A 3D printing process intelligent control method and system based on real-time feedback

Abstract

The present invention discloses a 3D printing process intelligent control method and system based on real-time feedback, which relates to the field of 3D printing technology. A data acquisition module collects print layer surface images and print data in real time, constructs image data sets and print data sets, and cleans and standardizes the data in a preprocessing module. The image analysis module uses a convolutional neural network to analyze print layer surface defects, generates a defect score D, and compares it with a preset defect score threshold D1, evaluates the print layer status and triggers an early warning. The comprehensive evaluation module combines the print data and defect score to calculate the crack generation probability Z. The intelligent control module compares the crack generation probability Z with a preset crack generation probability threshold Z1 to evaluate the printing status, and controls the printing parameters based on the feedback results, effectively preventing and reducing cracks and structural defects in the printing process, thereby improving printing accuracy and finished product quality.

Description

Technical Field

[0001] The present invention relates to the technical field of 3D printing, and in particular to a 3D printing process intelligent control method and system based on real-time feedback. Background Art

[0002] 3D printing technology, as a manufacturing method of adding materials layer by layer, is widely used in aerospace, medical, electronics, precision engineering and other fields. In recent years, with the development of materials science, 3D printing technology has gradually extended to the manufacture of ceramic materials. Ceramic materials have excellent properties such as high temperature resistance, corrosion resistance, and insulation. Therefore, they have important applications in the manufacture of high-temperature components, insulating components and wear-resistant structural parts. However, the ceramic 3D printing process is complex and requires high printing accuracy and real-time control. During the printing process, the deposition, cooling rate and temperature distribution of each layer need to be precisely controlled to ensure the dimensional accuracy and structural integrity of the printed parts.

[0003] The current ceramic 3D printing process still has some technical deficiencies. First, due to the brittle characteristics of ceramic materials, they are easily affected by temperature gradients and cooling rates during the printing process, generating internal stress, which leads to weak bonding between material layers and even cracks or tiny gaps. Second, existing printing systems usually use preset printing paths and printing parameters, lack dynamic adjustment capabilities based on real-time feedback, and cannot optimize the actual printing layer status. Therefore, during the printing process, if the surface quality of the printed layer is poor or the cooling process is uneven, the system cannot respond and adjust the parameters in time, resulting in unstable print quality and difficulty in achieving the expected accuracy and strength requirements.

[0004] This defect mainly stems from the complex thermodynamic changes, stress distribution, and material properties during the printing process of ceramic materials. When the temperature and cooling rate are not properly controlled during the printing process, the temperature gradient within the material will induce thermal stress, resulting in weakened adhesion between printed layers and even abnormal phenomena such as cracks and interlayer separation. These abnormalities will destroy the structural integrity of the ceramic parts, affect the mechanical properties and dimensional accuracy of the printed parts, and reduce their reliability in high temperature or high load environments. Therefore, achieving high-precision real-time monitoring and intelligent control of the ceramic 3D printing process, which can effectively reduce the probability of defects and improve the overall quality of printed parts by adjusting printing parameters in real time, has become an urgent need. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a 3D printing process intelligent control method and system based on real-time feedback, which solves the problems in the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent control system for a 3D printing process based on real-time feedback, comprising a data acquisition module, a preprocessing module, an image analysis module, a comprehensive evaluation module and an intelligent control module;

[0007] The data acquisition module is used to collect high-definition images of the printing layer surface and printing data during the 3D printing process, and to construct an image data set and a printing data set based on the collected high-definition images of the printing layer surface and printing data;

[0008] The pre-processing module is used to pre-process the high-definition image of the printing layer surface and the printing data in the image data set and the printing data set;

[0009] The image analysis module is used to construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1, compare and analyze the preset defect score threshold D1 with the printed layer defect score D, evaluate the printed layer status, and trigger an early warning mechanism to control the printed layer;

[0010] The comprehensive evaluation module is used to obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D;

[0011] The intelligent control module is used to preset a crack generation probability threshold Zl, compare and analyze the preset crack generation probability threshold Zl with the crack generation probability Z, evaluate the 3D printing status, and feed back data to the 3D printing data center based on the evaluated 3D printing status to control the 3D printing process.

[0012] Preferably, the data acquisition module includes an image data acquisition unit and a print data acquisition unit;

[0013] The image data acquisition unit is used to use a high-resolution camera installed above the print head to collect the surface image of the printed layer after each layer printing process is completed, and transmit the collected surface image of the printed layer to the 3D printing data center via wireless network technology to construct an image data set;

[0014] The printing data acquisition unit is used to deploy an intelligent sensor group at the bottom of the 3D printer platform to obtain printing data of the 3D printing process, wherein the printing data includes the printing layer temperature, printing layer stress and deposition layer thickness, and obtains the thermal expansion coefficient and effective elastic modulus of the material through the physical property database of the printing material and the material manual;

[0015] The printed layer temperature data refers to the temperature of the printed layer obtained by real-time monitoring using an infrared temperature sensor;

[0016] The printed layer stress refers to the use of a laser stress plate installed at the bottom of the printing platform to monitor the stress changes generated during the printing process in real time and obtain the printed layer stress data;

[0017] The thickness of the deposited layer refers to the use of a deployed optical sensor to detect the height change of the printed layer through the reflected light signal to obtain the thickness data of the deposited layer;

[0018] The obtained printing layer temperature data, printing layer stress data and deposition layer height data are transmitted to the 3D printing data center via wireless network technology to construct a printing data set.

[0019] Preferably, the pre-processing module includes a print data processing unit and an image processing unit;

[0020] The print data processing unit is used to perform noise filtering, data cleaning and normalization processing on the print data in the print data set;

[0021] The noise filtering refers to using a Kalman filter to remove random noise from the print data;

[0022] The data cleaning refers to detecting outliers using a box plot, calculating the interquartile range (IQR) in the box plot, and obtaining the normal data range (Q1-1.5×IQR, Q3+1.5×IQR). Printed data that is not within the normal data range is considered abnormal data and is removed.

[0023] The normalization process refers to performing dimensionless processing on the collected printing data, and unifying the units of the collected printing data.

[0024] Preferably, the image processing unit is configured to perform region cropping, RGB standardization, and boundary setting on the image data in the image data set;

[0025] The area cropping refers to using computer vision technology to crop out the printing area and remove background information;

[0026] The RGB standardization refers to performing pixel operations on image data and performing standard version processing on each pixel;

[0027] The boundary setting refers to marking image boundary points using an edge detection algorithm.

[0028] Preferably, the image analysis module includes a model building unit and a preliminary evaluation unit;

[0029] The model building unit is used to build a defect analysis model based on convolutional neural network technology, and in combination with a 3D printing database, obtain historical printing image data, input the historical printing image data as training data into the input layer of the defect analysis model, use back propagation and optimization algorithms to adjust the defect analysis model parameters, input the real-time collected printing image data into the trained defect analysis model, and obtain a print layer defect score D;

[0030] The defect analysis model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer;

[0031] The input layer is used to input the real-time collected printing image data into the defect analysis model;

[0032] The convolutional layer is used to extract local features of image data and generate a local feature map;

[0033] The pooling layer is used to reduce the dimension of the local feature map obtained by the convolutional layer;

[0034] The fully connected layer is used to map the features to the printed layer defect score D output;

[0035] The output layer is used to output the print layer defect score D. The print layer defect score D is specifically obtained as follows:

[0036]

[0037] Where, is the average edge strength, is the texture uniformity, is the color uniformity coefficient, 、 and are the weight values ​​of average edge intensity, texture uniformity and color uniformity coefficient respectively, and c is the bias term;

[0038] The average edge strength The way to obtain it is:

[0039] Perform edge detection on the image, use the Sobel operator to detect the edge strength of all pixels in the image, and obtain the horizontal gradient edge strength and vertical gradient edge strength , and based on the horizontal gradient edge strength and vertical gradient edge strength Get the edge strength G of each pixel:

[0040]

[0041] Get the average edge strength of the entire image based on the edge strength G of each pixel :

[0042]

[0043] Where, is the total number of pixels in the image, k is the pixel index of the image, is the edge strength of the k-th pixel;

[0044] The texture uniformity The way to obtain it is:

[0045] Convert the image into a grayscale image, obtain the grayscale co-occurrence matrix, and obtain the texture uniformity based on the grayscale co-occurrence matrix:

[0046]

[0047] Where a and b represent the two gray values ​​of the gray level co-occurrence matrix, is the relative frequency of the combination of a pixel with gray value a and its adjacent pixel with gray value b in the gray level co-occurrence matrix;

[0048] The color uniformity distribution coefficient The way to obtain it is:

[0049] Convert the image to HSV color space, including three channels: hue, saturation, and brightness. Calculate the standard deviation of each channel and obtain the color uniformity coefficient based on the standard deviation and mean of each channel. :

[0050]

[0051] Where, 、 and are the standard deviations of the hue, saturation, and brightness channels, respectively.

[0052] Preferably, the preliminary evaluation unit is used to preset a defect score threshold D1, and compare and analyze the obtained printing layer defect score D with the defect score threshold D1 to evaluate the 3D printing quality. The specific evaluation content is as follows:

[0053] If the print layer defect score D is less than the defect score threshold Dl, that is, D<Dl, the printing is determined to be in a normal risk state, indicating that the surface quality of the print layer is normal and no adjustment is required. The printing process is continuously monitored.

[0054] If the print layer defect score D is greater than or equal to the defect score threshold Dl, that is, D≥Dl, the printing is determined to be in an abnormal risk state, indicating that there are defects on the print layer surface. The alarm mechanism is triggered, and the printing is immediately stopped and the printing data is recorded and sent to the 3D printing controller until the 3D printing controller responds.

[0055] Preferably, the comprehensive evaluation module includes a printing data analysis unit and a crack probability prediction unit;

[0056] The print data analysis unit is used to obtain the average cooling rate according to the pre-processed print data set. and heat stress ;

[0057] The average cooling rate The way to obtain it is:

[0058]

[0059] Where, It's at the time When the instantaneous temperature of the printed layer is It's at the time The instantaneous temperature of the printed layer when M is the total number of sampling time points. is the i-th sampling time point, is the i-1th sampling time point, i=1, 2, ..., M, is the time interval between two adjacent data collection time points;

[0060] The thermal stress The way to obtain it is:

[0061]

[0062] Where, is the temperature difference at different locations within the printed material, is the thermal expansion coefficient of the printing material, is the effective elastic modulus of the printing material.

[0063] Preferably, the crack probability prediction unit is used to construct a crack probability prediction model and use historical printing data to train the crack probability prediction model;

[0064] The defect score D of the printed layer and the average cooling rate and heat stress The trained crack probability prediction model is input to obtain the crack generation probability Z. The crack generation probability Z is obtained as follows:

[0065]

[0066] Where, 、 and They are the defect score D of the printed layer, the average cooling rate and heat stress The weight coefficient of , q is the bias term.

[0067] Preferably, the intelligent control module is used to preset a crack generation probability threshold Z1, compare and analyze the crack generation probability threshold Z1 with the crack generation probability Z, and evaluate the 3D printing status. The specific evaluation content is as follows:

[0068] If the crack generation probability Z is greater than or equal to the crack generation probability threshold Zl, that is, Z ≥ Zl, the printing is determined to be in an abnormal state, triggering the alarm mechanism, immediately lowering the temperature to half of the original temperature, and replanning the printing path, adjusting the printing time interval between layers to twice the original time interval;

[0069] If the crack generation probability Z is less than the crack generation probability threshold Zl, that is, Z<Zl, it is determined that the printing is in a normal state and the printing state is relatively stable. The printing state is continuously monitored and a printing report is generated and stored in the 3D printing data center.

[0070] A 3D printing process intelligent control method based on real-time feedback includes the following steps:

[0071] Step 1: Collect high-definition images and printing data of the printing layer surface during the 3D printing process, and construct an image data set and a printing data set based on the collected high-definition images and printing data of the printing layer surface;

[0072] Step 2: pre-processing the high-definition image of the printed layer surface and the printing data in the image data set and the printing data set;

[0073] Step 3: Construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1. Compare and analyze the preset defect score threshold D1 with the printed layer defect score D to evaluate the printed layer status, trigger an early warning mechanism, and control the printed layer;

[0074] Step 4: Obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D;

[0075] Step 5: Preset a crack generation probability threshold Zl, and compare and analyze the preset crack generation probability threshold Zl and the crack generation probability Z to evaluate the 3D printing status. Based on the evaluated 3D printing status, feedback data is sent to the 3D printing data center to control the 3D printing process.

[0076] The present invention provides a 3D printing process intelligent control method and system based on real-time feedback, which has the following beneficial effects:

[0077] (1) The present invention realizes all-round monitoring and analysis of the surface image and printing data of the printing layer through real-time data acquisition and preprocessing. The data acquisition module obtains the temperature, stress, deposition layer thickness and high-definition image of the printing layer surface through a high-resolution camera and an intelligent sensor group, providing rich data support for subsequent analysis and judgment. The data is transmitted to the 3D printing data center through a wireless network. The system can construct and update the image data set and the printing data set in real time to ensure that the data in the printing process can be collected and processed comprehensively and quickly, avoiding the monitoring of printing quality affected by data loss. This real-time data acquisition and transmission method effectively improves the accuracy and timeliness of data acquisition and provides a solid data foundation for intelligent control.

[0078] (2) The image analysis module and comprehensive evaluation module in the system can efficiently analyze the surface defects of the printed layer and evaluate the risk of crack generation based on data such as cooling rate and thermal stress. In the image analysis module, the convolutional neural network model is used to analyze the surface image of the printed layer. It can not only extract features such as edge strength, texture uniformity and color uniformity, but also quantify and generate a defect score D, and preset a defect score threshold Dl for preliminary evaluation, thereby realizing a rapid evaluation of the surface quality of the printed layer. In addition, the comprehensive evaluation module calculates the probability of crack generation by combining the defect score, average cooling rate and thermal stress, and compares and analyzes the crack generation probability Z by presetting the crack generation probability threshold Zl, providing a comprehensive printing quality evaluation method. Compared with the traditional static monitoring method, the system can perform dynamic evaluation based on real-time data, so that the system can trigger corresponding control measures in the early stage of identifying anomalies, reducing the uncertainty and potential defects in the printing process.

[0079] (3) The intelligent control module provides timely and effective feedback control during the printing process. When the system detects that the probability of crack generation Z exceeds the preset threshold Zl, it will immediately trigger the alarm mechanism and automatically take adjustment measures, including lowering the printing temperature, replanning the printing path, and increasing the time interval between layers, thereby reducing thermal stress and cooling rate, and reducing the risk of crack generation. At the same time, when the system assesses that the printing status is normal, it will continuously monitor and generate a printing report, which will be stored in the 3D printing data center to provide a basis for subsequent quality tracking and optimization. This automated intelligent control and adjustment function can fine-tune and optimize the printing status without relying on human intervention, which not only improves the overall quality and reliability of the printed parts, but also enhances the automation and intelligence level of the 3D printing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 This is a block diagram of an intelligent control system for a 3D printing process based on real-time feedback according to the present invention.

[0081] Figure 2 This is a flow chart of a method for intelligent control of a 3D printing process based on real-time feedback according to the present invention. DETAILED DESCRIPTION

[0082] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention. Example 1

[0083] See also Figure 1 , the present invention provides a 3D printing process intelligent control system based on real-time feedback, a data acquisition module, a preprocessing module, an image analysis module, a comprehensive evaluation module and an intelligent control module;

[0084] The data acquisition module is used to collect high-definition images of the printing layer surface and printing data during the 3D printing process, and to construct an image data set and a printing data set based on the collected high-definition images of the printing layer surface and printing data;

[0085] The pre-processing module is used to pre-process the high-definition image of the printing layer surface and the printing data in the image data set and the printing data set;

[0086] The image analysis module is used to construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1, compare and analyze the preset defect score threshold D1 with the printed layer defect score D, evaluate the printed layer status, and trigger an early warning mechanism to control the printed layer;

[0087] The comprehensive evaluation module is used to obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D;

[0088] The intelligent control module is used to preset a crack generation probability threshold Zl, compare and analyze the preset crack generation probability threshold Zl with the crack generation probability Z, evaluate the 3D printing status, and feed back data to the 3D printing data center based on the evaluated 3D printing status to control the 3D printing process.

[0089] In this embodiment, through precise data acquisition, preprocessing, image analysis, comprehensive evaluation and intelligent control modules, the defects and crack generation risks in the printing process are monitored in real time, ensuring the stability and consistency of printing quality. The system uses dual acquisition of high-definition images and printing data, combined with convolutional neural networks for defect analysis, to effectively identify tiny defects on the surface of the printed layer, and optimize the printing process by dynamically adjusting printing parameters to reduce the occurrence of quality problems. In addition, through a comprehensive evaluation of the probability of crack generation, the system can predict and avoid potential structural defects in advance, improve production efficiency, reduce material waste and reduce the defective rate. Overall, the system realizes a fully automated intelligent control and feedback mechanism, which not only improves printing quality and accuracy, but also optimizes the production process, reduces human intervention and operational errors, and promotes the application of 3D printing technology in efficient and refined production. Example 2

[0090] Please refer to Figure 1 , specifically:

[0091] The data acquisition module includes an image data acquisition unit and a print data acquisition unit;

[0092] The image data acquisition unit is used to use a high-resolution camera installed above the print head to collect the surface image of the printed layer after each layer printing process is completed, and transmit the collected surface image of the printed layer to the 3D printing data center via wireless network technology to construct an image data set;

[0093] The printing data acquisition unit is used to deploy an intelligent sensor group at the bottom of the 3D printer platform to obtain printing data of the 3D printing process, wherein the printing data includes the printing layer temperature, printing layer stress and deposition layer thickness, and obtains the thermal expansion coefficient and effective elastic modulus of the material through the physical property database of the printing material and the material manual;

[0094] The printed layer temperature data refers to the use of infrared temperature sensors to monitor the temperature of the printed layer in real time. Temperature acquisition helps monitor and control the heating and cooling rates of the material during the printing process, avoiding changes in material properties or poor interlayer bonding due to temperature fluctuations;

[0095] The printed layer stress refers to the use of a laser stress plate installed at the bottom of the printing platform to monitor the stress changes generated during the printing process in real time and obtain the printed layer stress data;

[0096] The deposited layer thickness refers to the use of a deployed optical sensor to detect the height change of the printed layer through the reflected light signal to obtain the deposited layer thickness data. The deposited layer refers to each layer of material formed by the nozzle depositing the molten material layer by layer on the printing platform or the printed layer during the 3D printing process;

[0097] The obtained printing layer temperature data, printing layer stress data and deposition layer thickness data are transmitted to the 3D printing data center via wireless network technology to construct a printing data set.

[0098] In this embodiment, through the real-time collection of 3D printing process data, strong data support is provided for subsequent printing process control and quality analysis. The image data acquisition unit uses a high-resolution camera installed above the print head to collect high-precision surface images of the printed layer after each layer is printed. This image acquisition method can accurately capture the detailed structure of the printed layer, forming a high-definition image data set, which provides basic data for subsequent defect detection and print quality assessment. The application of wireless network technology enables image data to be quickly transmitted to the 3D printing data center, ensuring efficient and delay-free transmission of image data, and realizing real-time monitoring of the printing process. The printing data set stored in the data center helps to analyze abnormal changes in the printing process and establish a long-term management system for historical data, providing an important reference for subsequent intelligent control and optimization. By integrating real-time monitoring and data aggregation of multiple sensors, the system can promptly detect and respond to any abnormal conditions that occur during the printing process, ensuring the smooth progress of the 3D printing process and the quality of the final product. Example 3

[0099] Please refer to Figure 1 , specifically:

[0100] The pre-processing module includes a print data processing unit and an image processing unit;

[0101] The print data processing unit is used to perform noise filtering, data cleaning and normalization processing on the print data in the print data set;

[0102] The noise filtering refers to using a Kalman filter to remove random noise from the print data;

[0103] The data cleaning refers to detecting outliers using a box plot, calculating the interquartile range (IQR) in the box plot, and obtaining the normal data range (Q1-1.5×IQR, Q3+1.5×IQR). Printed data that is not within the normal data range is considered abnormal data and is removed.

[0104] The above Q1 and Q3 are the 1st quantile and the 3rd quantile respectively. By arranging the data from small to large, the data at the 25% and 75% positions in the data set are obtained as the 1st quantile and the 3rd quantile;

[0105] The normalization process refers to performing dimensionless processing on the collected printing data, and unifying the units of the collected printing data.

[0106] The image processing unit is to perform region cropping, RGB standardization and boundary setting on the image data in the image data set;

[0107] The area cropping refers to using computer vision technology to crop out the printing area and remove background information;

[0108] RGB normalization refers to performing pixel operations on image data, standardizing each pixel to reduce the impact of lighting changes;

[0109] The boundary setting refers to marking image boundary points using an edge detection algorithm.

[0110] In this embodiment, high-quality and accurate data support is provided for the subsequent intelligent control of the 3D printing process through multi-level cleaning and standardization operations on the printing data and image data. First, the printing data processing unit uses a Kalman filter to remove random noise in the data to ensure the stability and accuracy of temperature, stress and deposition layer thickness. At the same time, through box plot detection and elimination of outliers, abnormal data that may affect the analysis results are effectively filtered out, and the dimensional differences of various types of data are eliminated through normalization processing, making the comparison and calculation between data more consistent and scientific. These steps ensure the stability and reliability of the printing data and lay the foundation for the accurate prediction of the probability of crack generation. Secondly, the image processing unit uses computer vision technology to perform regional cropping on the printed image, remove background information, and make the image more focused on the printing area. Through RGB standardization, the influence of illumination changes on the consistency of image color is reduced. The image boundary is set through edge detection technology to clarify the effective area of ​​image analysis. These preprocessing measures improve the quality of image data, provide standardized data input for subsequent convolutional neural network defect analysis, ensure the accuracy and consistency of defect scoring results, enhance the precision and consistency of data, enable the intelligent control system to make accurate judgments based on high-quality data, effectively reduce the risk of error accumulation and misjudgment, thereby improving the stability and reliability of the printing process and ensuring the quality and structural integrity of the printed parts. Example 4

[0111] Please refer to Figure 1 , specifically:

[0112] The image analysis module includes a model building unit and a preliminary evaluation unit;

[0113] The model building unit is used to build a defect analysis model based on convolutional neural network technology, and in combination with a 3D printing database, obtain historical printing image data, input the historical printing image data as training data into the input layer of the defect analysis model, use back propagation and optimization algorithms to adjust the defect analysis model parameters, input the real-time collected printing image data into the trained defect analysis model, and obtain a print layer defect score D;

[0114] The defect analysis model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer;

[0115] The input layer is used to input the real-time collected printing image data into the defect analysis model;

[0116] The convolutional layer is used to extract local features of image data and generate a local feature map;

[0117] The pooling layer is used to reduce the dimension of the local feature map obtained by the convolutional layer;

[0118] The fully connected layer is used to map the features to the printed layer defect score D output;

[0119] The output layer is used to output the print layer defect score D. The print layer defect score D is specifically obtained as follows:

[0120]

[0121] Where, is the average edge strength, is the texture uniformity, is the color uniformity coefficient, 、 and are the weight values ​​of average edge intensity, texture uniformity and color uniformity coefficient respectively, and c is the bias term;

[0122] The average edge strength The way to obtain it is:

[0123] Perform edge detection on the image, use the Sobel operator to detect the edge strength of all pixels in the image, and obtain the horizontal gradient edge strength and vertical gradient edge strength , and based on the horizontal gradient edge strength and vertical gradient edge strength Get the edge strength G of each pixel:

[0124]

[0125] Get the average edge strength of the entire image based on the edge strength G of each pixel :

[0126]

[0127] Where, is the total number of pixels in the image, k is the pixel index of the image, is the edge strength of the k-th pixel;

[0128] The Sobel operator is an edge detection operator used in image processing to extract edge information from an image. It estimates the gradient of an image by calculating the grayscale change rate of a pixel in the horizontal and vertical directions, thereby highlighting areas in the image where grayscale changes dramatically.

[0129] The texture uniformity The way to obtain it is:

[0130] Convert the image into a grayscale image, obtain the grayscale co-occurrence matrix, and obtain the texture uniformity based on the grayscale co-occurrence matrix:

[0131]

[0132] Where a and b represent the two gray values ​​of the gray level co-occurrence matrix, is the relative frequency of the combination of a pixel with gray value a and its adjacent pixel with gray value b in the gray level co-occurrence matrix;

[0133] The color uniformity distribution coefficient The way to obtain it is:

[0134] Convert the image to HSV color space, including three channels: hue, saturation, and brightness. Calculate the standard deviation of each channel and obtain the color uniformity coefficient based on the standard deviation and mean of each channel. :

[0135]

[0136] Where, 、 and are the standard deviations of the hue, saturation, and brightness channels, respectively.

[0137] The standard deviations of the hue channel, saturation channel, and brightness channel are obtained by calculating the standard deviations of the pixel values ​​of the hue channel, saturation channel, and brightness channel;

[0138] The preliminary evaluation unit is used to preset a defect score threshold D1, and compare and analyze the obtained printing layer defect score D with the defect score threshold D1 to evaluate the 3D printing quality. The specific evaluation content is as follows:

[0139] If the print layer defect score D is less than the defect score threshold Dl, that is, D<Dl, the printing is determined to be in a normal risk state, indicating that the surface quality of the print layer is normal and no adjustment is required. The printing process is continuously monitored.

[0140] If the print layer defect score D is greater than or equal to the defect score threshold Dl, that is, D≥Dl, the printing is determined to be in an abnormal risk state, indicating that there are defects on the print layer surface. The alarm mechanism is triggered, and the printing is immediately stopped and the printing data is recorded and sent to the 3D printing controller until the 3D printing controller responds.

[0141] In this embodiment, by constructing a defect analysis model based on a convolutional neural network (CNN), efficient identification and evaluation of surface defects of printed layers during 3D printing are achieved, which has significant beneficial effects. First, through the model construction unit, the system uses historical printing image data to train the defect analysis model, and uses back propagation and optimization algorithms to continuously adjust the model parameters, so that the model can accurately extract features such as edges, textures, and colors of the printed layer surface. During the real-time printing process, the system inputs the collected images into the trained model and automatically calculates the defect score D. This score combines the average edge strength, texture uniformity, and color uniformity coefficient of the printed layer, making the defect detection results accurate and reliable. Secondly, the preliminary evaluation unit evaluates the printing quality by performing real-time comparative analysis between the preset defect score threshold D1 and the defect score D. When the defect score D is lower than the defect score threshold D1, the system determines that the printing process is in a normal state and continuously monitors without additional intervention. When the defect score D is greater than or equal to the defect score threshold D1, the system determines that it is in an abnormal risk state, immediately triggers the alarm mechanism, stops printing, and sends the data to the control personnel for further processing. The real-time feedback and intelligent control functions of this module improve the automation and intelligence level of the 3D printing process, can effectively reduce the time and energy required for human intervention, and avoid subsequent printing quality problems caused by surface defects of the printing layer. Example 5

[0142] Please refer to Figure 1 , specifically:

[0143] The comprehensive evaluation module includes a printing data analysis unit and a crack probability prediction unit;

[0144] The print data analysis unit is used to obtain the average cooling rate according to the pre-processed print data set. and heat stress ;

[0145] The average cooling rate The way to obtain it is:

[0146]

[0147] Where, It's at the time When the instantaneous temperature of the printed layer is It's at the time The instantaneous temperature of the printed layer when M is the total number of sampling time points. is the i-th sampling time point, is the i-1th sampling time point, i=1, 2, ..., M, is the time interval between two adjacent data collection time points;

[0148] The thermal stress The way to obtain it is:

[0149]

[0150] Where, is the temperature difference at different locations within the printed material, is the thermal expansion coefficient of the printing material, is the effective elastic modulus of the printing material.

[0151] The crack probability prediction unit is used to construct a crack probability prediction model, wherein the crack probability prediction model is constructed by using a random forest algorithm, and the crack probability prediction model is trained using historical printing data, and the crack probability prediction model parameters are optimized using a k-fold cross validation method;

[0152] The defect score D of the printed layer and the average cooling rate and heat stress The trained crack probability prediction model is input to obtain the crack generation probability Z. The crack generation probability Z is obtained as follows:

[0153]

[0154] Where, 、 and They are the defect score D of the printed layer, the average cooling rate and heat stress The weight coefficient of , q is the bias term.

[0155] In this embodiment, the print data analysis unit and the crack probability prediction unit are used to achieve efficient evaluation and accurate prediction of the print layer quality and crack risk, which has significant beneficial effects. The print data analysis unit first calculates the average cooling rate and thermal stress based on the preprocessed print data set. By real-time monitoring and calculating the instantaneous temperature at different time points during the printing process, the change trend of the cooling rate is analyzed, and the problem of thermal stress concentration caused by excessive cooling rate can be discovered in time, avoiding the impact of drastic temperature changes on the print layer quality. The thermal stress is obtained based on the temperature difference, thermal expansion coefficient and effective elastic modulus of the material, which effectively reflects the internal stress generated by temperature unevenness in the material. These data not only provide key information for the subsequent crack probability prediction, but also provide reference for the development of the industry. The key parameters also improve the comprehensiveness and accuracy of data analysis. The crack probability prediction unit is based on the obtained print layer defect score, average cooling rate and thermal stress. By constructing a crack probability prediction model and training the model with historical printing data, it can achieve accurate prediction of the probability of crack generation in the print layer. The three parameters of defect score, cooling rate and thermal stress are weighted and integrated to evaluate the influence of micro defects, temperature control and stress distribution on the surface of the print layer on crack generation, and finally obtain the crack generation probability Z. The crack generation probability Z can reduce the risk of crack generation in the printing process, improve the overall quality and structural integrity of the printed part, reduce rework and material waste caused by crack defects, and realize high-precision intelligent control in the 3D printing process. Example 6

[0156] Please refer to Figure 1 , specifically:

[0157] The intelligent control module is used to preset a crack generation probability threshold Z1, compare and analyze the crack generation probability threshold Z1 with the crack generation probability Z, and evaluate the 3D printing status. The specific evaluation content is as follows:

[0158] If the crack generation probability Z is greater than or equal to the crack generation probability threshold Zl, that is, Z ≥ Zl, the printing is determined to be in an abnormal state, triggering the alarm mechanism, immediately lowering the temperature to half of the original temperature, and replanning the printing path, adjusting the printing time interval between layers to twice the original time interval;

[0159] If the crack generation probability Z is less than the crack generation probability threshold Zl, that is, Z<Zl, it is determined that the printing is in a normal state and the printing state is relatively stable. The printing state is continuously monitored and a printing report is generated and stored in the 3D printing data center.

[0160] In this embodiment, by comparing and analyzing the crack generation probability threshold Zl with the crack generation probability Z, when the crack generation probability Z is greater than or equal to the crack generation probability threshold Zl, the printing state is automatically determined to be abnormal and the alarm mechanism is triggered, which can effectively reduce the thermal stress and cooling rate inside the material, alleviate the structural instability caused by temperature fluctuations, and thus significantly reduce the risk of crack generation. On the contrary, when the crack generation probability Z is less than the crack generation probability threshold Zl, the system determines that the printing state is normal, keeps the printing parameters unchanged, and continuously monitors the printing state. At this time, the system will generate a real-time printing status report and store it in the 3D printing data. This automated real-time feedback and control function enables the system to respond immediately when risks are detected without human intervention, ensuring the molding quality and precision of printed parts. At the same time, the dynamic adjustment and data storage functions not only optimize the printing process, but also provide valuable experience data for future printing tasks, thereby gradually improving the stability and reliability of the printing process and the intelligence level of the system. Through this intelligent control module, the system achieves high-precision quality control and autonomous optimization of the printing process, significantly reducing the incidence of cracks and improving the performance and durability of printed parts. Example 7

[0161] Please refer to Figure 2 , specifically:

[0162] A 3D printing process intelligent control method based on real-time feedback includes the following steps:

[0163] Step 1: Collect high-definition images and printing data of the printing layer surface during the 3D printing process, and construct an image data set and a printing data set based on the collected high-definition images and printing data of the printing layer surface;

[0164] Step 2: pre-processing the high-definition image of the printed layer surface and the printing data in the image data set and the printing data set;

[0165] Step 3: Construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1. Compare and analyze the preset defect score threshold D1 with the printed layer defect score D to evaluate the printed layer status, trigger an early warning mechanism, and control the printed layer;

[0166] Step 4: Obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D;

[0167] Step 5: Preset a crack generation probability threshold Zl, and compare and analyze the preset crack generation probability threshold Zl and the crack generation probability Z to evaluate the 3D printing status. Based on the evaluated 3D printing status, feedback data is sent to the 3D printing data center to control the 3D printing process.

[0168] In this embodiment, by real-time acquisition and preprocessing of printing data, combined with a convolutional neural network model for defect identification and status assessment, and realizing intelligent feedback control through crack generation probability prediction, the system can identify potential defects and assess the risk of crack generation in real time during the printing process. By automatically adjusting printing parameters, it can effectively prevent and reduce cracks and structural defects in the printing process, which not only improves printing accuracy and finished product quality, but also realizes efficient and intelligent management of the printing process, thereby improving the stability of 3D printing.

[0169] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent control system for 3D printing process based on real-time feedback, characterized by: It includes data acquisition module, pre-processing module, image analysis module, comprehensive evaluation module and intelligent control module; The data acquisition module is used to collect high-definition images of the printing layer surface and printing data during the 3D printing process, and to construct an image data set and a printing data set based on the collected high-definition images of the printing layer surface and printing data; The pre-processing module is used to pre-process the high-definition image of the printing layer surface and the printing data in the image data set and the printing data set; The image analysis module is used to construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1, compare and analyze the preset defect score threshold D1 with the printed layer defect score D, evaluate the printed layer status, and trigger an early warning mechanism to control the printed layer; The comprehensive evaluation module is used to obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D; The intelligent control module is used to preset a crack generation probability threshold Zl, and compare and analyze the preset crack generation probability threshold Zl with the crack generation probability Z to evaluate the 3D printing status. Based on the evaluated 3D printing status, the module feeds back data to the 3D printing data center to control the 3D printing process. The image analysis module includes a model building unit and a preliminary evaluation unit; The model building unit is used to build a defect analysis model based on convolutional neural network technology, and in combination with a 3D printing database, obtain historical printing image data, input the historical printing image data as training data into the input layer of the defect analysis model, use back propagation and optimization algorithms to adjust the defect analysis model parameters, input the real-time collected printing image data into the trained defect analysis model, and obtain a print layer defect score D; The defect analysis model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer; The input layer is used to input the real-time collected printing image data into the defect analysis model; The convolutional layer is used to extract local features of image data and generate a local feature map; The pooling layer is used to reduce the dimension of the local feature map obtained by the convolutional layer; The fully connected layer is used to map the features to the printed layer defect score D output; The output layer is used to output the print layer defect score D. The print layer defect score D is specifically obtained as follows: ； Where, is the average edge strength, is the texture uniformity, is the color uniformity coefficient, 、 and are the weight values ​​of average edge intensity, texture uniformity and color uniformity coefficient respectively, and c is the bias term; The average edge strength The way to obtain it is: Perform edge detection on the image, use the Sobel operator to detect the edge strength of all pixels in the image, and obtain the horizontal gradient edge strength and vertical gradient edge strength , and based on the horizontal gradient edge strength and vertical gradient edge strength Get the edge strength G of each pixel: ； Get the average edge strength of the entire image based on the edge strength G of each pixel : ； Where, is the total number of pixels in the image, k is the pixel index of the image, is the edge strength of the k-th pixel; The texture uniformity The way to obtain it is: Convert the image into a grayscale image, obtain the grayscale co-occurrence matrix, and obtain the texture uniformity based on the grayscale co-occurrence matrix: ； Where a and b represent the two gray values ​​of the gray level co-occurrence matrix, is the relative frequency of the combination of a pixel with gray value a and its adjacent pixel with gray value b in the gray level co-occurrence matrix; The color uniformity distribution coefficient The way to obtain it is: Convert the image to HSV color space, including three channels: hue, saturation, and brightness. Calculate the standard deviation of each channel and obtain the color uniformity coefficient based on the standard deviation and mean of each channel. : ； Where, 、 and are the standard deviations of the hue, saturation, and brightness channels, respectively; The preliminary evaluation unit is used to preset a defect score threshold D1, and compare and analyze the obtained printing layer defect score D with the defect score threshold D1 to evaluate the 3D printing quality. The specific evaluation content is as follows: If the print layer defect score D is less than the defect score threshold Dl, that is, D<Dl, the printing is determined to be in a normal risk state, indicating that the surface quality of the print layer is normal and no adjustment is required. The printing process is continuously monitored. If the print layer defect score D is greater than or equal to the defect score threshold Dl, that is, D≥Dl, the printing is determined to be in an abnormal risk state, indicating that there are defects on the print layer surface. The alarm mechanism is triggered, and the printing is immediately stopped and the printing data is recorded and sent to the 3D printing controller until the 3D printing controller responds.

2. The 3D printing process intelligent control system based on real-time feedback according to claim 1, characterized in that: The data acquisition module includes an image data acquisition unit and a print data acquisition unit; The image data acquisition unit is used to use a high-resolution camera installed above the print head to collect the surface image of the printed layer after each layer printing process is completed, and transmit the collected surface image of the printed layer to the 3D printing data center via wireless network technology to construct an image data set; The printing data acquisition unit is used to deploy an intelligent sensor group at the bottom of the 3D printer platform to obtain printing data of the 3D printing process, wherein the printing data includes the printing layer temperature, printing layer stress and deposition layer thickness, and obtains the thermal expansion coefficient and effective elastic modulus of the material through the physical property database of the printing material and the material manual; The printed layer temperature data refers to the temperature of the printed layer obtained by real-time monitoring using an infrared temperature sensor; The printed layer stress refers to the use of a laser stress plate installed at the bottom of the printing platform to monitor the stress changes generated during the printing process in real time and obtain the printed layer stress data; The thickness of the deposited layer refers to the use of a deployed optical sensor to detect the height change of the printed layer through the reflected light signal to obtain the thickness data of the deposited layer; The obtained printing layer temperature data, printing layer stress data and deposition layer height data are transmitted to the 3D printing data center via wireless network technology to construct a printing data set.

3. The 3D printing process intelligent control system based on real-time feedback according to claim 2, characterized in that: The pre-processing module includes a print data processing unit and an image processing unit; The print data processing unit is used to perform noise filtering, data cleaning and normalization processing on the print data in the print data set; The noise filtering refers to using a Kalman filter to remove random noise from the print data; The data cleaning refers to detecting outliers using a box plot, calculating the interquartile range (IQR) in the box plot, and obtaining the normal data range (Q1-1.5×IQR, Q3+1.5×IQR). Printed data that is not within the normal data range is considered abnormal data and is removed. The normalization process refers to performing dimensionless processing on the collected printing data, and unifying the units of the collected printing data.

4. The 3D printing process intelligent control system based on real-time feedback according to claim 3, characterized in that: The image processing unit is to perform region cropping, RGB standardization and boundary setting on the image data in the image data set; The area cropping refers to using computer vision technology to crop out the printing area and remove background information; The RGB standardization refers to performing pixel operations on image data and performing standard version processing on each pixel; The boundary setting refers to marking image boundary points using an edge detection algorithm.

5. The 3D printing process intelligent control system based on real-time feedback according to claim 4, characterized in that: The comprehensive evaluation module includes a printing data analysis unit and a crack probability prediction unit; The print data analysis unit is used to obtain the average cooling rate according to the pre-processed print data set. and heat stress ; The average cooling rate The way to obtain it is: ； Where, It's at the time When the instantaneous temperature of the printed layer is It's at the time The instantaneous temperature of the printed layer when M is the total number of sampling time points. is the i-th sampling time point, is the i-1th sampling time point, i=1, 2, ..., M, is the time interval between two adjacent data collection time points; The thermal stress The way to obtain it is: ； Where, is the temperature difference at different locations within the printed material, is the thermal expansion coefficient of the printing material, is the effective elastic modulus of the printing material.

6. The 3D printing process intelligent control system based on real-time feedback according to claim 5, characterized in that: The crack probability prediction unit is used to construct a crack probability prediction model and use historical printing data to train the crack probability prediction model; The defect score D of the printed layer and the average cooling rate and heat stress The trained crack probability prediction model is input to obtain the crack generation probability Z. The crack generation probability Z is obtained as follows: ； Where, 、 and They are the defect score D of the printed layer, the average cooling rate and heat stress The weight coefficient of , q is the bias term.

7. The intelligent control system for 3D printing process based on real-time feedback according to claim 6, characterized in that: The intelligent control module is used to preset a crack generation probability threshold Zl, compare and analyze the crack generation probability threshold Zl with the crack generation probability Z, and evaluate the 3D printing status. The specific evaluation content is as follows: If the crack generation probability Z is greater than or equal to the crack generation probability threshold Zl, that is, Z ≥ Zl, the printing is determined to be in an abnormal state, triggering the alarm mechanism, immediately lowering the temperature to half of the original temperature, and replanning the printing path, adjusting the printing time interval between layers to twice the original time interval; If the crack generation probability Z is less than the crack generation probability threshold Zl, that is, Z<Zl, it is determined that the printing is in a normal state and the printing state is relatively stable. The printing state is continuously monitored and a printing report is generated and stored in the 3D printing data center.

8. A method for intelligent control of a 3D printing process based on real-time feedback, using the intelligent control system for a 3D printing process based on real-time feedback according to any one of claims 1 to 7, characterized in that: The following steps are included: Step 1: Collect high-definition images and printing data of the printing layer surface during the 3D printing process, and construct an image data set and a printing data set based on the collected high-definition images and printing data of the printing layer surface; Step 2: pre-processing the high-definition image of the printed layer surface and the printing data in the image data set and the printing data set; Step 3: Construct a defect analysis model based on a convolutional neural network, input the preprocessed high-definition image of the printed layer surface into the defect analysis model, obtain the printed layer defect score D, and preset a defect score threshold D1. Compare and analyze the preset defect score threshold D1 with the printed layer defect score D to evaluate the printed layer status, trigger an early warning mechanism, and control the printed layer; Step 4: Obtain the crack generation probability Z based on the preprocessed printing data set and the printing layer defect score D; Step 5: Preset a crack generation probability threshold Zl, and compare and analyze the preset crack generation probability threshold Zl and the crack generation probability Z to evaluate the 3D printing status. Based on the evaluated 3D printing status, feedback data is sent to the 3D printing data center to control the 3D printing process.

Images