3D metal printing steel bar real-time detection system based on deep learning intelligent identification

Abstract

The invention discloses a 3D metal printing steel bar real-time detection system based on deep learning intelligent identification, and relates to the technical field of 3D metal printing steel bars. According to the method, appearance and internal structure data of 3D printing steel bars are collected in real time, a neural network model is utilized for anomaly detection, surface defects and internal invisible defects are classified and recognized, the system further analyzes the severity of the internal defects, and the stress concentration degree and the material fatigue life reduction condition are combined. The reinforcing steel bars with serious defects are reprinted, the reinforcing steel bars with middle defects are subjected to subsequent production optimization by adjusting printing parameters, and the reinforcing steel bars with slight defects are marked as qualified products, so that the precision, quality and safety of reinforcing steel bar production are effectively improved, and structural failure caused by hidden defects is prevented.

Description

Technical Field

[0001] The present invention relates to the technical field of 3D metal printed steel bars, and in particular to a 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition. Background Art

[0002] 3D metal printed rebar refers to the use of 3D printing technology to manufacture rebars with complex shapes and structures by stacking metal materials layer by layer. Compared with traditional manufacturing methods, this technology can accurately control the shape and size of the rebar, achieve personalized design and rapid production, and is particularly suitable for special needs in customized engineering structures. This type of 3D printed rebar usually has high strength and good corrosion resistance, and is suitable for various applications in the construction and industrial fields.

[0003] Real-time detection of 3D metal printed steel bars based on deep learning intelligent recognition refers to the use of deep learning algorithms to automatically identify and analyze the quality and shape of steel bars during 3D printing by training neural network models. This technology can monitor the printing process in real time, detect defects or deviations in a timely manner, and ensure the accuracy and consistency of the final product. This real-time detection can improve production efficiency and product quality, and is widely used in engineering and manufacturing fields that require high precision and high reliability. However, in the actual use of 3D printing technology to manufacture high-strength steel bars, internal delamination occurs due to slight fluctuations in printing parameters or problems with the material itself. This delamination makes the steel bars look intact on the surface and meet all appearance standards, but in fact, the internal delamination will cause the steel bars to break or deform when subjected to high loads. However, since the deep learning model mainly focuses on the appearance characteristics of the steel bars, it may ignore such internal defects, thereby judging steel bars with hidden defects as qualified products and putting them into use. At the same time, a serious consequence of this misidentification is that when a bridge or other building structure is subjected to external forces (such as wind, earthquake, vehicle load, etc.) during actual use, the steel bars with hidden defects inside may fail at a critical moment, causing partial or complete collapse of the structure, resulting in serious safety accidents and economic losses. Summary of the invention

[0004] The purpose of the present invention is to provide a 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition to solve the shortcomings of the background technology.

[0005] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: a 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition, comprising a data acquisition module, an anomaly detection module, an anomaly analysis module and a processing and adjustment module; Data acquisition module: during the 3D printing process, real-time collection of multi-dimensional data of the steel bar, the multi-dimensional data including the appearance image data and internal structure image data of the steel bar; Anomaly detection module: The multi-dimensional data of the steel bars collected in real time is input into the neural network model for anomaly detection, and the anomalies are classified into surface defect anomalies and internal invisible defect anomalies according to the detection results; Abnormal analysis module: classifies the detected internal cracks and delaminations, builds a data test model, analyzes the severity of the abnormal internal invisible defects of the steel bars according to the stress concentration degree of the steel bar defects and the reduction of the material fatigue life, determines whether the currently printed steel bars need to be processed, and marks the steel bars with abnormal surface defects as unqualified products; Processing adjustment module: If the severity of the internal invisible defects of the steel bar is high, it will be reprinted. If the severity of the internal invisible defects of the steel bar is medium, its printing speed and material flow rate will be dynamically adjusted to optimize the subsequent printing quality. If the severity of the internal invisible defects of the steel bar is low, it will be marked as a qualified product.

[0006] Preferably, in the anomaly detection module, anomaly detection and identification are performed on the collected multi-dimensional data of steel bars through a convolutional neural network model, specifically: The appearance image data is input into one branch of the CNN model, and the internal structure image data is input into another branch of the CNN model; in each branch, each convolution layer is applied to the input image data to automatically extract local features, and a pooling layer is added after each convolution layer to reduce the spatial size of the feature map; the convolution and pooling operations are repeated to form a deep feature map. At a certain layer of the model, the feature maps of the surface image branch and the internal structure image branch are fused to comprehensively consider the multi-dimensional features of the steel bars; the fused feature map is input into the fully connected layer for further processing. The fully connected layer will learn higher-level features and ultimately generate outputs for classification. The final output layer will determine whether the current input image data is abnormal based on the features learned by the model.

[0007] Preferably, in the abnormal analysis module, a stress intensity factor fluctuation index is generated according to the stress concentration degree at the steel bar defect, and the method for obtaining the stress intensity factor fluctuation index is: According to the actual geometry of the steel bar, a three-dimensional finite element model is established. The model should include the key features of the steel bar and potential defect areas. The geometric model is meshed, and the load conditions under the actual working conditions are applied to the model. The boundary conditions are set to simulate the fixation of the steel bar in the structure, and the stress intensity factor K is calculated. The calculation expression is: ; where σ is the nominal stress, a is the crack length or characteristic size of the defect, and Y is the geometric correction factor. For each cyclic load, the fluctuation amplitude of the stress intensity factor is calculated as , the expression is: ; Kmax is the stress intensity factor calculated under the maximum stress σmax, Kmin is the stress intensity factor calculated under the minimum stress σmin, the stress intensity factor is calculated for multiple load cycles, and several ΔK values ​​are obtained to form a fluctuation data set of the stress intensity factor, and the stress intensity factor fluctuation index FK is calculated. The expression is: ; mean(ΔK) is the average value of the stress intensity factor fluctuation amplitude ΔK under all cyclic loads, and std(ΔK) is the standard deviation of the stress intensity factor fluctuation amplitude ΔK.

[0008] Preferably, the fatigue crack sensitivity index is generated according to the fatigue life reduction of the material, and the method for obtaining the fatigue crack sensitivity index is: The crack length a is measured in real time under N load cycles. According to the measured crack length and the applied stress level, the corresponding stress intensity factor range ΔK is calculated. The ratio of the change in crack length to the corresponding number of load cycles is calculated. ; Get the crack growth rate; According to the Paris law, the crack growth rate and stress intensity factor range ΔK are expressed as: ;in: is the crack growth rate, ΔK is the stress intensity factor range, C and m are empirical constants related to the material, where m is the exponent in the Paris law. The fatigue crack sensitivity index is calculated as follows: ; Where FCSI represents fatigue crack sensitivity index, It is the sensitivity of the crack growth rate to the range of stress intensity factors, that is, the rate of change of the crack growth rate with ΔK.

[0009] Preferably, the stress intensity factor fluctuation index and the fatigue crack sensitivity index are converted into a first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the severity value label of the internal invisible defect anomaly of the steel bar with each group of first eigenvectors as the prediction target, and minimizes the sum of the prediction errors of the severity value labels of the internal invisible defect anomalies of all steel bars as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The severity value of the internal invisible defect anomaly of the steel bar is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0010] Preferably, the acquired severity value of the abnormal internal invisible defect of the steel bar is compared with a gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the severity value of the abnormal internal invisible defect of the steel bar is compared with the first standard threshold and the second standard threshold respectively; If the severity value of the abnormal internal invisible defect of the steel bar is greater than the second standard threshold, it means that the severity of the abnormal internal invisible defect of the steel bar is high, and a first-level warning signal is generated; If the severity value of the abnormal internal invisible defect of the steel bar is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the degree of the internal invisible defect of the steel bar is medium, and a secondary warning signal is generated; If the severity value of the abnormal internal invisible defect of the steel bar is less than the first standard threshold, it means that the abnormal degree of the internal invisible defect of the steel bar is low, and no warning signal is generated at this time.

[0011] Preferably, in the processing and adjustment module, if the severity of the abnormal invisible defects of the steel bars is moderate, that is, the severity value of the abnormal invisible defects of the steel bars generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the data generated within a subsequent fixed time period that are greater than or equal to the first standard threshold and less than or equal to the second standard threshold are collected to establish a corresponding data set, the mean and standard deviation of the data set are calculated, and after analyzing them, the printing speed and material flow rate of the steel bars are dynamically adjusted according to the analysis results to optimize the subsequent printing quality.

[0012] Preferably, if the mean of the severity values ​​in the data set is greater than or equal to the reference threshold of the mean of the severity values, and the standard deviation of the severity values ​​is less than the reference threshold of the standard deviation of the severity values, it indicates that the stability of the printing process is good but the degree of defects is high, and the printing speed and material flow rate are reduced; If the mean value of the severity value is greater than or equal to the reference threshold value of the mean value of the severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold value of the standard deviation of the severity value, it indicates that the stability of the printing process is poor and there are many systematic defects, and the printing speed and the flow rate of the material are reduced; If the mean value of the severity value is less than the reference threshold value of the mean value of the severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold value of the standard deviation of the severity value, it indicates that the printing process is unstable, and the printing speed is reduced while keeping the material flow rate unchanged; If the mean of the severity values ​​is less than the reference threshold of the mean of the severity values, and the standard deviation of the severity values ​​is less than the reference threshold of the standard deviation of the severity values, it indicates that the printing process is not only stable but also of high quality, and the existing printing speed and material flow rate are maintained.

[0013] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention collects and analyzes multi-dimensional data of 3D printed steel bars in real time, combines the stress concentration level and material fatigue life reduction, and accurately classifies and evaluates the severity of the internal invisible defects of the steel bars. It can automatically detect and mark steel bars with surface defects, and take corresponding treatment measures according to the severity of the internal defects, such as reprinting, dynamically adjusting printing parameters, or directly marking them as qualified products, thereby ensuring the quality and safety of the products.

[0014] 2. The present invention dynamically analyzes the data generated during the printing process, and the system can continuously optimize the printing process and improve the stability and efficiency of the production process. In particular, when moderate defects are detected, the system fine-tunes the printing speed and material flow rate according to the analysis results of the data set, further improving the final quality of the steel bars. Overall, it not only significantly improves the production quality and consistency of 3D printed steel bars, but also reduces the potential risks caused by invisible defects, and is suitable for high-precision and high-reliability engineering application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0016] Figure 1 The present invention is a flow chart of the method.

[0017] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] For examples, see Figure 1 and 2 As shown, the 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition described in this embodiment includes a data acquisition module, an anomaly detection module, an anomaly analysis module and a processing and adjustment module; Data acquisition module: during the 3D printing process, real-time collection of multi-dimensional data of the steel bar, the multi-dimensional data including the appearance image data and internal structure image data of the steel bar; Anomaly detection module: The multi-dimensional data of the steel bars collected in real time is input into the neural network model for anomaly detection, and the anomalies are classified into surface defect anomalies and internal invisible defect anomalies according to the detection results; Abnormal analysis module: classifies the detected internal cracks and delaminations, builds a data test model, analyzes the severity of the abnormal internal invisible defects of the steel bars according to the stress concentration degree of the steel bar defects and the reduction of the material fatigue life, determines whether the currently printed steel bars need to be processed, and marks the steel bars with abnormal surface defects as unqualified products; Processing adjustment module: If the severity of the internal invisible defects of the steel bar is high, it will be reprinted. If the severity of the internal invisible defects of the steel bar is medium, its printing speed and material flow rate will be dynamically adjusted to optimize the subsequent printing quality. If the severity of the internal invisible defects of the steel bar is low, it will be marked as a qualified product.

[0020] In the data acquisition module, during the 3D printing process, multi-dimensional data of the steel bar is collected in real time, and the multi-dimensional data includes the appearance image data and the internal structure image data of the steel bar, specifically: The acquisition of appearance image data specifically includes: Install high-resolution industrial cameras around the working area of ​​the 3D printer to ensure that the cameras can cover all angles of the printing area. Multiple cameras can be used to synchronously capture images from different angles to obtain comprehensive appearance data of the steel bars. Through real-time image acquisition by high-resolution cameras, details of the steel bar surface can be captured, such as surface texture, finish, dimensional deviation and possible surface defects (such as cracks, dents, uneven surface roughness, etc.). This image data provides the basis for subsequent appearance inspection.

[0021] Equipped with a special light source (such as a ring LED light) and optical system, the camera is calibrated to eliminate imaging errors caused by uneven lighting or reflections, ensuring image quality and consistency. The calibrated optical system can provide uniform lighting, avoid shadow or reflection interference, improve image clarity and contrast, and enable tiny surface defects to be accurately identified.

[0022] The appearance image data collected by the camera is transmitted to the central processing unit (CPU) or image processing server in real time through a high-speed data bus (such as USB 3.0 or GigE), and is stored and backed up in real time. High-speed data transmission and storage ensures the continuity and real-time nature of data collection, avoids data loss, and enables subsequent real-time analysis and detection to proceed smoothly.

[0023] The acquisition of internal structure image data specifically includes: Integrating non-destructive testing (NDT) equipment in 3D printing equipment, such as X-ray scanners or ultrasonic probes, can penetrate the material to obtain the internal structure image of the steel bar during the printing process. Through NDT equipment, the internal structural data of the steel bar can be obtained in real time, including material uniformity, internal cracks, pores, delamination and other defect information. These internal image data are crucial for identifying hidden defects and ensuring the quality of steel bars.

[0024] Using multi-level scanning technology, by adjusting the angle and frequency of X-rays or ultrasound, image data at different depths of the steel bar is obtained to generate a three-dimensional internal structure model. At the same time, the scanning and 3D printing process are synchronized to ensure the real-time nature of the data. Multi-level scanning provides a comprehensive view of the internal structure of the steel bar, so that hidden defects at any depth can be discovered. This synchronous imaging technology can ensure that the data matches the printing process and prevent data misalignment or delay.

[0025] The acquired internal structure image data is preprocessed through image processing algorithms (such as edge detection, image enhancement, filtering, etc.) to improve the clarity and recognizability of the image, and then the processed image data is input into the deep learning model or image analysis system for real-time analysis. Through image processing, noise can be filtered out and the contrast of defects can be enhanced, so that the deep learning model can more accurately identify and classify internal defects, ensuring the reliability and accuracy of real-time detection.

[0026] Anomaly detection module: The multi-dimensional data of steel bars collected in real time is input into the neural network model for anomaly detection, and the anomalies are classified into surface defect anomalies and internal invisible defect anomalies according to the detection results.

[0027] Among them, the neural network models that can be used for multi-dimensional data anomaly detection of steel bars mainly include: Convolutional Neural Network (CNN): CNNs excel in processing image data and are particularly suitable for detecting surface defects in steel bar appearance images. They can automatically extract spatial features in images and detect anomalies such as cracks, dents, and surface roughness. They can automatically extract low-level and high-level features in images, reducing reliance on manual feature engineering. They have a good tolerance for noise and image deformation. Common structures include LeNet, AlexNet, VGG, ResNet, etc.

[0028] 3D CNN (3D Convolutional Neural Network): 3D CNN is suitable for processing three-dimensional data. For example, in internal structure image detection, it can analyze X-ray, CT scan or ultrasound image data to detect internal hidden defects (such as delamination, cracks, pores, etc.). It can extract features in three-dimensional space and is suitable for processing volume data or multi-angle data. It extracts deep features from multi-layer data to help identify subtle abnormalities in internal structures.

[0029] Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM): RNNs and LSTMs are suitable for processing sequence data, especially for detecting time series anomalies in the printing process (such as changes in printing temperature and pressure). They can detect potential defects caused by continuous changes in the steel bar manufacturing process. They can remember dependencies over long time spans and are suitable for analyzing continuous production data. They can identify anomalies in dynamic changes in the production process and help predict and prevent defects.

[0030] The convolutional neural network model is used to detect and identify anomalies in the multi-dimensional data of the collected steel bars. Specifically: De-noise the appearance image data of the steel bars to remove noise and interference in the image. Image enhancement techniques (such as contrast enhancement and edge enhancement) can be used to improve the clarity of the image. Filter and normalize the internal structure image data (such as obtained by X-ray or ultrasound) to ensure data quality and consistency. Normalize all image data to the same numerical range (such as between 0 and 1) to ensure input consistency from different data sources.

[0031] Convert the processed image data into a tensor format that meets the input requirements of the convolutional neural network. For two-dimensional images (such as surface images), convert them into three-dimensional tensors of the form (height, width, number of channels); for three-dimensional images (such as internal structure images), convert them into four-dimensional tensors of the form (depth, width, height, number of channels).

[0032] The appearance image data is fed into one branch of the CNN model. This branch is usually a two-dimensional convolutional neural network (2D CNN) that extracts spatial features of the image, such as edges, textures, and shapes.

[0033] The internal structure image data is fed into another branch of the CNN model. This branch can be a two-dimensional or three-dimensional convolutional neural network (3D CNN) to extract the characteristics of the internal structure of the steel bar, such as material delamination, pores, cracks, etc.

[0034] In each branch, multiple convolutional layers are applied to the input image data to automatically extract local features. The convolution kernel (filter) scans different areas of the image and extracts important feature maps.

[0035] Adding a pooling layer (such as max pooling or average pooling) after each convolutional layer reduces the spatial size of the feature map, thereby reducing computational complexity and enhancing feature invariance.

[0036] The above convolution and pooling operations can be repeated multiple times to form deep feature maps that contain high-level features of the surface and internal images of the steel bars.

[0037] At a certain layer of the model, the feature maps of the surface image branch and the internal structure image branch are fused (e.g., by splicing or weighted averaging) to comprehensively consider the multi-dimensional characteristics of the steel bars.

[0038] The fused feature map is input into a fully connected layer (or dense layer) for further processing. The fully connected layer learns higher-level features and ultimately generates output for classification.

[0039] The final output layer will determine whether there is an anomaly in the current input image data based on the features learned by the model, and output the probability or confidence of anomaly detection.

[0040] At the output layer, the detected anomalies are classified into two categories: surface defect anomalies and internal hidden defect anomalies. Surface defect anomalies include cracks, uneven surface roughness, etc., while internal hidden defect anomalies include internal cracks, material delamination, pores, etc. If the same input data may have multiple anomalies at the same time, the model can output the confidence of multiple categories (for example, using multi-label classification) to identify multiple defects.

[0041] Abnormal analysis module: classify the detected internal cracks and delamination, build a data test model, analyze the severity of the abnormal internal invisible defects of the steel bars according to the stress concentration degree and material fatigue life reduction at the steel bar defects, determine whether the currently printed steel bars need to be processed, and mark the steel bars with abnormal surface defects as unqualified products.

[0042] The 3D CNN model is used to perform deep convolution processing on the internal structure image to extract relevant feature maps. The feature maps contain information about the location, morphological characteristics, and size of cracks and delamination.

[0043] Key feature identification: Crack characteristics: including the length, width, depth, shape (linear, bifurcated, curved, etc.) of the crack, the location of the crack inside the material, and the stress concentration around the crack. Layer characteristics: including the thickness of the layer, the size of the gap between the layers, the area of ​​the layer, the shape of the layer (parallel, non-parallel), and the distance between the layer and the outer surface of the steel bar.

[0044] The above features are converted into feature vectors for classification analysis. The feature vector of each crack or delamination may contain position coordinates, size parameters, morphological parameters, etc.

[0045] The feature vectors are preliminarily classified according to the preset criteria or the classification boundaries determined by learning. For example, crack classification: cracks can be classified into shallow cracks, deep cracks, through cracks, etc. based on the length and depth of the cracks.

[0046] Based on the area of ​​the stratification and the size of the gaps between the layers, the stratification can be divided into minor stratification, significant stratification, major stratification, etc.

[0047] Clustering algorithms such as K-means and DBSCAN are used to cluster the extracted crack and delamination features and automatically group similar defects into one category. These algorithms can classify cracks and delaminations based on similarities in feature space.

[0048] Support Vector Machine (SVM) classification: Use supervised learning algorithms such as SVM to classify cracks and delamination. This method is suitable for situations where labeled data is already available and can accurately classify newly detected defects based on known defect types.

[0049] Assign classification labels to each detected internal crack and delamination. For example: Crack 1: Deep crack, 10mm in length, located in the center of the steel bar. Delamination 1: Significant delamination, with a gap of 0.5mm between the layers, close to the surface of the steel bar.

[0050] The classification results are visualized to generate a 3D model or annotation diagram of internal cracks and delamination, so that users can intuitively view the specific conditions of each defect and the classification results. The classification results are stored in the database together with the original data for subsequent analysis and quality control.

[0051] The stress intensity factor fluctuation index is generated according to the stress concentration degree at the steel bar defect. The method for obtaining the stress intensity factor fluctuation index is: Build a 3D finite element model based on the actual steel bar geometry. The model should include the key features of the steel bar as well as potential defect areas (e.g., cracks, delamination). Mesh the geometry, especially using denser meshes in defect areas to improve computational accuracy. Usually, finer meshes are required near the crack tip to capture stress concentration effects. Use either hexahedral or tetrahedral meshes, the choice depends on the geometry complexity and computational requirements. Apply actual load conditions to the model. Cyclic loads are usually required to simulate stress fluctuations in the steel bar during service. Among them, the maximum stress σmax is the maximum stress that the steel bar may withstand during service. The minimum stress σmin is the minimum stress that the steel bar may withstand during service. The applied stress can be tensile, compressive, or bending, and should be selected based on the actual working conditions.

[0052] Set boundary conditions to simulate the fixation of the steel bar in the structure. Usually, a fixed constraint can be applied to one end of the steel bar and a load can be applied to the other end. Calculate the stress intensity factor K, the calculation expression is: ; where σ is the nominal stress, which depends on the applied load conditions (can be σmax or σmin), a is the crack length or characteristic size of the defect, and Y is the geometric correction factor, which depends on the geometry and location of the crack and is usually obtained through FEA analysis. For each cyclic load, the amplitude of the stress intensity factor fluctuation is calculated , the expression is: ; Kmax is the stress intensity factor calculated under the maximum stress σmax, Kmin is the stress intensity factor calculated under the minimum stress σmin, the stress intensity factor is calculated for multiple load cycles, and several ΔK values ​​are obtained to form a fluctuation data set of the stress intensity factor, and the stress intensity factor fluctuation index FK is calculated. The expression is: ; mean(ΔK) is the average value of the stress intensity factor fluctuation amplitude ΔK under all cyclic loads, and std(ΔK) is the standard deviation of the stress intensity factor fluctuation amplitude ΔK.

[0053] The FK index measures the stability of the fluctuations of the stress intensity factor over multiple load cycles. A higher FK value indicates that the fluctuations of the stress intensity factor are smaller and the stress concentration effect at the defect is more stable. A lower FK value indicates that the stress intensity factor fluctuates more, indicating that the defect is more sensitive to fatigue growth and may more easily lead to material failure.

[0054] The fatigue crack sensitivity index is generated according to the fatigue life reduction of the material. The method for obtaining the fatigue crack sensitivity index is: The crack length a is measured in real time under N load cycles. According to the measured crack length and the applied stress level, the corresponding stress intensity factor range ΔK is calculated. The ratio of the change in crack length to the corresponding number of load cycles is calculated. ; Get the crack growth rate.

[0055] According to the Paris law, the crack growth rate and stress intensity factor range ΔK are expressed as: ;in: is the crack growth rate, ΔK is the stress intensity factor range, C and m are empirical constants related to the material, which can be obtained by fitting the experimental data, where m is the exponent in the Paris law, reflecting the sensitivity of the crack growth rate to the stress intensity factor range. The fatigue crack sensitivity index is calculated as follows: ; Where FCSI represents fatigue crack sensitivity index, It is the sensitivity of the crack growth rate to the range of stress intensity factors, that is, the rate of change of the crack growth rate with ΔK.

[0056] It reflects the sensitivity of the crack growth rate to changes in the stress intensity factor, that is, the degree of influence of a small change in the stress intensity factor on the crack growth rate. This index is usually directly related to the fatigue performance of the material. A higher FCSI means that the material is more sensitive to changes in the stress intensity factor, which means that the material has a faster crack growth rate under fatigue load and a shorter fatigue life. A low FCSI indicates that the material has a slower crack growth under stress fluctuations and a longer fatigue life.

[0057] The stress intensity factor fluctuation index and the fatigue crack sensitivity index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the severity value label of the internal invisible defect anomaly of the steel bar with each group of first eigenvectors as the prediction target, and minimizes the sum of the prediction errors of the severity value labels of the internal invisible defects of all steel bars as the training target. The machine learning model is trained until the sum of the prediction errors converges, and the model training is stopped. The severity value of the internal invisible defect anomaly of the steel bar is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0058] The method for obtaining the severity value of the internal invisible defect anomaly of the steel bar is: from the first eigenvector training data of the trained machine learning model, obtain the corresponding function expression: ; In the formula, is the output function of the model, is the stress intensity factor fluctuation index, FCSI is the fatigue crack sensitivity index, It is the severity value of the abnormal internal invisible defects of the steel bars.

[0059] The acquired severity value of the internal invisible defect abnormality of the steel bar is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the severity value of the internal invisible defect abnormality of the steel bar is compared with the first standard threshold and the second standard threshold respectively; If the severity value of the abnormal internal invisible defect of the steel bar is greater than the second standard threshold, it means that the abnormal internal invisible defect of the steel bar is of high severity. At this time, a first-level warning signal is generated, indicating that the abnormal internal invisible defect of the steel bar is serious and needs to be handled immediately; If the severity value of the abnormal internal invisible defects of the steel bar is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the internal invisible defects of the steel bar are of medium severity. At this time, a secondary warning signal is generated, and it may need to be repaired, downgraded or further evaluated; If the severity value of the abnormal internal invisible defects of the steel bar is less than the first standard threshold, it means that the abnormal degree of the internal invisible defects of the steel bar is low. At this time, no warning signal is generated. It is generally considered acceptable and does not require special treatment. It can be put into use directly.

[0060] If surface defects (such as cracks, dents or excessive surface roughness) are judged to be beyond the acceptable range, the system will automatically mark these steel bars as unqualified products and remove them from the production process to prevent steel bars that do not meet quality standards from entering subsequent use or sales, thereby ensuring the safety and reliability of the final product.

[0061] It should be noted here that the importance of the first-level warning signal is greater than that of the second-level warning signal, and the importance of the second-level warning signal is greater than that of the third-level warning signal. Relevant personnel can take corresponding measures according to different warning signal levels.

[0062] Processing adjustment module: If the severity of the internal invisible defects of the steel bar is high, it will be reprinted. If the severity of the internal invisible defects of the steel bar is medium, its printing speed and material flow rate will be dynamically adjusted to optimize the subsequent printing quality. If the severity of the internal invisible defects of the steel bar is low, it will be marked as a qualified product.

[0063] If the severity of the internal invisible defect abnormality of the steel bar is assessed as high, the system will automatically trigger the reprint process. This means that the currently printed steel bar cannot meet the quality standards due to serious internal defects (such as deep cracks, severe delamination, etc.), and its safety and reliability in actual use cannot be guaranteed. Therefore, in order to ensure that the product meets the strict engineering requirements, the steel bar must be reprinted to produce a qualified product that meets the quality standards.

[0064] If the severity of the abnormal invisible defects of the steel bars is moderate, that is, the severity value of the abnormal invisible defects of the steel bars generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the values ​​greater than or equal to the first standard threshold and less than or equal to the second standard threshold generated within a subsequent fixed time period are collected to establish a corresponding data set, the mean and standard deviation of the data set are calculated, and after analyzing them, the printing speed and material flow rate of the steel bars are dynamically adjusted according to the analysis results to optimize the subsequent printing quality.

[0065] If the mean of the severity values ​​in the data set is greater than or equal to the reference threshold of the mean of the severity values, and the standard deviation of the severity values ​​is less than the reference threshold of the standard deviation of the severity values, it indicates that the abnormality of the internal invisible defects of the steel bars is generally high (the mean is higher than the reference threshold), but the fluctuation of these defects in different printing batches is small (the standard deviation is lower than the reference threshold). This indicates that the stability of the printing process is good but the degree of defects is high. Reduce the printing speed and material flow rate, which can reduce the occurrence of stress concentration effects to ensure that the material can fully solidify and fill possible defect areas more evenly.

[0066] If the mean value of the severity value is greater than or equal to the reference threshold of the mean value of the severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold of the standard deviation of the severity value, it indicates that the overall abnormality of the internal invisible defects of the steel bars is high, and the severity of the defects fluctuates greatly between different printing batches. This indicates that the stability of the printing process is poor and there are many systematic defects. Reducing the printing speed and the flow rate of the material will help reduce the problem of uneven material accumulation or insufficient interlayer adhesion.

[0067] If the mean severity value is less than the reference threshold of the mean severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold of the standard deviation of the severity value, this indicates that the overall abnormality of the internal invisible defects of the steel bars is low, but the severity of the defects between different batches fluctuates greatly. This means that the overall quality is acceptable, but the process stability is low, and large defects occasionally occur. Improve the consistency of printing by reducing the printing speed, and at the same time, keep the material flow rate unchanged to reduce occasional defects caused by fluctuations.

[0068] If the mean severity value is less than the reference threshold of the mean severity value, and the standard deviation of the severity value is less than the reference threshold of the standard deviation of the severity value, this indicates that the internal invisible defects of the steel bars are generally less abnormal, and the degree of defects between different printing batches fluctuates less. This is an ideal situation, indicating that the printing process is not only stable, but also of high quality. In this case, the system should maintain the existing printing speed and material flow rate, because the current settings have achieved stable and high-quality printing. If production time or cost needs to be optimized, you can try to slightly increase the printing speed or material flow rate while maintaining quality, but be careful to continue monitoring defects to prevent quality degradation caused by increased speed.

[0069] If the severity of the internal invisible defect anomaly of the steel bar is assessed as low, the system will automatically mark the steel bar as a qualified product. This means that the internal defects of the current steel bar (such as cracks, delamination, etc.) are within the acceptable range and will not have a significant impact on the structural integrity and performance of the steel bar. Therefore, the steel bar can smoothly pass the quality inspection and enter the subsequent production process or be put into use without additional processing.

[0070] In this application, by collecting and analyzing the data set of abnormal severity of internal invisible defects, and based on the analysis results of the mean and standard deviation, the system can effectively adjust the printing speed and material flow rate to optimize the subsequent printing quality. According to different severity distributions, the system adopts different adjustment strategies to ensure that the internal quality of the steel bars remains within a controllable range and maximize production efficiency and product quality.

[0071] In this embodiment, during the 3D printing process, the system collects the appearance image and internal structure image data of the steel bar in real time through the data acquisition module, and inputs these multi-dimensional data into the neural network model for abnormality detection, which is classified into surface defect abnormalities and internal invisible defect abnormalities. The abnormality analysis module further classifies the detected internal cracks and delamination conditions, and uses the data test model to comprehensively analyze the stress concentration degree and material fatigue life reduction, judge the severity of the internal invisible defects of the steel bar, and decide whether to process it. For steel bars with abnormal surface defects, the system will mark them as unqualified products. The processing adjustment module takes corresponding measures according to the severity of the internal invisible defects: steel bars with high severity will be reprinted, steel bars with medium severity will optimize the subsequent printing quality by adjusting the printing speed and material flow rate, and steel bars with low severity will be marked as qualified products.

[0072] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0073] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0074] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0075] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0076] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0077] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition, characterized by: It includes a data acquisition module, an anomaly detection module, an anomaly analysis module and a processing and adjustment module; Data acquisition module: during the 3D printing process, real-time collection of multi-dimensional data of the steel bar, the multi-dimensional data including the appearance image data and internal structure image data of the steel bar; Anomaly detection module: The multi-dimensional data of the steel bars collected in real time is input into the neural network model for anomaly detection, and the anomalies are classified into surface defect anomalies and internal invisible defect anomalies according to the detection results; Abnormal analysis module: classifies the detected internal cracks and delaminations, builds a data test model, analyzes the severity of the abnormal internal invisible defects of the steel bars according to the stress concentration degree of the steel bar defects and the reduction of the material fatigue life, determines whether the currently printed steel bars need to be processed, and marks the steel bars with abnormal surface defects as unqualified products; Processing adjustment module: If the severity of the internal invisible defects of the steel bar is high, it will be reprinted. If the severity of the internal invisible defects of the steel bar is medium, its printing speed and material flow rate will be dynamically adjusted to optimize the subsequent printing quality. If the severity of the internal invisible defects of the steel bar is low, it will be marked as a qualified product.

2. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 1 is characterized in that: In the anomaly detection module, the convolutional neural network model is used to detect and identify anomalies in the multi-dimensional data of the collected steel bars. Specifically: The appearance image data is fed into one branch of the CNN model, and the internal structure image data is fed into another branch of the CNN model; in each branch, various convolutional layers are applied to the input image data to automatically extract local features, and a pooling layer is added after each convolutional layer to reduce the spatial size of the feature map; the convolution and pooling operations are repeated to form a deep feature map, and at a certain layer of the model, the feature maps of the surface image branch and the internal structure image branch are fused to comprehensively consider the multi-dimensional features of the steel bars; The fused feature map is input into the fully connected layer for further processing. The fully connected layer will learn higher-level features and ultimately generate output for classification. The final output layer will determine whether there is an abnormality in the current input image data based on the features learned by the model.

3. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 1 is characterized in that: In the abnormal analysis module, the stress intensity factor fluctuation index is generated according to the stress concentration degree at the steel bar defect. The method for obtaining the stress intensity factor fluctuation index is: According to the actual geometry of the steel bar, a three-dimensional finite element model is established. The model should include the key features of the steel bar and potential defect areas. The geometric model is meshed, and the load conditions under the actual working conditions are applied to the model. The boundary conditions are set to simulate the fixation of the steel bar in the structure, and the stress intensity factor K is calculated. The calculation expression is: ; where σ is the nominal stress, a is the crack length or characteristic size of the defect, and Y is the geometric correction factor. For each cyclic load, the fluctuation amplitude of the stress intensity factor is calculated as , the expression is: ; Kmax is the stress intensity factor calculated under the maximum stress σmax, and Kmin is the stress intensity factor calculated under the minimum stress σmin. The stress intensity factor is calculated for multiple load cycles to obtain several ΔK values ​​to form a fluctuation data set of the stress intensity factor. The stress intensity factor fluctuation index FK is calculated, and the expression is: ; mean(ΔK) is the average value of the stress intensity factor fluctuation amplitude ΔK under all cyclic loads, and std(ΔK) is the standard deviation of the stress intensity factor fluctuation amplitude ΔK.

4. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 3 is characterized in that: The fatigue crack sensitivity index is generated according to the fatigue life reduction of the material. The method for obtaining the fatigue crack sensitivity index is: The crack length a is measured in real time under N load cycles. According to the measured crack length and the applied stress level, the corresponding stress intensity factor range ΔK is calculated. The ratio of the change in crack length to the corresponding number of load cycles is calculated. ; Get the crack growth rate; According to the Paris law, the crack growth rate and stress intensity factor range ΔK are expressed as: ;in: is the crack growth rate, ΔK is the stress intensity factor range, C and m are empirical constants related to the material, where m is the exponent in the Paris law. The fatigue crack sensitivity index is calculated as follows: ; Where FCSI represents fatigue crack sensitivity index, It is the sensitivity of the crack growth rate to the range of stress intensity factors, that is, the rate of change of the crack growth rate with ΔK.

5. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 4 is characterized in that: The stress intensity factor fluctuation index and the fatigue crack sensitivity index are converted into the first eigenvector, and the first eigenvector is used as the input of the machine learning model. The machine learning model predicts the severity value label of the internal invisible defect anomaly of the steel bar with each group of first eigenvectors as the prediction target, and minimizes the sum of the prediction errors of the severity value labels of the internal invisible defects of all steel bars as the training target. The machine learning model is trained until the sum of the prediction errors converges, and the model training is stopped. The severity value of the internal invisible defect anomaly of the steel bar is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

6. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 5 is characterized in that: The acquired severity value of the internal invisible defect abnormality of the steel bar is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the severity value of the internal invisible defect abnormality of the steel bar is compared with the first standard threshold and the second standard threshold respectively; If the severity value of the abnormal internal invisible defect of the steel bar is greater than the second standard threshold, it means that the severity of the abnormal internal invisible defect of the steel bar is high, and a first-level warning signal is generated; If the severity value of the abnormal internal invisible defect of the steel bar is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the degree of the internal invisible defect of the steel bar is medium, and a secondary warning signal is generated; If the severity value of the abnormal internal invisible defect of the steel bar is less than the first standard threshold, it means that the abnormal degree of the internal invisible defect of the steel bar is low, and no warning signal is generated at this time.

7. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 1 is characterized in that: In the processing and adjustment module, if the severity of the internal invisible defect anomaly of the steel bar is moderate, that is, the severity value of the internal invisible defect anomaly of the steel bar generated within a fixed time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the severity values ​​greater than or equal to the first standard threshold and less than or equal to the second standard threshold generated within a subsequent fixed time period are collected to establish a corresponding data set, the mean and standard deviation of the data set are calculated, and after analyzing them, the printing speed and material flow rate of the steel bar are dynamically adjusted according to the analysis results to optimize the subsequent printing quality.

8. The 3D metal printed steel bar real-time detection system based on deep learning intelligent recognition according to claim 7 is characterized in that: If the mean of the severity values ​​in the data set is greater than or equal to the reference threshold of the mean of the severity values, and the standard deviation of the severity values ​​is less than the reference threshold of the standard deviation of the severity values, it indicates that the stability of the printing process is good but the degree of defects is high, and the printing speed and material flow rate are reduced; If the mean value of the severity value is greater than or equal to the reference threshold value of the mean value of the severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold value of the standard deviation of the severity value, it indicates that the stability of the printing process is poor and there are many systematic defects, and the printing speed and the flow rate of the material are reduced; If the mean value of the severity value is less than the reference threshold value of the mean value of the severity value, and the standard deviation of the severity value is greater than or equal to the reference threshold value of the standard deviation of the severity value, it indicates that the printing process is unstable, and the printing speed is reduced while keeping the material flow rate unchanged; If the mean of the severity values ​​is less than the reference threshold of the mean of the severity values, and the standard deviation of the severity values ​​is less than the reference threshold of the standard deviation of the severity values, it indicates that the printing process is not only stable but also of high quality, and the existing printing speed and material flow rate are maintained.

Images