local fractal dimension of the defect candidate region molten pool image data matrix based on the adaptive fractal scale : :

[0063]

[0064] Among them, represents the local fractal dimension at the position coordinates of the defect candidate region is the local fractal dimension at the position coordinates of the defect candidate region represents the minimum number of boxes required to cover the position coordinates of the defect candidate region at the fractal scale ;

[0065] S44. Determine the adaptive segmentation threshold of the microcrack defect region according to the local fractal dimension and the local gray-scale difference feature :

[0066]

[0067] Among them, is the position coordinates of the defect candidate region Adaptive segmentation threshold for the microcrack defect area and is the weight coefficient;

[0068] S45. Adopt the adaptive segmentation threshold of the microcrack defect area to perform local threshold binary segmentation on the data matrix of the molten pool image in the defect candidate area to obtain the data matrix of the molten pool image in the microcrack defect area after fine segmentation :

[0069] .

[0070] Optionally, the specific steps of S5 are as follows:

[0071] S51. Based on the data matrix of the molten pool image in the microcrack defect area after fine segmentation construct a microcrack boundary detection operator to extract the boundary contour point set of the microcrack area :

[0072]

[0073] wherein, represents the boundary contour point set of the microcrack defect area, represents the data matrix of the molten pool image in the microcrack defect area after fine segmentation gradient, is the edge detection threshold;

[0074] S52. Calculate the length of the microcrack defect , use the minimum circumscribed rectangle method to fit the boundary points of the microcrack area, and define the microcrack length as the major axis of the minimum circumscribed rectangle:

[0075]

[0076] wherein, represents the length of the microcrack defect, and are any two points in the microcrack boundary contour respectively, represents the Euclidean distance between the two points;

[0077] S53. Use the normal projection method to calculate the average width in the direction of the microcrack center line:

[0078]

[0079] wherein, represents the average width of the microcrack defect, The coordinates of the positive boundary point where the g-th sampling point extends outward along the normal direction to the crack edge, The coordinates of the negative boundary point where the g-th sampling point contracts inward along the normal direction to the crack edge, is the total number of sampling points on the center line of the microcrack;

[0080] S54. Based on the molten pool image data matrix of the microcrack defect area after fine segmentation Extract the crack edge contour of the microcrack defect area to determine the crack morphology characteristics , and the crack morphology characteristics are described by the crack edge contour curvature characteristics:

[0081]

[0082] Among them, is the crack morphology characteristic, reflecting the bending degree of the crack area contour, , respectively represent the first-order derivatives in the horizontal and vertical directions of the -th edge contour point, , respectively represent the second-order derivatives in the horizontal and vertical directions of the -th edge contour point, represents the total number of points sampled on the crack edge contour;

[0083] S55. Based on the length , width of the microcrack and the crack morphology characteristic , calculate the severity score of the microcrack defect:

[0084]

[0085] Among them, , and are weight coefficients;

[0086] S56. Set the microcrack defect level according to the severity score of the microcrack defect. The basis for defect level division is:

[0087] Defect level division

[0088] Among them, and are respectively the low-level and high-level severity thresholds of microcrack defects in additive manufacturing, forming a quantitative evaluation result of microcrack defects in additive manufacturing.

[0089] The beneficial effects of the present invention are:

[0090] (1) The present invention introduces an adaptive memory weight and temperature field constraint mechanism to make the search for defect areas more accurate. By constructing the fitness function of the eagle group individuals and combining the temperature field molten pool data, the search mode is dynamically adjusted to ensure that the eagle group individuals can focus on the areas with high gray-scale differences, thereby improving the accuracy of micro-crack detection.

[0091] (2) For the detection of micro-crack defects in the additive manufacturing process, the present invention adopts a fractal segmentation algorithm. By calculating the local gray-scale difference features and the fractal dimension, the crack segmentation threshold is adaptively adjusted to achieve the accurate extraction of micro-cracks. It can adaptively adjust the segmentation strategy under different complex backgrounds, and is particularly suitable for scenarios where the size of micro-cracks varies greatly. In addition, the combination of the fractal segmentation algorithm and the adaptive scale adjustment makes the extraction of the crack boundary smoother, and can effectively reduce the interference of pseudo-cracks.

[0092] (3) Based on the identification of micro-crack defects, the present invention establishes an intelligent early warning system based on the defect severity score, and combines it with the process parameter optimization model to achieve the dynamic control of defect formation. Based on the quantitative evaluation results of the defects, the key process parameters such as laser power and scanning speed are dynamically adjusted, thereby reducing the probability of defect generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0094] Figure 1 is a flowchart of a defect identification and early warning system in an additive manufacturing process proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0095] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0096] Refer to Figure 1 , a defect identification and early warning system in an additive manufacturing process, comprising:

[0097] A molten pool image data acquisition module, used to acquire visible light molten pool image data and temperature field molten pool image data during the metal additive manufacturing process, and form a time series molten pool image data matrix;

[0098] A molten pool image data preprocessing module, used to preprocess the time series molten pool image data matrix to form a preprocessed molten pool image data matrix;

[0099] Harris Hawk Algorithm Defect Region Location Module, which is used to perform global search using the improved Harris Hawk algorithm based on the preprocessed molten pool image data matrix, determine the candidate regions of microcrack defects, and output the molten pool image data matrix of the defect candidate regions;

[0100] Fractal Segmentation Algorithm Microcrack Defect Fine Recognition Module, which is used to calculate the local gray difference features of the microcrack defect candidate regions based on the molten pool image data matrix of the defect candidate regions, and perform local threshold segmentation on the microcrack defect candidate regions to obtain the molten pool image data matrix of the finely segmented microcrack defect regions;

[0101] Microcrack Defect Feature Extraction and Quantitative Evaluation Module, which is used to extract the geometric features and morphological features of the microcracks based on the molten pool image data matrix of the finely segmented microcrack defect regions, and form the quantitative evaluation results of the microcrack defects;

[0102] Microcrack Defect Early Warning Module, which is used to trigger the real-time warning feedback mechanism based on the quantitative evaluation results of the microcrack defects.

[0103] A method for defect identification and early warning in the additive manufacturing process, applied to a defect identification and early warning system in the additive manufacturing process, includes the following steps:

[0104] S1. During the metal additive manufacturing process, use an imaging device to monitor the molten pool in real time and obtain the molten pool image data matrix of continuous time series;

[0105] S2. Preprocess the time series molten pool image data matrix to obtain the preprocessed molten pool image data matrix;

[0106] S3. Use the Harris Hawk algorithm to perform global search on the preprocessed molten pool image data matrix, automatically locate the candidate regions of microcrack defects in the molten pool image data matrix, and form the molten pool image data matrix of the defect candidate regions;

[0107] S4. Apply the fractal segmentation algorithm to analyze the fractal features of the microcracks in the molten pool image data of the defect candidate regions to obtain the molten pool image data matrix of the finely segmented microcrack defect regions;

[0108] S5. Extract the geometric features and morphological features of the microcracks based on the molten pool image data matrix of the finely segmented microcrack defect regions to form the quantitative evaluation results of the microcrack defects;

[0109] S6. Based on the quantitative evaluation results of microcrack defects, if the severity score exceeds the set threshold, different levels of warnings are triggered, including visual alarms, production parameter adjustment suggestions, and automated control strategies. Combining the molten pool temperature field data, analyze the main influencing factors of crack generation, and optimize the additive manufacturing process parameters, including laser power, scanning speed, and powder feeding rate. The system dynamically adjusts the process parameters based on real-time data feedback to inhibit crack propagation and improve the forming quality and stability.

[0110] In this embodiment, S1 specifically includes the following steps:

[0111] S11. During the metal additive manufacturing process, use an imaging device to monitor the molten pool in real time. The imaging device includes a high-speed camera and an infrared thermal imager. The high-speed camera is used to obtain visible light molten pool image data of the molten pool area, and the infrared thermal imager is used to collect temperature field molten pool image data of the molten pool area, and synchronously store the obtained visible light molten pool image data and temperature field molten pool image data to form multi-modal molten pool image data;

[0112] S12. Perform data fusion on the collected multi-modal molten pool image data to construct a time-series molten pool image data matrix :

[0113]

[0114] Among them, is the time-series molten pool image data matrix, is the visible light molten pool image data matrix, is the temperature field molten pool image data matrix, represents the time-series frame index, represents the image space index.

[0115] In this embodiment, a high-speed camera and an infrared thermal imager are used in the acquisition of molten pool image data to achieve multi-modal fusion of visible light information and temperature field information of the molten pool, improve data integrity and accuracy. The time-series synchronous processing ensures the continuity and traceability of the data. The adaptive sampling strategy optimizes data storage, improves the spatio-temporal resolution, provides high-quality input data for subsequent microcrack defect detection, enhances the robustness and real-time performance of the detection system, effectively reduces false detections and missed detections caused by environmental changes, and improves the quality control ability in the additive manufacturing process.

[0116] In this embodiment, S2 specifically includes the following steps:

[0117] S21. Perform image noise removal processing on the time-series molten pool image data matrix using median filtering on the time-series molten pool image data matrix Denoise each frame of image data to eliminate the random noise interference caused by high-temperature gasification in the molten pool environment, and obtain the molten pool image data matrix after preliminary denoising. ;

[0118] S22. Perform gray normalization on the molten pool image data matrix after preliminary denoising so that the gray value of each frame of molten pool image data in the molten pool image data matrix is mapped to the unified interval [0, 1], and obtain the molten pool image data matrix after gray normalization ;

[0119] S23. Perform contrast enhancement on each frame of molten pool image in the molten pool image data matrix after gray normalization to enhance the gray difference between the microcrack region and the molten pool background region, highlight the characteristics of the crack region, and form the molten pool image data matrix after contrast enhancement ;

[0120] S24. Apply edge filtering to each frame of molten pool image in the molten pool image data matrix after contrast enhancement The filtering uses an edge-preserving Gaussian bilateral filter to enhance the clarity of the crack region boundary, and form the preprocessed molten pool image data matrix .

[0121] The molten pool image data preprocessing method proposed in this embodiment improves the image clarity, enhances the contrast and recognizability of the microcrack region through noise removal, gray normalization, contrast enhancement and edge filtering. The adaptive edge filtering technology can effectively suppress background noise while maintaining the crack boundary details, and optimize the input quality of the subsequent algorithm. The storage structure of the time series molten pool image data matrix ensures data stability, improves the dynamic crack detection accuracy, lays a foundation for the precise positioning and morphological analysis of microcracks, and enhances the overall performance of the detection system.

[0122] In this embodiment, S3 specifically includes the following steps:

[0123] S31. During the metal additive manufacturing process, initialize the hawk swarm position matrix of the Harris hawk algorithm based on the preprocessed molten pool image data matrix , and the initial position of each hawk swarm individual in the molten pool image data matrix is expressed as:

[0124]

[0125] where represents the initial position coordinates of the th hawk swarm individual in the molten pool image data matrix, is the total number of hawk swarm individuals, ​is the two-dimensional spatial coordinate of the k-th individual in the eagle group, representing the specific horizontal and vertical coordinates of the individual in the molten pool image data matrix;

[0126] S32. Establish the fitness function of the eagle group individuals. The fitness function of the eagle group individuals is calculated based on the gray-scale difference between the current position of the eagle group individuals and the average gray-scale value of the molten pool image background in their neighborhood:

[0127]

[0128] where, is the fitness function value of the k-th individual in the eagle group at the t-th iteration, is at the -th iteration, the -th position coordinate of the k-th individual in the eagle group, is a neighborhood window of a set size centered on the position coordinate of the eagle group individual, is the preprocessed molten pool image data matrix at the position coordinate of the eagle group individual in the matrix, is the neighborhood window the total number of pixels within, is the preprocessed molten pool image data matrix in the neighborhood window at the pixel point in the matrix;

[0129] is used to calculate the fitness of the eagle group individuals in the search process for the microcrack defect area, that is, to measure the difference degree between the pixel gray-scale value at the current position of the individual and the average gray-scale value of its neighborhood area. A larger fitness value indicates that this position may be located in the microcrack area, while a smaller fitness value indicates that it is in the background area. Through the formula, the Harris hawk algorithm can preferentially select the area with a higher fitness as the potential microcrack defect area during the search process, improving the accuracy and stability of defect detection.

[0130] S33. Construct the memory weight of the eagle group individuals according to the historical search information of the eagle group and update the position of the eagle group individuals:

[0131]

[0132] where, is at the -th iteration, the -th position coordinate of the k-th individual in the eagle group, is the memory weight of the eagle group individuals:

[0133]

[0134] Among them, is the fitness function value of the i-th individual in the eagle group at the t-th iteration;

[0135] The memory weight of the individual in the eagle group is determined according to the difference between the current fitness function value of the individual in the eagle group and the historical fitness function value of the individual in the eagle group, so that the position of the individual in the eagle group moves towards the area with a higher gray-scale difference;

[0136] S34. Construct an automatic judgment index for the eagle group search strategy :

[0137]

[0138] Among them, is the automatic judgment index for the eagle group search strategy, is the highest fitness function value of the individual in the eagle group, is the total number of individuals in the eagle group. When the automatic judgment index for the eagle group search strategy is less than the preset threshold the eagle group search automatically switches to the local fine mode; otherwise, it is the global exploration mode;

[0139] S35. In the global exploration mode, the position update method of the individual in the eagle group is:

[0140]

[0141] Among them, is the position of the individual in the eagle group at the next iteration in the global exploration mode, is the current optimal position of the individual in the eagle group, represents a position of an individual in the eagle group randomly selected from the current search space at the (t + 1)-th iteration, and are random factors within the interval [0, 1];

[0142] Under the action of the formula, the individual in the eagle group can explore the microcrack defect area globally, improving the adaptability and robustness of the Harris hawk algorithm in the complex environment of the molten pool in metal additive manufacturing.

[0143] S36. In the local fine mode, the position of the individual in the eagle group is restricted by the temperature field of the molten pool. When the gray-scale value of the temperature field molten pool image data corresponding to the position of the individual in the eagle group is higher than the temperature anomaly threshold the position update of the individual in the eagle group is:

[0144]

[0145] Among them, is the position of the individual in the eagle group in the next iteration under the local exploration mode, is the grayscale value at the current position coordinate of the individual in the eagle group in the temperature field molten pool image data matrix, representing the temperature field intensity at the current position coordinate, is the abnormal threshold of the molten pool temperature, is the temperature field constraint coefficient;

[0146] By introducing temperature field data, the influence of the molten pool environment on the identification of microcracks is strengthened, making the eagle group search more accurate, effectively reducing false detections and missed detections, and improving the quality monitoring ability in the metal additive manufacturing process.

[0147] S37. After the eagle group search is completed, select the position of the optimal eagle group individual in the final iteration as the central position of the microcrack defect candidate area, and intercept a molten pool image area with a fixed size centered on the central position of the microcrack defect candidate area to form a molten pool image data matrix of the defect candidate area .

[0148] This embodiment uses an improved Harris hawk algorithm for automatic positioning of microcrack defects, constructs an adaptive fitness function based on molten pool image features and temperature field data to improve the search accuracy for defect areas, proposes an eagle group search historical memory weight and temperature field constraint strategy, enables the algorithm to have stronger global exploration and local optimization capabilities in a complex molten pool environment, and a dynamic search mode switching mechanism improves the search efficiency, avoids falling into local optima, and improves the stability and accuracy of microcrack detection.

[0149] In this embodiment, S4 specifically includes the following steps:

[0150] S41. Based on the molten pool image data matrix of the defect candidate area Calculate the local grayscale difference feature of each defect candidate area in the molten pool image data matrix of the defect candidate area :

[0151]

[0152] Among them, represents the local grayscale difference feature of the defect candidate area centered on the position coordinate of the defect candidate area, represents the grayscale value at the position coordinate of the defect candidate area in the molten pool image data matrix of the defect candidate area, is the overall average grayscale value of the defect candidate area, are the number of rows and columns of the local window respectively;

[0153] The local grayscale difference feature formula is used to calculate the local grayscale difference feature of each pixel point in the defect candidate region. The local grayscale difference reflects the grayscale change of the molten pool image in the microcrack region, which helps to distinguish the crack region from the normal molten pool region.

[0154] S42. Determine the adaptive fractal scale for the local grayscale difference feature The adaptive fractal scale is dynamically adjusted according to the local grayscale difference feature:

[0155]

[0156] where is the adaptive fractal scale at the position coordinates of the defect candidate region in the defect candidate region , is the initially set standard fractal box size, is the adaptive adjustment coefficient;

[0157] S43. Calculate the local fractal dimension of the molten pool image data matrix of the defect candidate region based on the adaptive fractal scale :

[0158]

[0159] where represents the local fractal dimension at the position coordinates of the defect candidate region, represents the minimum number of boxes required to cover the position coordinates of the defect candidate region at the fractal scale ;

[0160] S44. Determine the adaptive segmentation threshold of the microcrack defect region according to the local fractal dimension and the local grayscale difference feature :

[0161]

[0162] where is the adaptive segmentation threshold of the microcrack defect region at the position coordinates of the defect candidate region, , are the weight coefficients;

[0163] S45. Use the adaptive segmentation threshold of the microcrack defect region to process the molten pool image data matrix of the defect candidate region ​​Perform local threshold binarization segmentation to obtain the molten pool image data matrix of the microcrack defect area after fine segmentation :

[0164] 。

[0165] In this embodiment, an improved adaptive fractal segmentation algorithm is used to accurately segment the defect candidate area. An adaptive fractal scale calculation method driven by local gray difference features is proposed, making the calculation of the fractal dimension more targeted, improving the fine-grained extraction ability of the crack area. The adaptive threshold segmentation strategy combines the local fractal dimension and gray features to improve the recognition accuracy of the crack area, effectively distinguish the molten pool texture from the real microcracks, and improve the accuracy of microcrack segmentation.

[0166] In this embodiment, S5 specifically includes the following steps:

[0167] S51. Based on the molten pool image data matrix of the microcrack defect area after fine segmentation Construct a microcrack boundary detection operator to extract the boundary contour point set of the microcrack area :

[0168]

[0169] Among them, represents the boundary contour point set of the microcrack defect area, represents the molten pool image data matrix of the microcrack defect area after fine segmentation gradient, is the edge detection threshold;

[0170] S52. Calculate the length of the microcrack defect , use the minimum circumscribed rectangle method to fit the boundary points of the microcrack area, and define the microcrack length as the major axis of the minimum circumscribed rectangle:

[0171]

[0172] Among them, represents the length of the microcrack defect, and are any two points in the microcrack boundary contour respectively, represents the Euclidean distance between two points;

[0173] S53. Use the normal projection method to calculate the average width in the direction of the microcrack center line :

[0174]

[0175] Among them, represents the average width of microcrack defects, is the coordinate of the positive boundary point extending outward from the g-th sampling point along the normal direction to the crack edge, The g-th sampling point shrinks inward along the normal direction to the negative boundary point coordinate of the crack edge. is the total number of sampling points on the center line of the microcrack;

[0176] S54, based on the molten pool image data matrix of the microcrack defect area after fine segmentation Extract the crack edge contour of the microcrack defect area and determine the crack morphology characteristics , the crack morphology is described by the crack edge contour curvature characteristics:

[0177]

[0178] in, is the crack morphology characteristic, reflecting the degree of curvature of the crack area contour. , Respectively represent The first-order derivatives in the horizontal and vertical directions of the edge contour points, , Respectively represent The second-order derivatives in the horizontal and vertical directions of the edge contour points, Represents the total number of points sampled on the crack edge contour;

[0179] In practical applications, the higher the crack curvature, the more dramatic the change in the crack edge, which may be a microcrack caused by rapid growth or molten pool fluctuation. The value can be used for crack severity scoring, assisting in process parameter optimization, and preventing further crack propagation.

[0180] S55, based on the length of microcracks ,width and crack morphology characteristics , calculate the severity score of microcrack defects :

[0181]

[0182] in, , and is the weight coefficient;

[0183] S56. Score based on the severity of microcrack defects Set the microcrack defect level, and the defect level is divided according to:

[0184] Defect grade classification

[0185] Among them, and are the low - level and high - level severity thresholds of micro - crack defects in additive manufacturing respectively, forming the quantitative evaluation result of micro - crack defects in additive manufacturing.

[0186] In this embodiment, for the finely segmented micro - crack region, a multi - feature fusion method is adopted for defect quantification analysis. The crack length and width are accurately calculated through boundary contour extraction, minimum circumscribed rectangle fitting, and normal projection methods. At the same time, combined with the curvature feature of the crack edge contour, the ability to describe the defect morphology is enhanced, a comprehensive severity scoring function is constructed, and the crack defects are classified and evaluated to improve the scientificity and accuracy of micro - crack severity determination. The defect quantitative analysis results can be used for real - time warning and process optimization, improving the intelligent quality control ability of the metal additive manufacturing process.

[0187] Example 1:

[0188] During the night shift production on March 15, 2024, in the additive manufacturing workshop of an aviation manufacturing enterprise, the selective laser melting technology was being used to print Inconel718 nickel - based alloy turbine blades. The operator started the printing task at 22:37, and the set process parameters were laser power 400W, scanning speed 800mm / s, layer thickness 30μm, and scanning spacing 0.1mm. Since the turbine blade is a key component of an aero - engine and has extremely high mechanical property requirements, any tiny micro - crack will lead to part rejection and even affect flight safety.

[0189] When the printing reached the 134th layer (time: 23:02), the defect recognition and warning system of the present invention began to detect abnormal signals. The system monitored abnormal gray - scale changes (local gray - scale difference reached 42.6%) and temperature distribution deviation (local temperature gradient increased abnormally by 6.2%) in the molten pool area, which are the precursors of micro - crack formation.

[0190] The real - time data records are as follows:

[0191] Layer number: 134 layers; Abnormal area coordinates: (X: 23.7mm, Y: 18.4mm); Abnormal gray - scale change: Enhanced by 42.6% (normal range is 20% - 30%); Abnormal temperature value: Local temperature gradient increased by 6.2%; Abnormal duration: 3.7 seconds; Peak value of the Harris hawk optimization algorithm (HHO) fitness function: 0.91 (exceeding the defect threshold of 0.85).

[0192] The system automatically triggered the Harris hawk optimization algorithm (HHO) at 23:02:14 for fine - grained search of the defect area and confirmed the existence of potential micro - crack risks in this area. To avoid further expansion of the defect, the system automatically adjusted the process parameters according to the defect prediction model:

[0193] The laser power is reduced by 3% (from 400 W to 388 W); the scanning speed is increased by 50 mm / s (from 800 mm / s to 850 mm / s); the scanning strategy is optimized to reduce the number of remelting times and the direction of the laser path is adjusted.

[0194] After adjustment, the system continuously monitors this area, and the results are as follows:

[0195] Layer 136 (Time: 23:07): The abnormal gray-scale change is reduced to 28.4% (returning to the normal range); Layer 138 (Time: 23:12): The temperature gradient is restored to the standard value (±1.3%); the fitness function of the eagle group is reduced to 0.79 (lower than the defect threshold).

[0196] After printing is completed, X-ray computed tomography (XCT) detection is carried out at 03:42, and by comparing with the samples produced using traditional fixed process parameters, the following differences are found:

[0197]

[0198] In addition, XCT detection finds that the fatigue life of the turbine blades produced by the method of the present invention is increased by about 27% in the high-cycle fatigue test, which proves the effectiveness of the defect identification and early warning system.

[0199] The method of the present invention can identify the abnormal conditions of the molten pool in the additive manufacturing process in real time, quickly adjust the process parameters, and reduce the defect occurrence rate. Compared with the traditional method, the defect detection rate is higher, and the process can be optimized and adjusted within 1.2 seconds, effectively reducing the generation of microcracks and improving the part quality and production stability.

[0200] The present invention introduces an adaptive memory weight and a temperature field constraint mechanism to make the search for defect areas more accurate. By constructing the fitness function of the eagle group individuals and combining the temperature field molten pool data, the search mode is dynamically adjusted to ensure that the eagle group individuals can focus on the areas with high gray-scale differences, improving the accuracy of microcrack detection.

[0201] For the detection of microcrack defects in the additive manufacturing process, the present invention adopts a fractal segmentation algorithm. By calculating the local gray-scale difference features and the fractal dimension, the crack segmentation threshold is adaptively adjusted to achieve the accurate extraction of microcracks. It can adaptively adjust the segmentation strategy under different complex backgrounds, and is especially suitable for scenarios where the size of microcracks changes greatly. In addition, the combination of the fractal segmentation algorithm and the adaptive scale adjustment makes the extraction of the crack boundary smoother, effectively reducing the interference of pseudo-cracks.

[0202] Based on the identification of microcrack defects, the present invention establishes an intelligent early warning system based on defect severity scoring, and combines a process parameter optimization model to achieve dynamic control of defect formation. Based on the quantitative evaluation results of defects, key process parameters such as laser power and scanning speed are dynamically adjusted, thereby reducing the probability of defect generation.

[0203] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A defect recognition and early warning system in an additive manufacturing process, characterized in that: include: A molten pool image data acquisition module is used to obtain visible light molten pool image data and temperature field molten pool image data in the metal additive manufacturing process to form a time series molten pool image data matrix; A molten pool image data preprocessing module is used to preprocess the time series molten pool image data matrix to form a preprocessed molten pool image data matrix; Harris Eagle algorithm defect area positioning module, which is used to perform global search based on the preprocessed molten pool image data matrix using the improved Harris Eagle algorithm to determine the candidate areas of microcrack defects and output the molten pool image data matrix of the defect candidate areas; The micro-crack defect fine recognition module of the fractal segmentation algorithm is used to calculate the local grayscale difference characteristics of the micro-crack defect candidate area based on the molten pool image data matrix of the defect candidate area, and perform local threshold segmentation on the micro-crack defect candidate area to obtain the molten pool image data matrix of the micro-crack defect area after fine segmentation; The microcrack defect feature extraction and quantitative evaluation module is used to extract the geometric and morphological features of microcracks based on the finely segmented molten pool image data matrix of the microcrack defect area to form a quantitative evaluation result of microcrack defects; The microcrack defect early warning module is used to trigger a real-time early warning feedback mechanism based on the quantitative evaluation results of microcrack defects.

2. A defect recognition and early warning method in an additive manufacturing process, applied to a defect recognition and early warning system in an additive manufacturing process as claimed in claim 1, characterized in that: The following steps are involved: S1. During the metal additive manufacturing process, the molten pool is monitored in real time using imaging equipment to obtain a continuous time series molten pool image data matrix; S2, preprocessing the time series molten pool image data matrix to obtain a preprocessed molten pool image data matrix; S3, using the Harris Eagle algorithm to perform a global search on the preprocessed molten pool image data matrix, automatically locate the candidate area of ​​microcrack defects in the molten pool image data matrix, and form a molten pool image data matrix of defect candidate areas; S4, applying a fractal segmentation algorithm to analyze the fractal characteristics of microcrack defects in the molten pool image data of the defect candidate area, and obtaining a molten pool image data matrix of the microcrack defect area after fine segmentation; S5. Extracting geometric and morphological features of microcracks based on the finely segmented molten pool image data matrix of the microcrack defect area to form a quantitative evaluation result of microcrack defects; S6. Based on the quantitative evaluation results of microcrack defects, a real-time early warning feedback mechanism is triggered.

3. A defect identification and early warning method in an additive manufacturing process according to claim 2, characterized in that: The S1 specifically includes the following steps: S11. During the metal additive manufacturing process, the molten pool is monitored in real time using an imaging device, wherein the imaging device includes a high-speed camera and an infrared thermal imager, wherein the high-speed camera is used to obtain visible light molten pool image data of the molten pool area, and the infrared thermal imager is used to collect temperature field molten pool image data of the molten pool area, and the obtained visible light molten pool image data and temperature field molten pool image data are synchronously stored to form multimodal molten pool image data; S12, fusing the collected multi-modal melt pool image data to construct a time series melt pool image data matrix : in, is the time series melt pool image data matrix, is the visible light melt pool image data matrix, is the temperature field melt pool image data matrix, represents the time series frame index, Represents the image space index.

4. The defect identification and early warning method in the additive manufacturing process according to claim 3, characterized in that: The S2 specifically includes the following steps: S21, the time series molten pool image data matrix Image noise removal is performed, and median filtering is used to process the time series melt pool image data matrix The image data of each frame is denoised to eliminate the random noise interference caused by high temperature gasification in the molten pool environment, and the molten pool image data matrix after preliminary denoising is obtained. ; S22, the molten pool image data matrix after the preliminary denoising Grayscale normalization is performed to map the grayscale value of each frame of the molten pool image data in the molten pool image data matrix to a uniform interval [0,1], and the grayscale normalized molten pool image data matrix is ​​obtained. ; S23, the molten pool image data matrix after grayscale normalization The contrast of each frame of the molten pool image is enhanced to enhance the grayscale difference between the microcrack area and the molten pool background area, highlight the characteristics of the crack area, and form a contrast-enhanced molten pool image data matrix ; S24, contrast-enhanced melt pool image data matrix The edge filtering is applied to each frame of the molten pool image. The edge-preserving Gaussian bilateral filter is used to enhance the clarity of the crack area boundary to form a preprocessed molten pool image data matrix. .

5. A defect identification and early warning method in an additive manufacturing process according to claim 4, characterized in that: The S3 specifically includes the following steps: S31. In the metal additive manufacturing process, based on the pre-processed melt pool image data matrix Initialize the hawk group position matrix of the Harris Hawk algorithm , the initial position of the eagle group individuals in the melt pool image data matrix is ​​expressed as: in, Indicates The initial position coordinates of the eagle group individuals in the melt pool image data matrix, is the total number of eagle individuals, is the two-dimensional spatial coordinate of the kth eagle group individual, indicating the specific horizontal and vertical coordinates of the individual in the melt pool image data matrix; S32, establish the individual fitness function of the eagle group, the individual fitness function of the eagle group is calculated by the grayscale difference between the current position of the eagle group individual and the average grayscale of the background of the molten pool image in its neighborhood: in, is the fitness function value of the kth eagle group individual at the tth iteration, For the The first iteration The individual position coordinates of the eagle group, The individual position coordinates of the eagle group The neighborhood window of set size for the center, is the preprocessed melt pool image data matrix Individual position coordinates of the eagle group The gray value at Neighborhood window The total number of pixels within is the preprocessed melt pool image data matrix In the neighborhood window Inner pixel The gray value at ; S33. Construct individual memory weights of eagle groups based on historical search information of eagle groups , update the individual positions of the eagle group : in, For the The first iteration The individual position coordinates of the eagle group, Memory weights for individual eagles: in, is the fitness function value of the i-th eagle group individual at the t-th iteration; Eagle group individual memory weight According to the difference between the current fitness function value of the eagle group and the fitness function value of the historical eagle group, the position of the eagle group individuals is moved to the area with higher grayscale difference; S34. Build eagle group search strategy to automatically determine indicators : in, Automatically determine indicators for eagle group search strategies, is the highest individual fitness function value of the eagle group, is the total number of individuals in the swarm, when the swarm search strategy automatically determines the index Less than the preset threshold When , the eagle group search automatically switches to the local refinement mode, otherwise it switches to the global exploration mode; S35. In the global exploration mode, the individual positions of the eagle group are updated as follows: in, is the next iteration position of the eagle group individuals in the global exploration mode, is the current optimal individual position of the eagle group, represents the position of an individual eagle group randomly selected from the current search space at the t + 1th iteration, and is a random factor in the interval [0,1]; S36. In the local fine mode, the individual positions of the eagle group are constrained by the temperature field of the molten pool. When the gray value of the temperature field molten pool image data corresponding to the individual position of the eagle group Above the temperature abnormality threshold When , the individual positions of the eagle group are updated to: in, is the next iteration position of the eagle group individuals in the local exploration mode, is the gray value at the current position coordinate of the eagle group individual in the temperature field melt pool image data matrix, representing the temperature field intensity at the current position coordinate. is the abnormal temperature threshold of the molten pool, is the temperature field constraint coefficient; S37. After the eagle group search is completed, select the optimal eagle group individual position of the final iteration As the center position of the microcrack defect candidate area, a fixed-size molten pool image area is intercepted with the center position of the microcrack defect candidate area as the center to form a molten pool image data matrix of the defect candidate area .

6. A defect identification and early warning method in an additive manufacturing process according to claim 5, characterized in that: The S4 specifically comprises the following steps: S41, based on the defect candidate area melt pool image data matrix Calculate the local grayscale difference characteristics of each defect candidate area in the defect candidate area molten pool image data matrix : in, Indicates the position coordinates of the defect candidate area The local grayscale difference characteristics of the defect candidate area centered on Represents the position coordinates of the defect candidate area in the molten pool image data matrix of the defect candidate area The gray value at is the overall average gray value of the defect candidate area, are the number of rows and columns of the local window respectively; S42, local grayscale difference features Adaptive fractal scaling Dynamic adjustment based on local grayscale difference characteristics: in, The defect candidate area is the position coordinate of the defect candidate area Adaptive fractal scaling at is the initial standard fractal box size, is the adaptive adjustment coefficient; S43, based on adaptive fractal scale Calculate the local fractal dimension of the melt pool image data matrix in the defect candidate area : in, Indicates the position coordinates of the defect candidate area The local fractal dimension at , Represented on a fractal scale Position coordinates of the candidate area of ​​the lower coverage defect The minimum number of boxes required at the location; S44, based on the local fractal dimension and local grayscale difference features Determine the adaptive segmentation threshold of microcrack defect area : in, is the position coordinate of the defect candidate area The adaptive segmentation threshold of the microcrack defect area, , is the weight coefficient; S45, using adaptive segmentation threshold of microcrack defect area The molten pool image data matrix of the defect candidate area Perform local threshold binary segmentation to obtain the molten pool image data matrix of the microcrack defect area after fine segmentation : 。 7. A defect identification and early warning method in an additive manufacturing process according to claim 6, characterized in that: The S5 specifically includes the following steps: S51, based on the molten pool image data matrix of the microcrack defect area after fine segmentation Construct a microcrack boundary detection operator to extract the boundary contour point set of the microcrack area : in, Represents the boundary contour point set of the microcrack defect area, Represents the melt pool image data matrix of the microcrack defect area after fine segmentation The gradient of is the edge detection threshold; S52. Calculate the length of microcrack defects , the minimum circumscribed rectangle method is used to fit the boundary points of the microcrack area and define the microcrack length is the major axis of the minimum enclosing rectangle: in, represents the length of the microcrack defect, and are any two points in the microcrack boundary contour, Represents the Euclidean distance between two points; S53, using the normal projection method to calculate the average width of the microcrack centerline direction : in, represents the average width of microcrack defects, is the coordinate of the positive boundary point extending outward from the g-th sampling point along the normal direction to the crack edge, The g-th sampling point shrinks inward along the normal direction to the negative boundary point coordinate of the crack edge. is the total number of sampling points on the center line of the microcrack; S54, based on the molten pool image data matrix of the microcrack defect area after fine segmentation Extract the crack edge contour of the microcrack defect area and determine the crack morphology characteristics , the crack morphology is described by the crack edge contour curvature characteristics: in, is the crack morphology characteristic, reflecting the degree of curvature of the crack area contour. , Respectively represent The first-order derivatives in the horizontal and vertical directions of the edge contour points, , Respectively represent The second-order derivatives in the horizontal and vertical directions of the edge contour points, Represents the total number of points sampled on the crack edge contour; S55, based on the length of microcracks ,width and crack morphology characteristics , calculate the severity score of microcrack defects : in, , and is the weight coefficient; S56. Score based on the severity of microcrack defects Set the microcrack defect level, and the defect level is divided according to: Defect grade classification in, and They are the low-level and high-level severity thresholds of microcrack defects in additive manufacturing, respectively, forming the quantitative evaluation results of microcrack defects in additive manufacturing.

Images