Real-time Monitoring and Control Method for 3D Printing Process of Laser Sintered Powder Materials

Abstract

The present invention provides a method for real-time monitoring and regulation in the 3D printing process of laser-sintered powdered materials. Step 100: The process of data collection by the multi-dimensional perception system. Step 200: The step of data processing by the data processing system. Step 300: The process of calculation and decision-making by the intelligent decision-making system. Step 400: The process of controlling the printing by the intelligent control system. The method of the present invention is applicable to the high-precision preparation of laser-sintered powdered material specimens in rock mechanics experiments. Through multi-modal information fusion and closed-loop control strategies, the accurate replication of complex structural features such as micro-cracks and pores inside natural rocks can be achieved, providing key technical support for the research of rock engineering materials.

Description

Technical Field

[0001] The present invention relates to the technical field of additive manufacturing, and specifically to a real-time monitoring and regulation module for the laser 3D printing process that integrates multi-sensor data acquisition, digital twin modeling, visualization analysis, and intelligent control technologies. Background Art

[0002] With the continuous development of 3D printing technology, its significant advantages in aspects such as personalized customization, complex structure forming, and rapid prototype development are deeply integrating into multiple fields such as industrial manufacturing, medical and health, construction engineering, and aerospace, driving the innovation and upgrading of traditional manufacturing paradigms. In the field of rock mechanics research, traditional experimental methods face dual technical bottlenecks: the randomness of internal defects in natural rock specimens (such as the uncontrollability of crack distribution and the repeatability problem of standardized preparation) and the scale effect limitation of engineering simulation, resulting in the repeatability of experimental results and the measurement accuracy of parameters being difficult to meet the requirements of modern research. The emergence of the 3D printing technology of laser sintering powdered materials (such as laser ceramics) provides an innovative solution path for this field. Through the "CT scanning - digital modeling - 3D printing" technology chain, high-precision digital replication of the internal microstructure of natural rocks can be achieved, providing a reliable technical means for the standardized development of rock mechanics experiments.

[0003] However, when using the 3D printing technology of laser sintering powdered materials to prepare rock specimens with complex internal structures, especially when it is necessary to accurately reproduce the microcrack characteristics of natural rocks, micron-level printing errors can lead to distortion of crack morphology (such as blockage, offset, etc.). Since it is difficult to effectively evaluate the quality of internal defect replication of the specimen after printing through conventional non-destructive testing means, there are significant differences in the physical and mechanical properties between the final specimen and the original rock, thereby significantly reducing the accuracy and credibility of experimental data. Therefore, constructing a real-time defect monitoring module during the printing process has become a key technical requirement for the engineering application of this technology.

[0004] Currently, the 3D printing technology of laser sintering powdered materials has four technical limitations: (1) There is a problem of missing information dimensions in single physical quantity monitoring (such as only monitoring the molten pool temperature field); (2) There is a lack of a real-time tracking mechanism for the printing path, making it difficult to achieve closed-loop control of the process; (3) The path deviation compensation mechanism relies on offline calibration and cannot adapt to dynamic printing conditions; (4) Defect repair requires manual intervention, restricting the improvement of printing efficiency. To break through the above technical bottlenecks, it is urgent to develop a comprehensive monitoring module that integrates intelligent monitoring, parameter adaptive adjustment, and printing process control to reduce the dependence on professional technical personnel.

[0005] The monitoring method proposed by the present invention adopts multi-physical field coupling real-time monitoring and intelligent regulation technology. By constructing a multi-modal perception network, developing an intelligent decision-making algorithm, and designing a closed-loop control module, precise regulation of the entire 3D printing process is achieved. The implementation of this technical solution will effectively improve the accuracy and reliability of rock specimen preparation, providing new technical support for rock mechanics experimental research. Summary of the Invention

[0006] The present invention improves the technical bottlenecks existing in the prior art, such as single monitoring dimension, insufficient response timeliness, and low defect repair efficiency. By building a multi-modal monitoring network including laser power, sintering zone temperature field, sintering process images, and spot trajectory parameters, and integrating spatio-temporal synchronous data acquisition technology and intelligent control algorithms, closed-loop regulation of the entire printing process is realized. This module innovatively adopts a deep coupling architecture of energy flow monitoring, geometric shape control, and environmental perception, effectively breaking through the dual constraints of traditional additive manufacturing technology in terms of real-time regulation ability and module robustness.

[0007] To solve the problems in the prior art, the present invention provides a method for real-time monitoring and regulation of the 3D printing process of laser-sintered powder materials, including the following steps:

[0008] Step 100, the multi-dimensional perception system collects data: Obtain real-time data during the printing process, including laser power, powder bed temperature, sintering zone temperature, laser spot temperature, sintering zone images, and laser spot path information data; Add a timestamp to the data frame header, and pack and send the laser power data, temperature field data, and image data to the data processing system;

[0009] Step 200, the data processing system processes the data: Process the real-time collected data, including time alignment, noise filtering, abnormal data cleaning, and image conversion, extract important features of the image and the spot position, and establish a six-dimensional data cube: X, Y, Z, P, T, t, with a time alignment error ≤ 1 μs, construct a multi-physical field data fusion analysis framework, where X, Y, and Z represent spatial positions, P represents laser power, T represents temperature, and t represents time, save the data and transmit the data to the intelligent decision-making system through the main control board;

[0010] Step 300, the intelligent decision-making system analyzes and compares to generate the result of data feedback: Compare the real-time collected data with the original preset standard data, identify defects in the image data, compare the actual printing path with the preset printing path, and take corresponding decision-making measures according to the comparison results; The defect identification method is as follows:

[0011] Image defect identification;

[0012] Path deviation monitoring;

[0013] Path compensation threshold calculation;

[0014] Path prediction and compensation value calculation;

[0015] The adjustment strategy generation module generates different adjustment strategies according to different signals;

[0016] The results of generating data feedback include deviation values, defect marks, and adjustment strategies;

[0017] Step 400, the intelligent control system controls the printing, adjusts the printing process and parameters according to the results of data feedback, and activates the corresponding response mechanism to ensure that the printing process proceeds accurately.

[0018] As a further improvement of the present invention, the path compensation threshold is calculated as follows:

[0019] (1) Gaussian mixture model GMM: In path deviation monitoring, a GMM model is built for historical normal path deviation data to obtain the probability distribution of different deviation values. As the printing process progresses, the GMM model is continuously updated according to new data, and the adaptive coefficient is obtained through reinforcement learning based on the probability distribution, and then the path compensation threshold is dynamically adjusted;

[0020] (2) The calculation method of the compensation threshold is:

[0021]

[0022] Where α 、 β and δ are adaptive coefficients obtained by the reinforcement learning algorithm through sorting and analyzing historical printing data.

[0023] As a further improvement of the present invention, the path prediction and compensation mechanism is as follows:

[0024] (1) Establishment of the printing path dynamic model and prediction path establishment: According to the principle of the model predictive control algorithm MPC, a printing path dynamic model is established:

[0025]

[0026] Where represents the predicted printing state vector at the next moment, represents the current printing state vector, including the relevant states of the galvanometer, the current printing coordinates, and path information; A and B are coefficient matrices that describe the influence of state transition and control input on the state, and the coefficients of A and B can both be calculated from historical printing state change data; is the control input signal function at the current k-1 moment, is the process noise, which follows a normal distribution with a mean of 0 and a covariance matrix of Q; through this printing path dynamics model, the printing state and the coordinates of the printing position point at the next moment within a very short time can be predicted based on the current printing state; after obtaining the printing coordinates at the next moment, the current printing coordinates and the predicted coordinate values are connected by a cubic line interpolation algorithm to obtain the predicted path;

[0027] (2) Optimize the predicted path: Optimize the objective function through the model predictive control algorithm MPC:

[0028]

[0029] In this objective function, represents the predicted state vector, represents the desired state vector, which is the preset printing state. Q and R are the weight matrices of the state error and the control input respectively; within a certain prediction horizon N, by optimizing the control input , the actual state is made as close as possible to the desired state , which is measured by . The Q matrix can adjust the weights of different state variables and at the same time limit the size of the control input, which is realized by . The R matrix is used to adjust the weights of different control input variables to avoid unnecessary impacts on the equipment; by optimizing this objective function, the optimal control input sequence can be obtained to achieve the collaborative optimization of path compensation correction and optimization; when the finally calculated optimal path compensation value is 0, no path compensation operation is performed, and when it is not 0, compensation is required. The final compensation amount is compared with the compensation threshold. When the final compensation amount is less than the compensation threshold, the printing path is adjusted directly in one step. When the compensation amount exceeds the threshold, the printing path is adjusted step by step to avoid damaging the printing equipment due to excessive compensation amount.

[0030] As a further improvement of the present invention, in step 300, the image defect recognition method is as follows: The defect recognition monitoring module constructs a high-precision defect recognition system based on the CNN-YOLOv8n hybrid neural network; this architecture innovatively integrates the CSPDarknet53 backbone network and the PANet path aggregation structure; the system adopts a transfer learning strategy and is intensively trained based on more than 100,000 high-precision labeled samples. The samples cover typical printing defects, forming a perfect defect feature recognition model;

[0031] In the actual printing monitoring process, this module uses the input image through the adaptive NMS dynamic threshold adjustment algorithm:

[0032]

[0033] The function of this formula is to dynamically adjust the threshold of non-maximum suppression according to the heights of different layers in the neural network. Among them, 0.5 is a basic threshold, which is the initial threshold when the factor of the height of the neural network layer is not considered, and 0.01 is a proportionality coefficient used to control the influence degree of the height of the neural network layer on the threshold. As the height of the neural network layer increases, the threshold will gradually increase by 0.01 times the layer height.

[0034] Through the adaptive NMS dynamic threshold adjustment algorithm, the image defect monitoring module can perform millisecond-level defect detection on real-time images collected at high speed. After detecting a defect, the system immediately generates structured data and sends a feedback instruction to the adjustment strategy module to trigger subsequent adjustment strategies.

[0035] As a further improvement of the present invention, in step 300, the path deviation monitoring method is as follows:

[0036] Dynamic time warping algorithm: The printing path is transformed into a time series for processing. For the actual printing path, the time stamps and position coordinates of the collected images are converted into time series information as the actual printing time series. For the preset printing path, according to parameters such as the designed printing texture and speed, the printing position at the theoretical moment is generated, and this theoretical moment and the theoretical printing position are used as the preset printing time series, where the time series of the actual printing path is:

[0037] ,

[0038] where represents the position vector of the actual path at the i-th time point, that is:

[0039] ;

[0040] The time series of the preset printing path is:

[0041] ,

[0042] Similarly represents the position vector of the preset path at the j-th time point, that is:

[0043] ,

[0044] where i = j, that is, the actual printing coordinates and the preset printing coordinates at the same printing moment;

[0045] Taking the actual path time series as rows and the preset path time series as columns, a two-dimensional distance matrix D is constructed. The matrix D is an m×n matrix, and the deviation between the actual coordinates and the preset coordinates is calculated:

[0046]

[0047] Each element in the matrix represents the distance between the position at a certain moment in the actual path and the position at a certain moment in the preset path, and the sum of the distances within this extremely short time is recorded as the path deviation value.

[0048] As a further improvement of the present invention, in step 300, the adjustment strategy generation module generates different adjustment strategies according to different signals, and the specific method is as follows:

[0049] (1) Construct a laser power-temperature two-parameter collaborative regulation model. When the monitored value is within the safe threshold range, the module triggers three types of response strategies based on the threshold comparison result:

[0050] Equivalent state: Maintain the current parameters;

[0051] Positive deviation: Execute the parameter attenuation instruction, and the power / temperature decreases;

[0052] Negative deviation: Start the parameter enhancement instruction, and the power / temperature increases;

[0053] (2) Molding path compensation module: The module receives the compensation amount data packet from the online monitoring module and generates a smooth transition path by using an adaptive interpolation algorithm.

[0054] As a further improvement of the present invention, in step 300, when the intelligent decision-making system calculates and makes a decision:

[0055] The data parsing module receives the data transmitted by the data processing system and parses the data of the original print file;

[0056] The data synchronization module uses the PTP + Kalman filtering algorithm with the print timestamp as the standard to align the collected data with the preset data of the original print file, and only aligns the data time for the image data;

[0057] The feature extraction module extracts four data of laser power, temperature, image, and spot path at each time point;

[0058] The feature extraction module includes the following sub-steps:

[0059] 1. For actual path extraction, the template matching algorithm and the image centroid method are used to calculate the spot coordinates, and the spot positions of several frames of images are plotted into a line to obtain the spot path;

[0060] 2. For preset path parsing, the G-code parsing module extracts the target coordinates corresponding to the actual path time, and the preset spot path is obtained by the cubic spline interpolation method.

[0061] As a further improvement of the present invention, in step 300, when the intelligent decision-making system calculates the decision: the deviation calculation module calculates the deviation between the actual data and the preset data, including the laser power deviation , the temperature deviation and the path deviation ;

[0062] The deviation calculation module includes the following sub-steps:

[0063] 1. Laser power deviation Calculated according to the following formula:

[0064]

[0065] Where and represent the actual laser power and the preset laser power respectively;

[0066] 2. The temperature deviation needs to calculate the deviation of the two temperatures of the powder bed and the laser spot respectively , calculated according to the following formula:

[0067]

[0068] Where and represent the actual temperature and the preset temperature respectively.

[0069] As a further improvement of the present invention, in step 300, when the intelligent decision-making system calculates the decision: the machine learning model compares and analyzes and mines the real-time data and the historical database through the Bayesian optimization algorithm, establishes a knowledge model and a rule base, and summarizes the optimal process parameter combination and defect handling strategy by analyzing the printing quality and defect conditions under different process parameters; according to the current printing conditions and monitoring data, automatically adjusts the control strategy and parameters to ensure the stability of the printing quality; the adjustment strategy integration module generates adjustment strategy data according to the defect monitoring results and the control parameters of the machine learning module, and classifies the adjustment data into two categories: "can continue printing" and "cannot continue printing".

[0070] As a further improvement of the present invention, in step 400, the intelligent control system controls the printing, specifically as follows:

[0071] The strategy data receiving module receives the adjustment strategy information, analyzes the data results, and performs different operations according to different data marks;

[0072] When the adjustment data is marked as "can continue printing", the start parameter adjustment module is activated. Its execution content includes recording the response level, adjusting various printing parameters of the printing device, and giving a path compensation value to adjust the printing path. Among them, the adjusted printing parameters include laser power, powder bed temperature, and sintering temperature. The operation process of this module and the parameter adjustment records will be stored in the historical database, so as to continuously update and optimize the historical database.

[0073] When the adjustment data is marked as "cannot continue printing", the termination printing module is activated. Its execution content is to record the fatal defects, stop printing, and issue an alarm. If this module is executed, it indicates that an irreparable error has occurred during the printing process, and at this time, manual intervention is required and readjustment is needed.

[0074] The beneficial effects of the present invention are:

[0075] The method of the present invention is applicable to the high-precision preparation of laser-sintered powder material specimens in rock mechanics experiments. Through multi-modal information fusion and closed-loop control strategies, it can accurately reproduce complex structural features such as micro-cracks and pores inside natural rocks, providing key technical support for the research of rock engineering materials.

[0076] (1) Multi-physical field coupling real-time monitoring technology: A distributed sensor array is used to achieve cross-physical field data synchronous acquisition, and an adaptive Kalman filter algorithm is combined to construct a multi-dimensional signal fusion processing module, breaking through the monitoring limitation of a single physical field in traditional 3D printing monitoring technology, and forming the ability of multi-modal data collaborative analysis.

[0077] (2) Visual real-time monitoring technology: A real-time defect detection model optimized based on the CNN-YOLOV8n architecture is built, deeply integrating the attention mechanism and the lightweight network architecture, and realizing dynamic visual monitoring during the continuous printing process through an industrial-grade CMOS camera, achieving a millisecond-level dynamic response while ensuring the detection accuracy, and constructing a full-process visualization monitoring system.

[0078] (3) Laser spot path tracking technology: An innovative dynamic comparison and monitoring method for the laser path and the preset path is proposed, and a sub-micron-level high-precision coordinate analysis and dynamic comparison module is developed, upgrading the traditional ex-post deviation analysis mode to a process-level root cause monitoring system, and realizing real-time quantitative analysis and traceability of the printing path deviation.

[0079] (4)Print path deviation recognition technology: Innovatively combines the dynamic time warping algorithm (DTW) with the Gaussian mixture model (GMM). The DTW algorithm converts the print path into a time series for processing, calculates the difference between the actual and preset print path time series, quantifies the path deviation, and effectively addresses print path fluctuations caused by factors such as equipment vibration and uneven spreading of material powder. As a probability model, GMM models the historical normal path deviation data to obtain the probability distribution of deviation values, and updates the model during the printing process to dynamically adjust the path compensation threshold. The combination of the two can accurately capture subtle path deviations and adaptively adjust the abnormal judgment criteria based on real-time and historical data, improving the accuracy and reliability of path monitoring.

[0080] (5)Print path compensation technology: Establishes a printing power model and an objective optimization function based on the principle of the model predictive control algorithm (MPC) to achieve print path prediction and optimization, integrating path monitoring and compensation. The printing power model predicts the printing state and print coordinates at the next moment within a very short time based on the current printing state and print path, obtains the predicted path through cubic spline interpolation, and then optimizes the objective function to adjust the predicted path through the objective optimization function. By comparing the final compensation amount with the compensation threshold, the path compensation method is determined to ensure the accuracy of the print path, improve printing efficiency and product quality.

[0081] (6)Intelligent three-level response module: Innovatively constructs a three-level intelligent response and control module through in-depth mining and analysis of multi-physical field coupling monitoring data. According to different error types that occur during the printing process, a three-level processing mechanism is accurately implemented, completely abandoning the traditional "one-size-fits-all" extensive control scheme, and greatly improving the stability and reliability of the printing process. Description of the Drawings

[0082] Figure 1 It is a schematic flow diagram of the method for real-time monitoring and regulation of the 3D printing process of the laser sintering powder material of the present invention. Detailed Embodiments

[0083] The present invention will be further described below with reference to the drawings.

[0084] As Figure 1 shown, the method for real-time monitoring and intelligent regulation of the 3D printing process of the laser sintering powder material of the present invention includes the following steps:

[0085] Step 100: The process of data collection by the multi-dimensional perception system.

[0086] Obtain real-time data during the printing process, including data such as laser power, powder bed temperature, sintering zone temperature, laser spot temperature, sintering zone image, and laser spot path information;

[0087] Step 200: Data processing steps of the data processing system.

[0088] Process the real-time collected data, including time alignment, noise filtering, abnormal data cleaning, and image conversion, etc.

[0089] Step 300: Calculation and decision-making process of the intelligent decision-making system.

[0090] Compare the real-time collected data with the original preset standard data, identify defects in the image data, compare the actual printing path with the preset printing path, and take corresponding decision-making measures based on the comparison results.

[0091] Step 400: Control the printing process by the intelligent control system.

[0092] Adjust the printing process and parameters according to the results of the data feedback, and start the corresponding response mechanism to ensure that the printing process proceeds accurately.

[0093] The result of the data feedback in this application is generated in the intelligent decision-making system of step 300. It is the optimal adjustment strategy obtained by the adjustment strategy generation module in step 300 through comprehensive analysis of these data based on the laser power deviation, temperature deviation, path deviation, path threshold compensation value, and path optimal compensation calculation value. This generated optimal adjustment strategy serves as the control instruction for the intelligent control of the printing process in step 400. That is, the monitoring system first judges the problems occurring in the printing process through the previous data analysis, generates adjustment solutions according to different printing problems, and feeds them back to the printing process, and the printer adjusts according to the adjustment solutions.

[0094] The closed-loop logic of the data feedback of the present invention is: the multi-dimensional perception system collects data → the data processing system cleans / aligns → the intelligent decision-making system analyzes and compares → generates feedback results (deviation values, defect marks, adjustment strategies) → the intelligent control system executes adjustments.

[0095] Based on the real-time parameters of the device, the real-time printing image, and the printing path, calculate the task execution status of the printing device respectively, and then comprehensively determine whether the device has a printing error. Compared with the traditional method, this monitoring process is more intelligent and efficient, can accurately monitor the printing quality of the laser ceramic 3D printing device, effectively reduce and even prevent the occurrence of printing errors, and significantly improve the printing efficiency.

[0096] Regarding step 100, the multi-dimensional perception system is used to achieve real-time data synchronization and acquisition, which is realized through the following modules: Laser power data acquisition module: A beam splitter assembly is set in the laser optical path. This assembly directs 95% of the main beam to the printing area, and the other 5% of the secondary beam is converted into an electrical signal through a photoelectric conversion device. A high-speed analog-to-digital converter is used to sample the electrical signal, and a digital signal processing algorithm is used to eliminate noise interference, and finally the real-time value of the laser power is successfully obtained.

[0097] Temperature data acquisition module: An infrared temperature sensor array is vertically installed on the top of the printing chamber. With a specific lens, it can cover a rectangular area of 100mm×100mm, realizing a point-plane combination monitoring method, so as to collect temperature field data including powder bed temperature and laser spot temperature in this area.

[0098] Vision acquisition module: A CMOS industrial camera is installed on the top of the printing chamber through a rotating assembly. This rotating assembly is connected to the galvanometer module of the printing device and rotates synchronously with the polarization mirror module to ensure that the shooting angle is 45° with the laser incident direction. The camera is connected to a narrowband pass filter through a data interface for image acquisition. The acquired image area is 100mm×100mm, and the image content covers the laser sintering area and the spot position. 100 frames of images are acquired each time.

[0099] Data transmission module: Considering the differences in the data acquisition frequencies among the sensors, in order to ensure a high degree of consistency in the acquisition time during subsequent data processing, a 32-bit timestamp (unit: µs) is added to the data frame header, and the laser power data, temperature field data, and image data are packed and sent to the data processing system.

[0100] The multi-dimensional perception system integrates advanced technologies such as laser power monitoring, temperature field reconstruction, sintering image acquisition, and laser spot position tracking. Through deep integration and efficient collaboration, it accurately captures key information during the laser processing process. Each sub-technology plays a unique advantage, real-time collects multi-dimensional data, and after strict calibration and error correction, it provides accurate real-time data support for subsequent monitoring, analysis, and decision-making processes, ensuring the efficiency, stability, and accuracy of the entire technological process.

[0101] Regarding step 200, the working method of the data processing system is used for efficient data processing and subsequent processing, which is realized through the following modules:

[0102] Time synchronization module: The data and images collected by the sensors and cameras are transmitted to the time synchronization module through the data interface. The time synchronization module processes the data with the timestamps carried by the data and images as tags, so as to ensure a high degree of consistency in the acquisition time of the data.

[0103] Data Processing Module: The data processing module performs preliminary data processing tasks. Data processing includes operations such as outlier cleaning, data noise filtering, and data format conversion.

[0104] Image Processing Module: Image processing includes steps such as "histogram equalization to enhance contrast - Gaussian filtering for noise reduction - Canny edge detection to extract contours - spot recognition algorithm based on Hu moments" to extract important features of the image and the position of the light spot.

[0105] Data Saving and Transmission Module: A six-dimensional data cube (X, Y, Z, P, T, t, with time alignment error ≤ 1 μs) is established, and a multi-physical-field data fusion analysis framework is constructed, where X, Y, and Z represent spatial positions, P represents laser power, T represents temperature, and t represents time. The data is saved and transmitted to the intelligent decision-making algorithm module through the main control board.

[0106] The data processing system builds a multi-physical-field data fusion analysis framework by constructing a spatio-temporal synchronous database cube, providing raw data support for subsequent defect recognition based on the CNN-YOLOV8n hybrid neural network, parameter adjustment based on Bayesian optimization, and path compensation for model predictive control, forming the core basis of the "perception - decision - execution" full-chain closed-loop control.

[0107] Regarding step 300, the working process of the intelligent decision-making system for efficient decision-making is implemented through the following modules:

[0108] Data Parsing Module: Receives the data transmitted by the data processing system and parses the data of the original print file (the file format is the print build slice G-code file). The file data contains printing parameters such as laser power, powder bed temperature, sintering temperature, and printing path.

[0109] Data Synchronization Module: Using the print timestamp as the standard and adopting the PTP + Kalman filtering algorithm, aligns the collected data with the preset data of the original print file (error ≤ 1 μs). Since there is no print image data in the preset data of the original print file, the image data is only aligned with the data time.

[0110] Feature Extraction Module: Extracts four data items of laser power, temperature, image, and light spot path at each time point.

[0111] The feature extraction module includes the following sub-steps:

[0112] 1. For actual path extraction, the template matching algorithm and the image centroid method are used to calculate the light spot coordinates, and the light spot positions of 100 frames of images are plotted as a line to obtain the light spot path;

[0113] 2. The preset path is parsed to extract the target coordinates corresponding to the actual path time through the G-code parsing module, and the preset light spot path is obtained by the cubic spline interpolation (continuous curvature) method.

[0114] Deviation calculation module: Calculate the deviation between the actual data and the preset data, including the laser power deviation ΔP, the temperature deviation ΔT, and the path deviation ΔD.

[0115] The deviation calculation module includes the following sub-steps:

[0116] 1. The laser power deviation is calculated according to the following formula:

[0117]

[0118] Where and represent the actual laser power and the preset laser power respectively. Indicates that the actual laser power is greater than the preset laser power, Indicates that the actual laser power is less than the preset laser power.

[0119] 2. The temperature deviation needs to calculate the deviation of the two temperatures of the powder bed and the laser light spot respectively , and is calculated according to the following formula:

[0120]

[0121] Where and represent the actual temperature (including the powder bed temperature and the laser light spot temperature) and the preset temperature (including the powder bed temperature and the laser light spot temperature) respectively. Indicates that the actual laser power is greater than the preset laser power, Indicates that the actual laser power is less than the preset laser power.

[0122] Defect monitoring module: Includes a threshold judgment module, a defect identification module, and an adjustment strategy generation module. The threshold judgment module compares the deviation value obtained by the deviation calculation module with the preset threshold. When the deviation value exceeds the threshold, an abnormal signal is sent, and when the deviation value is less than the threshold, a normal signal is output. The defect identification module identifies and marks the defects in the image and sends out corresponding marking signals. The signals sent by the threshold judgment module and the defect identification module are sent to the adjustment strategy generation module, and this module generates different countermeasures according to the signals.

[0123] The defect monitoring module includes the following sub-steps:

[0124] The method for identifying image defects is as follows:

[0125] The defect recognition and monitoring module constructs a high-precision defect recognition system based on the CNN-YOLOv8n hybrid neural network. By innovatively integrating the CSPDarknet53 backbone network and the PANet path aggregation structure, this architecture significantly improves the feature extraction efficiency and multi-scale defect detection ability while maintaining lightweight characteristics. The system adopts a transfer learning strategy and is intensively trained based on more than 100,000 high-precision annotated samples. The samples cover 8 types of typical printing defects such as pores, cracks, and sintered nodules, forming a perfect defect feature recognition model.

[0126] During the actual printing monitoring process, this module uses the input image and adjusts the dynamic threshold of adaptive NMS (Non-Maximum Suppression) through the following algorithm:

[0127]

[0128] The role of this formula is to dynamically adjust the threshold of non-maximum suppression according to the different layer heights in the neural network ( ). Among them, 0.5 is a basic threshold, which is the initial threshold when the layer height factor of the neural network is not considered. And 0.01 is a proportional coefficient used to control the influence degree of the neural network layer height on the threshold. As the neural network layer height ( ) increases, the threshold will gradually increase by 0.01 times the layer height. The purpose of this design is to enable the algorithm to flexibly adjust the threshold according to the feature conditions of different layers of the neural network. Usually, in deeper layers, the semantic information of the feature map is richer, and more target boxes of different scales and confidences may be detected. By dynamically increasing the threshold, the truly effective target boxes can be more strictly screened out, and those redundant and low-confidence boxes can be suppressed, thereby improving the accuracy and efficiency of defect detection. ) increases, the threshold will gradually increase by 0.01 times the layer height. The purpose of this design is to enable the algorithm to flexibly adjust the threshold according to the feature conditions of different layers of the neural network. Usually, in deeper layers, the semantic information of the feature map is richer, and more target boxes of different scales and confidences may be detected. By dynamically increasing the threshold, the truly effective target boxes can be more strictly screened out, and those redundant and low-confidence boxes can be suppressed, thereby improving the accuracy and efficiency of defect detection.

[0129] Through the adaptive NMS dynamic threshold adjustment algorithm, the image defect monitoring module can perform millisecond-level defect detection on real-time images collected at high speed. After detecting a defect, the system immediately generates structured data containing information such as defect type, size, and spatial coordinates, and sends a feedback instruction to the control unit through a high-speed communication protocol to trigger subsequent adjustment strategies.

[0130] 1. The path deviation monitoring method is as follows:

[0131] Dynamic Time Warping (DTW) algorithm: The printing path is converted into a time series for processing. For the actual printing path, the time stamps and position coordinates of the collected images are converted into time series information as the actual printing time series. For the preset printing path, according to parameters such as the designed printing texture and speed, the printing position at the theoretical moment is generated, and this theoretical moment and the theoretical printing position are used as the preset printing time series. Among them, the time series of the actual printing path is:

[0132] ,

[0133] wherein represents the position vector of the actual path at the i-th time point, that is:

[0134] ;

[0135] The time series of the preset printing path is:

[0136] ,

[0137] Similarly represents the position vector of the preset path at the j-th time point, that is:

[0138] ,

[0139] wherein , that is, the actual printing coordinates and the preset printing coordinates at the same printing moment.

[0140] Taking the actual path time series as rows and the preset path time series as columns, a two-dimensional distance matrix D is constructed. The matrix D is an m×n matrix, and the deviation between the actual coordinates and the preset coordinates is calculated by the Euclidean geometric distance algorithm:

[0141]

[0142] Each element in the matrix represents the distance between the position at a certain moment in the actual path and the position at a certain moment in the preset path, and the sum of the distances in this very short period of time is recorded as the path deviation value.

[0143] 2. The calculation method of the path compensation threshold is as follows:

[0144] (1) Gaussian mixture model (GMM): In path deviation monitoring, by performing GMM modeling on historical normal path deviation data, the probability distribution of different deviation values can be obtained. As the printing process progresses, the GMM model is continuously updated according to new data, and the adaptive coefficient is obtained through reinforcement learning based on the probability distribution, and then the path compensation threshold is dynamically adjusted.

[0145] (2) The calculation method of the compensation threshold is:

[0146]

[0147] Among them, α, β, and δ are adaptive coefficients obtained by the reinforcement learning algorithm through sorting and analyzing historical printing data. The reinforcement learning algorithm can continuously adjust the values of the coefficients according to the feedback information of the module, making the compensation threshold more adaptable to different printing conditions and improving the accuracy and effectiveness of path deviation compensation.

[0148] 3. The path prediction and compensation mechanism is as follows:

[0149] (1) Establishment of the printing path dynamic model and prediction path: According to the principle of the model predictive control (MPC) algorithm, establish the printing path dynamic model:

[0150]

[0151] Where represents the predicted printing state vector at the next moment, represents the current printing state vector, including the relevant states of the galvanometer (current position, deflection angle, angular velocity), the current printing coordinates, and path information; A and B are coefficient matrices that describe the influence of state transition and control input on the state, and the coefficients of A and B can both be calculated from the historical printing state change data; is the control input signal function at the current k - 1 moment, is the process noise, which follows a normal distribution with a mean of 0 and a covariance matrix of Q. Through this printing path dynamic model, the printing state and the printing position point coordinates at the next moment within a very short time (50 ms) can be predicted based on the current printing state. After obtaining the printing coordinates at the next moment, the current printing coordinates and the predicted coordinate values are connected by the cubic spline interpolation algorithm to obtain the predicted path.

[0152] (2) Optimize the predicted path: Optimize the objective function through the model predictive control (MPC) algorithm:

[0153]

[0154] to dynamically adjust the predicted printing path. In this objective function, represents the predicted state vector, represents the desired state vector, which is the preset printing state, and Q and R are the weight matrices of the state error and the control input respectively. Within a certain prediction horizon N, by optimizing the control input , make the actual state as close as possible to the desired state (measured by , and the Q matrix can adjust the weights of different state variables), and at the same time limit the size of the control input (by Implementation. The R matrix is used to adjust the weights of different control input variables to avoid unnecessary impacts on the device. When the finally calculated optimized path compensation value is 0, no path compensation operation is performed. When it is not 0, compensation is required. The final compensation amount is compared with the compensation threshold. When the final compensation amount is less than the compensation threshold, the printing path is adjusted directly in one step. When the compensation amount exceeds the threshold, the printing path is adjusted step by step to avoid damaging the printing device due to excessive compensation amount.

[0155] 4. The adjustment strategy generation module generates different adjustment strategies according to different signals. The specific method is as follows:

[0156] (1) Construct a laser power-temperature two-parameter collaborative regulation model. When the monitored value is within the safe threshold range, the module triggers three types of response strategies based on the threshold comparison results: equal value state: maintain the current parameters ( ), positive deviation: execute the parameter attenuation instruction (power / temperature reduction), and negative deviation: start the parameter enhancement instruction (power / temperature increase).

[0157] (2) Molding path compensation module: The module receives the compensation amount data packet from the online monitoring module and uses an adaptive interpolation algorithm to generate a smooth transition path.

[0158] Machine learning module: The machine learning model compares and analyzes the real-time data with the historical database through the Bayesian optimization algorithm, establishes a knowledge model and a rule base, and summarizes the optimal combination of process parameters and defect handling strategies by analyzing the printing quality and defect conditions under different process parameters. According to the current printing conditions and monitoring data, the control strategy and parameters are automatically adjusted to ensure the stability of printing quality.

[0159] Adjustment strategy integration module: Generate adjustment strategy data according to the defect monitoring results and the control parameters of the machine learning module, and mark the adjustment data into two categories: "can continue printing" and "cannot continue printing".

[0160] The intelligent decision-making system, based on the synchronous information collected by multiple sensors in real time, analyzes through machine learning whether there are errors in the current printing parameters, sintering screen, and printing path and sets an error threshold. When it is detected that the printing parameter and path errors exceed the threshold, the corresponding adjustment instructions are sent to the intelligent control system to perform the corresponding adjustment of printing parameters and printing steps.

[0161] Regarding step 400, the specific working process of the intelligent control system, which is used to receive the regulation instructions and execute the corresponding operation strategies, is implemented through the following modules:

[0162] Strategy data receiving module: Receive the adjustment strategy information, analyze the data results, and perform different operations according to different data marks.

[0163] Parameter adjustment module: This module is activated when the adjustment data is marked as "printable continuously". Its execution content includes recording the response level, adjusting various printing parameters of the printing device, and giving a path compensation value to adjust the printing path. Among them, the adjusted printing parameters include laser power, powder bed temperature, and sintering temperature. The operation process of this module and the parameter adjustment records will be stored in the historical database, so as to continuously update and optimize the historical database.

[0164] Termination printing module: This module is activated when the adjustment data is marked as "not printable continuously". Its execution content is to record fatal defects, stop printing, and issue an alarm. If this module is executed, it indicates that there are irreparable errors during the printing process, such as cracks, sintered nodules, or the printing specimen is misaligned due to excessive deviation of the printing path. At this time, manual intervention and re-adjustment are required.

[0165] As a key system for implementing the new strategy of adjusting printing parameters and printing paths, the intelligent control system achieves a complete closed-loop of "data acquisition - intelligent decision - execution feedback" with the help of the hierarchical response module.

[0166] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product.

[0167] Finally, it should be noted that: What is disclosed in the real-time monitoring and control method for the 3D printing process of laser-sintered powdered materials disclosed in the embodiments of the present invention is only the preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention, rather than limiting them; Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features.

Claims

1. A real-time monitoring and control method for the 3D printing process of laser sintered powdered materials, characterized in that, It includes the following steps: Step 100, the multi-dimensional perception system collects data: Obtain real-time data of the printing process, including laser power, powder bed temperature, sintering area temperature, laser spot temperature, sintering area image, and laser spot path information data; Add a timestamp to the data frame header, and pack and send the laser power data, temperature field data, and image data to the data processing system; Step 200, the data processing system processes data: Process the real-time collected data, including time alignment, noise filtering, abnormal data cleaning, and image conversion, extract important features of the image and the spot position, and establish a six-dimensional data cube: X, Y, Z, P, T, t, with a time alignment error ≤ 1 μs, construct a multi-physical field data fusion analysis framework, where X, Y, Z represent spatial positions, P represents laser power, T represents temperature, and t represents time, save the data and transmit the data to the intelligent decision-making system through the main control board; Step 300, the intelligent decision-making system analyzes and compares to generate the result of data feedback: Compare the real-time collected data with the original preset standard data, identify defects in the image data, compare the actual printing path with the preset printing path, and take corresponding decision-making measures based on the comparison results; The defect identification method is as follows: Image defect identification; Path deviation monitoring; Path compensation threshold calculation; Path prediction and compensation value calculation; The adjustment strategy generation module generates different adjustment strategies according to different signals; The result of generating data feedback includes deviation value, defect mark, and adjustment strategy; Step 400, the intelligent control system controls the printing, adjusts the printing process and parameters according to the result of data feedback, and starts the corresponding response mechanism to ensure the accurate progress of the printing process; The path compensation threshold, the calculation method is as follows: (1) Gaussian mixture model GMM: In path deviation monitoring, perform GMM modeling on historical normal path deviation data to obtain the probability distribution of different deviation values. As the printing process progresses, continuously update the GMM model according to new data, obtain the adaptive coefficient through reinforcement learning based on the probability distribution, and then dynamically adjust the path compensation threshold; (2) The calculation method of the compensation threshold is: threshold(t) = α · layer_height + β · laser_power + δ · scan_speed Where α, β, and δ are adaptive coefficients obtained by organizing and analyzing historical printing data through the reinforcement learning algorithm.

2. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 1, wherein: The path prediction and compensation mechanism is as follows: (1) Establish the printing path dynamic model and the predicted path: According to the principle of the model predictive control algorithm MPC, establish the printing path dynamic model: x k = Ax k-1 + Bu k-1 + w k-1 where x k represents the predicted printing state vector at the next moment, and x k-1 represents the current printing state vector, including the relevant states of the galvanometer scanner, the current printing coordinates, and path information; A and B are coefficient matrices that describe the influence of state transition and control input on the state, and the coefficients of A and B can both be calculated from the historical printing state change data; u k-1 is the control input signal function at the current k-1 moment, w k-1 is the process noise, which follows a normal distribution with a mean of 0 and a covariance matrix of Q; through this printing path dynamic model, the printing state and the coordinates of the printing position point at the next moment within a very short time can be predicted based on the current printing state; after obtaining the printing coordinates at the next moment, the current printing coordinates and the predicted coordinate values are connected by a cubic line interpolation algorithm to obtain the predicted path; (2) Optimize the predicted path: Optimize the objective function through the model predictive control algorithm MPC: In this objective function, x k represents the predicted state vector, and represents the desired state vector, which is the preset printing state. Q and R are the weight matrices of the state error and the control input respectively. Within a certain prediction time domain N, by optimizing the control input μ k , the actual state x k is made to be as close as possible to the desired state . This is measured by . The Q matrix can adjust the weights of different state variables and at the same time limit the magnitude of the control input, which is achieved by . The R matrix is used to adjust the weights of different control input variables to avoid unnecessary impacts on the device. By optimizing this objective function, the optimal control input sequence can be obtained to achieve the collaborative optimization of path compensation correction and optimization. When the finally calculated optimized path compensation value is 0, no path compensation operation is performed. When it is not 0, compensation is required. The final compensation amount is compared with the compensation threshold. When the final compensation amount is less than the compensation threshold, the printing path is adjusted directly in one step. When the compensation amount exceeds the threshold, the printing path is adjusted step by step to avoid damaging the printing device due to excessive compensation amount.

3. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 2, characterized in that: In step 300, the image defect recognition method is as follows: The defect recognition monitoring module constructs a high-precision defect recognition system based on the CNN-YOLOv8n hybrid neural network; this system innovatively integrates the CSPDarknet53 backbone network and the PANet path aggregation structure; the system adopts a transfer learning strategy and conducts intensive training based on more than 100,000 high-precision labeled samples, and the samples cover typical printing defects, forming a perfect defect feature recognition model; During the actual printing monitoring process, this module uses the input image and adopts the adaptive NMS dynamic threshold adjustment algorithm: filtering_threshold = 0.5 + 0.01 × layer_number The function of this formula is to dynamically adjust the non-maximum suppression threshold according to the height of different layers in the neural network; among them, 0.5 is a basic threshold, which is the initial threshold when not considering the factor of the neural network layer height, and 0.01 is a proportional coefficient used to control the influence degree of the neural network layer height on the threshold; as the neural network layer height increases, the threshold will gradually increase by 0.01 times the layer height; Through the adaptive NMS dynamic threshold adjustment algorithm, the image defect monitoring module can perform millisecond-level defect detection on the real-time images collected at high speed; after detecting a defect, the system immediately generates structured data and sends a feedback instruction to the adjustment strategy module to trigger subsequent adjustment strategies.

4. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 2, characterized in that: In step 300, the path deviation monitoring method is as follows: Dynamic time warping algorithm: The printing path is converted into a time series for processing. For the actual printing path, the timestamp and position coordinates of the collected image are converted into time series information as the actual printing time series. For the preset printing path, according to parameters such as the designed printing texture and speed, the printing position at the theoretical moment is generated, and this theoretical moment and the theoretical printing position are used as the preset printing time series. Among them, the time series of the actual printing path is: P = {p1, p2, … p m}, where p i represents the position vector of the actual path at the i-th time point, that is: p i =(x i , y i , z i ); The time series of the preset printing path is: Q = {q1, q2, … q n}, Similarly for q j represents the position vector of the preset path at the j-th time point, i.e.: q j =(x j , y j , z j ), where i = j, that is, the actual printing coordinates and the preset printing coordinates at the same printing moment; Taking the actual path time series as rows and the preset path time series as columns, a two-dimensional distance matrix D is constructed. Matrix D is an m×n matrix, and the deviation between the actual coordinates and the preset coordinates is calculated: Each element D(i,j) in the matrix represents the distance between the position at a certain moment in the actual path and the position at a certain moment in the preset path, and the sum of the distances in this very short period of time is recorded as the path deviation value.

5. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 2, characterized in that: In step 300, the adjustment strategy generation module generates different adjustment strategies according to different signals. The specific method is as follows: (1) Construct a laser power-temperature dual-parameter collaborative regulation model. When the monitored value is within the safe threshold range, the module triggers three types of response strategies based on the threshold comparison result: Equivalent state: Maintain the current parameters; Positive deviation: Execute the parameter attenuation instruction, and the power / temperature decreases; Negative deviation: Start the parameter enhancement instruction, and the power / temperature increases; (2) Forming path compensation module: The module receives the compensation amount data packet from the online monitoring module and uses the adaptive interpolation algorithm to generate a smooth transition path.

6. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 2, characterized in that: In step 300, when the intelligent decision-making system calculates the decision: The data parsing module receives the data transmitted by the data processing system and parses the data of the original print file; The data synchronization module takes the print timestamp as the standard and uses the PTP + Kalman filtering algorithm to align the collected data with the preset data of the original print file. The image data is only aligned with the data time; The feature extraction module extracts four data of laser power, temperature, image, and spot path at each time point; The feature extraction module includes the following sub-steps:

1. For actual path extraction, the template matching algorithm and the image centroid method are used to calculate the spot coordinates, and the spot positions of several frames of images are plotted as lines to obtain the spot path; 2. For preset path parsing, the G-code parsing module extracts the target coordinates corresponding to the actual path time, and the preset spot path is obtained through cubic spline interpolation.

7. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 6, characterized in that: In step 300, when the intelligent decision-making system calculates the decision: The deviation calculation module calculates the deviation between the actual data and the preset data, including the laser power deviation ΔP, the temperature deviation ΔT, and the path deviation ΔD; The deviation calculation module includes the following sub-steps:

1. The laser power deviation ΔP is calculated according to the following formula: where P real and P preset represent the actual laser power and the preset laser power, respectively; 2. The temperature deviation needs to calculate the deviation ΔT of the two temperatures of the powder bed and the laser spot respectively, and is calculated according to the following formula: where T real and T preset represent the actual temperature and the preset temperature respectively.

8. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 7, characterized in that: In step 300, when the intelligent decision-making system calculates the decision: The machine learning model conducts comparative analysis and mining on the real-time data and the historical database through the Bayesian optimization algorithm, establishes a knowledge model and a rule base, and summarizes the optimal process parameter combination and defect handling strategy by analyzing the printing quality and defect conditions under different process parameters; Automatically adjust the control strategy and parameters according to the current printing conditions and monitoring data to ensure the stability of the printing quality; The adjustment strategy integration module generates adjustment strategy data according to the defect monitoring results and the control parameters of the machine learning module, and classifies the adjustment data into two categories: "can continue printing" and "cannot continue printing".

9. The real-time monitoring and control method for the 3D printing process of laser sintering powdered materials according to claim 8, characterized in that: In step 400, the intelligent control system controls the printing, specifically as follows: The strategy data receiving module receives the adjustment strategy information, analyzes the data results, and performs different operations according to different data marks; When the adjustment data mark is "can continue printing", the parameter adjustment module is started. Its execution content includes recording the response level, adjusting the various printing parameters of the printing device, and giving a path compensation value to adjust the printing path; among them, the adjusted printing parameters include laser power, powder bed temperature, and sintering temperature; the operation process and parameter adjustment records of this module will be stored in the historical database to continuously update and optimize the historical database; When the adjustment data mark is "cannot continue printing", the print termination module is started. Its execution content is to record the fatal defect, stop printing, and issue an alarm; if this module is executed, it indicates that an irreparable error has occurred during the printing process, and at this time, manual intervention and re-adjustment are required.

Images