Method and system for monitoring, regulating and controlling temperature of powder bed in real time

Abstract

The invention relates to the technical field of powder bed temperature control, and discloses a powder bed temperature real-time monitoring, regulating and controlling method and system, and the method comprises the steps: dividing a powder bed into n1 actual temperature control regions, setting a target temperature value and an allowable deviation for each actual temperature control region, and forming an initial temperature distribution matrix; obtaining a first temperature time sequence, a thermal image time sequence and a production environment time sequence, and generating a first comprehensive feature sequence; an optimal powder bed temperature sequence prediction model and an optimal temperature distribution model of the current printing task are extracted, a temperature prediction smooth curve and an allowable temperature change range of each actual temperature control area are obtained, and graded early warning is carried out on temperature abnormity; obtaining a second temperature distribution matrix according to the temperature abnormity early warning level, performing temperature regulation and control on each actual temperature control area, and evaluating the regulation and control effect; real-time monitoring and intelligent regulation and control of the temperature of the powder bed are achieved, and the stability of the printing quality is effectively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of powder bed temperature control technology, and more specifically, to a method and system for real-time monitoring and control of powder bed temperature. Background Technology

[0002] With the rapid development of 3D printing technology, resistance heating powder-spreading 3D printing has become a widely used additive manufacturing method. In resistance heating powder-spreading 3D printing, precise control of the powder bed temperature is crucial for ensuring print quality. While some methods for powder bed temperature control exist in existing technologies, they still have some shortcomings.

[0003] Patent application CN114850498A discloses a method for controlling uniform preheating of a powder bed and an additive manufacturing apparatus. This method determines the center point and farthest point of the area to be preheated in the powder bed, determines the minimum and maximum preheating power based on these points, and determines the preheating power at each point according to the ratio of the distance from each point in the scanning path to the center point to the distance from the farthest point to the center point, thereby achieving uniform preheating of the powder bed. However, this method mainly focuses on the preheating stage and does not address real-time temperature monitoring and control during the printing process. Furthermore, this method does not consider the influence of production environment parameters on temperature, making it difficult to cope with complex actual production situations.

[0004] Chinese patent application CN118305332B discloses a control method for preparing high-strength stainless steel using laser powder bed technology. This method collects time series data on scanning speed and real-time temperature using sensors, and employs data analysis and processing techniques based on deep learning neural networks to extract temporal implicit features and interactive responses of the scanning speed and real-time temperature, thereby adaptively controlling the scanning speed at the current time point. While this method can adaptively adjust the scanning speed according to the current process state, it does not involve fine-grained management and control of the powder bed temperature distribution. Furthermore, this method is primarily designed for laser powder bed technology and is difficult to directly apply to resistance heating powder bed 3D printing.

[0005] In summary, existing powder bed temperature control methods lack refined management and real-time monitoring of powder bed temperature distribution, fail to fully consider the influence of production environment parameters on temperature, and lack temperature anomaly detection and intelligent control methods for high-performance materials such as titanium alloys, copper alloys, and cemented carbide in resistance heating powder-laying 3D printing, making it difficult to meet the requirements of different alloy material thermophysical differences. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, this invention provides a method and system for real-time monitoring and control of powder bed temperature, which can finely manage the temperature distribution of the powder bed, comprehensively consider various influencing factors, and promptly detect and control temperature anomalies, thereby effectively ensuring the quality stability of resistance heating powder spreading 3D printing.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] Methods for real-time monitoring and control of powder bed temperature include:

[0009] The powder bed is divided into n1 actual temperature control zones. A target temperature value and allowable deviation are set for each actual temperature control zone to form an initial temperature distribution matrix. The initial temperature distribution matrix is ​​stored in the process database. The first temperature time series, thermal image time series and production environment time series are obtained. Based on the first temperature time series, thermal image time series and production environment time series, a first comprehensive feature sequence is generated. n1 is a positive integer.

[0010] Extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database. Obtain the temperature prediction smoothing curve for each actual temperature control area based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model. Obtain the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task based on the optimal temperature distribution model. Perform graded early warning for temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range.

[0011] Based on the temperature anomaly warning level, a second temperature distribution matrix is ​​obtained; the current temperature of each actual temperature control area is collected in real time to form a second temperature data matrix; the second temperature data matrix is ​​compared with the second temperature distribution matrix to generate a temperature deviation matrix; the temperature of each actual temperature control area is adjusted according to the temperature deviation matrix; after the temperature adjustment is completed, the adjustment effect is evaluated.

[0012] Furthermore, acquiring the first temperature time series, the thermal image time series, and the production environment time series includes:

[0013] The temperature time series is formed by collecting the temperature change sequence of each actual temperature control area through a temperature sensor array; the real-time thermal images of the entire powder bed surface at different time points are collected by an infrared thermal imager to form a thermal image time series; the production environment data change sequence during the 3D printing process is collected to form a production environment time series; the production environment time series includes the time series of n3 environmental parameters, including temperature, humidity, oxygen concentration and dust concentration.

[0014] Further, generating the first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series includes:

[0015] Feature extraction is performed on the first temperature time series to obtain the temperature feature series;

[0016] Feature extraction is performed on the time series of thermal images to obtain the feature sequence of thermal images;

[0017] Feature extraction is performed on the time series of the production environment to obtain the production environment feature sequence;

[0018] The temperature feature sequence, thermal image feature sequence, and production environment feature sequence are aligned along the time dimension to form the first comprehensive feature sequence.

[0019] The feature extraction of the first temperature time series includes:

[0020] The first temperature time series is divided into n² time windows, each containing a fixed number of temperature sampling points; n² is a positive integer.

[0021] For each time window, calculate its mean, variance, maximum, and minimum temperature values ​​to generate a temperature feature vector.

[0022] The temperature feature vectors of each time window are arranged in chronological order to form a temperature feature sequence.

[0023] Furthermore, the feature extraction from the thermal image time series includes:

[0024] Each thermal image is divided into sub-regions corresponding to n1 actual temperature control areas. The grayscale histogram, gradient histogram and texture features of each sub-region are extracted to form an image feature vector.

[0025] The image feature vectors of all sub-regions at the same time are concatenated into a complete thermal image feature vector; the thermal image feature vectors at all times are arranged in chronological order to obtain the thermal image feature sequence.

[0026] The feature extraction of the production environment time series includes a first level and a second level; the first level is to extract features from the time series of a single environmental parameter to obtain the feature sequence of the single environmental parameter; the second level is to combine the feature sequences of different environmental parameters.

[0027] Furthermore, the optimal powder bed temperature sequence prediction model for extracting the current printing task from the process database includes:

[0028] Extract key process parameters from the process configuration file of the current printing task to form a process parameter tuple; the key process parameters include powder bed size, material type, layer thickness and printing speed;

[0029] The process parameter tuples are compared item by item with the process parameter combinations corresponding to the pulverized bed temperature sequence prediction model in the process database. If all key process parameters are equal, they are considered to be a perfect match. The pulverized bed temperature sequence prediction model that is a perfect match is taken as the optimal pulverized bed temperature sequence prediction model.

[0030] If no perfectly matching pulverized bed temperature sequence prediction model is found, the similarity between the process parameter tuple and the process parameter combination corresponding to the pulverized bed temperature sequence prediction model in the process database is calculated, and the one with the highest similarity is selected as the optimal pulverized bed temperature sequence prediction model.

[0031] Furthermore, the allowable temperature variation range of each actual temperature control zone of the toner bed in the current printing task, obtained according to the optimal temperature distribution model, includes:

[0032] By combining the powder bed size and the actual temperature control area, the optimal temperature distribution model is adaptively modified to generate the target temperature distribution matrix; the upper and lower thresholds of the target temperature distribution matrix are calculated to obtain the allowable temperature variation range of each actual temperature control area.

[0033] Furthermore, the optimal temperature distribution model includes the division of temperature control zones and the target temperature value for each temperature control zone;

[0034] The adaptive modification of the optimal temperature distribution model to generate the target temperature distribution matrix includes:

[0035] Align the temperature control region of the optimal temperature distribution model with the actual temperature control region of the powder bed. If the particle size of the temperature control region of the optimal temperature distribution model is inconsistent with that of the actual temperature control region, then perform interpolation or merging processing on the optimal temperature distribution model to match the temperature control region of the optimal temperature distribution model with that of the actual temperature control region.

[0036] Geometric transformation of the optimal temperature distribution model is performed based on the powder bed size to obtain an optimal temperature distribution transformation model adapted to the powder bed size;

[0037] Fine-tune the target temperature value in the optimal temperature distribution transformation model to generate the adjusted optimal temperature distribution transformation model;

[0038] The adjusted optimal temperature distribution transformation model is discretized to generate a target temperature distribution matrix corresponding to the actual temperature control area.

[0039] Furthermore, the step of classifying and issuing early warnings for temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range includes:

[0040] The temperature prediction smoothing curve is compared with the allowable temperature change range to determine whether a temperature anomaly will occur within the prediction time. If a temperature anomaly occurs, the magnitude and duration of the temperature exceeding the limit in the temperature prediction smoothing curve are calculated. Based on the magnitude and duration of the exceeding the limit, a graded warning for the temperature anomaly is issued, generating a temperature anomaly warning level.

[0041] The temperature anomaly warning levels include: Level 1 warning, Level 2 warning, Level 3 warning, and Level 4 warning;

[0042] Level 1 warning: Exceeding limit amplitude AM < 5%; Duration DU < 1 min;

[0043] Level II Warning: 5% ≤ Exceedance Amplitude AM < 10%; 1 min ≤ Duration DU < 5 min;

[0044] Level 3 warning: 10% ≤ Exceedance amplitude AM < 20%; 5min ≤ Duration DU < 10min;

[0045] Level 4 warning: Exceeding the limit AM ≥ 10%; Duration DU ≥ 10 min.

[0046] Furthermore, the evaluation of the regulatory effect includes:

[0047] Real-time acquisition of the second temperature time series after temperature control, the second thermal image time series of the entire powder bed surface, and the second production environment time series during the 3D printing process; extraction of the second temperature feature sequence of the second temperature time series, the second thermal image feature sequence of the second thermal image time series, and the second production environment feature sequence of the second production environment time series; and generation of the second comprehensive feature sequence.

[0048] The second comprehensive feature sequence is input into the optimal powder bed temperature sequence prediction model to generate the second temperature prediction sequence after temperature regulation.

[0049] Compare the second temperature prediction sequence with the first temperature prediction sequence to assess the difference in temperature change trends before and after temperature regulation; calculate the temperature distribution deviation between the second temperature prediction sequence and the target temperature distribution matrix.

[0050] The comprehensive control effect evaluation score is obtained based on the differences in temperature change trends and temperature distribution deviations.

[0051] A real-time monitoring and control system for powder bed temperature, used to implement the above-mentioned real-time monitoring and control method for powder bed temperature, the system comprising:

[0052] Feature generation module: used to divide the powder bed into n1 actual temperature control zones, set target temperature values ​​and allowable deviations for each actual temperature control zone, form an initial temperature distribution matrix, and store the initial temperature distribution matrix into the process database; acquire the first temperature time series, thermal image time series, and production environment time series, and generate the first comprehensive feature sequence based on the first temperature time series, thermal image time series, and production environment time series; n1 is a positive integer;

[0053] The graded early warning module is used to extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database. Based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model, it obtains the temperature prediction smoothing curve for each actual temperature control area. Based on the optimal temperature distribution model, it obtains the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task. Based on the temperature prediction smoothing curve and the allowable temperature variation range, it provides graded early warnings for temperature anomalies.

[0054] Control module: Based on the temperature anomaly warning level, obtain the second temperature distribution matrix; collect the current temperature of each actual temperature control area in real time to form a second temperature data matrix; compare the second temperature data matrix with the second temperature distribution matrix to generate a temperature deviation matrix; adjust the temperature of each actual temperature control area according to the temperature deviation matrix; evaluate the control effect after the temperature control is completed.

[0055] An electronic device includes a memory, a central processing unit (CPU), and a computer program stored in the memory and executable on the CPU. When the CPU executes the computer program, it implements the aforementioned method for real-time monitoring and control of powder bed temperature.

[0056] A computer-readable storage medium storing a computer program, which, when executed, implements the above-described method for real-time monitoring and control of powder bed temperature.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] This invention divides the powder bed into multiple temperature control zones, setting target temperatures and allowable deviations for each zone to suit materials such as titanium alloys, copper alloys, and cemented carbide. This achieves refined management and control of the powder bed temperature during the manufacturing process of these materials, enhancing the targeted nature of temperature process control and avoiding localized overheating or underheating caused by uneven temperature distribution. It also adapts to the differences in melting point and thermal conductivity of different alloy materials. By comprehensively considering real-time temperature data collected by temperature sensors, thermal images of the powder bed surface collected by thermal imagers, and production environment parameters, a comprehensive feature sequence is generated to fully reflect the real-time temperature status of the powder bed, providing a reliable basis for subsequent temperature prediction and anomaly detection. By extracting the optimal powder bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task, and combining this with the comprehensive feature sequence to predict the temperature change trend of each temperature control zone, the allowable temperature change range is obtained. This allows for timely detection of temperature anomalies and tiered early warnings, providing a proactive guarantee for precise process control and reducing the risk of temperature runaway. Based on the temperature anomaly warning level, the target temperature distribution is dynamically adjusted to generate a second temperature distribution matrix. The temperature of each temperature-controlled zone is collected in real time and compared with the target value to generate a temperature deviation matrix. Closed-loop temperature control is then implemented to ensure the temperature remains within a reasonable range. After temperature control, the effect is objectively evaluated by comparing the differences in temperature change trends and temperature distribution deviations before and after control. A comprehensive control effect evaluation score quantifies the control performance, providing feedback for optimizing the temperature control strategy. The entire temperature monitoring and control process is fully automated and intelligent, making process control for materials such as titanium alloys, copper alloys, and cemented carbide more efficient and precise. This significantly improves temperature management efficiency, reduces the need for manual intervention, and ensures the continuous stability of the printing process for different alloy materials, providing a reliable guarantee for the molding of high-performance alloy parts. Simultaneously, the constructed closed-loop control system also has self-learning and iterative optimization capabilities, continuously accumulating experience in actual production, dynamically adjusting relevant prediction models and strategy parameters, and optimizing process control logic to continuously improve temperature control performance. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the principle of the real-time monitoring and control method for powder bed temperature in this invention.

[0061] Figure 2 This is a flowchart of the method for generating the first comprehensive feature sequence in the real-time monitoring and control method for powder bed temperature of the present invention;

[0062] Figure 3 This is a flowchart of the method for feature extraction of the first temperature time series in the real-time monitoring and control method of powder bed temperature of the present invention.

[0063] Figure 4 This is a flowchart of the method for feature extraction of thermal image time series in the real-time monitoring and control method of powder bed temperature of the present invention.

[0064] Figure 5 This is a flowchart of the method for retrieving the optimal powder bed temperature sequence prediction model in the real-time monitoring and control method for powder bed temperature of the present invention;

[0065] Figure 6 This is a flowchart of the method for adaptively modifying the optimal temperature distribution model and generating the target temperature distribution matrix in the real-time monitoring and control method of powder bed temperature of the present invention.

[0066] Figure 7 This is a flowchart of the method for generating the second temperature distribution matrix in the real-time monitoring and control method for powder bed temperature of the present invention.

[0067] Figure 8 This is a functional module diagram of the real-time monitoring and control system for powder bed temperature in this invention. Detailed Implementation

[0068] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Example 1

[0070] Please see Figure 1 As shown, this embodiment provides a method for real-time monitoring and control of powder bed temperature, including:

[0071] Step S1000: Divide the powder bed into n1 actual temperature control zones, set a target temperature value and allowable deviation for each actual temperature control zone to form an initial temperature distribution matrix, and store the initial temperature distribution matrix in the process database; obtain the first temperature time series, thermal image time series and production environment time series, and generate the first comprehensive feature sequence based on the first temperature time series, thermal image time series and production environment time series; n1 is a positive integer;

[0072] Further, step S1000 includes:

[0073] Step S1100: Divide the powder bed into n1 actual temperature control zones, set a target temperature value and allowable deviation for each actual temperature control zone, form an initial temperature distribution matrix, and store the initial temperature distribution matrix in the process database; n1 is a positive integer.

[0074] Specifically, the rectangular toner bed is divided into n1 rectangular regions, each of which is an independent temperature control unit. The specific value of n1 needs to be considered in conjunction with factors such as the toner bed size and printing accuracy requirements. The granularity of the region division should be fine enough to ensure that the temperature gradient between adjacent regions is not too large, but also moderate to avoid excessive control and calculation overhead. Taking a typical 300mm×300mm toner bed 3D printer as an example, dividing the toner bed into 3×3, or 9 control regions, is a reasonable choice.

[0075] For each actual temperature control zone, a target temperature value is set based on the material's optimal printing temperature. The target temperature is typically 5-10°C lower than the material's melting point. For example, acrylonitrile-butadiene-styrene copolymer (ABS) has a melting point of 180-200°C, so the target temperature could be set at around 175°C. Simultaneously, considering the lag in temperature regulation and the tolerance of material properties, a certain allowable deviation should be set on both sides of the target temperature value, forming a target temperature range. The deviation setting needs to balance print quality and energy consumption. A narrow range is beneficial for temperature stability and high print quality, but consumes more energy; a wide range saves power, but causes large temperature fluctuations and reduced print quality. For ABS, ±5°C is a relatively balanced choice.

[0076] The aforementioned region division scheme, target temperature, and allowable deviation information are integrated into an initial temperature distribution matrix and stored in the process database. This matrix defines the baseline operating state of the powder bed heating system and serves as the starting point for subsequent dynamic temperature control. Persistently storing this matrix provides a data foundation for subsequent process optimization and experience accumulation, and also meets the need for knowledge reuse. For identical or similar printing tasks, the existing matrix can be directly invoked, eliminating the hassle of repeated setup.

[0077] The process database is a knowledge base that encompasses a wealth of printing experience data under various combinations of printing materials and process parameters. It serves as a crucial data foundation for optimizing printing processes.

[0078] The process database stores two types of key models:

[0079] Powder bed temperature sequence prediction model: This model can predict the trend of powder bed temperature change over a period of time in the future based on the current environmental characteristics.

[0080] Optimal Temperature Distribution Model: The model provides the optimal temperature distribution scheme for the toner bed under a specific printing task.

[0081] The process database indexes models using key attributes such as material type and process parameters, making it easy to quickly retrieve the most relevant models based on the attributes of the current printing task.

[0082] The purpose of the process database is to reuse past experience and avoid repeated trial and error. For new printing tasks, the best matching powder bed temperature sequence prediction model and optimal temperature distribution model can be directly retrieved, which simplifies the control process and ensures that temperature control is on the optimal track from the beginning.

[0083] The process database is a continuously accumulating and optimized knowledge base. Each completed printing task adds data to the database, accumulating over time to continuously refine the model's quantitative parameters. The richer the data, the more mature the model, and the higher the quality of printing control.

[0084] The zoned heating scheme has several advantages: first, it can specifically eliminate the non-uniformity of the powder bed temperature field, improving the molding quality of the parts; second, it simplifies the complexity of the control system by dividing and conquering; and third, it can flexibly adapt to 3D printing powder beds of different sizes and material systems. However, the zone settings should not be too complex; overly fine divisions will increase the control difficulty and computational burden, which is counterproductive. At the same time, the thermal coupling effect between adjacent zones cannot be ignored.

[0085] Step S1200: The temperature of each actual temperature control area is collected over time using a temperature sensor array to form a first temperature time series; real-time thermal images of the entire powder bed surface at different time points are collected using an infrared thermal imager to form a thermal image time series; the production environment data during the 3D printing process is collected over time to form a production environment time series; the production environment time series includes the time series of n3 environmental parameters, including temperature, humidity, oxygen concentration, and dust concentration;

[0086] Specifically, step S1200 involves using multiple sensors to collect real-time data on the powder bed temperature distribution, thermal image data, and production environment data during the 3D printing process, and arranging these data into a corresponding time series. Among these:

[0087] The temperature sensor array consists of several temperature sensors distributed across different areas of the powder bed. Each sensor is responsible for collecting real-time temperature data for its designated area and associates the collected data with a timestamp to form a temperature-time binary. By arranging all the binary data collected by the sensors in chronological order, the first temperature time series is obtained, which reflects the temperature change over time in each actual temperature control area.

[0088] Infrared thermal imagers can sense and record the infrared radiation intensity at various points on the surface of a powder bed, generating a thermal distribution image. By continuously imaging the powder bed surface, a series of thermal images at different times can be obtained. Arranging these thermal images in chronological order forms a thermal image time series. The thermal images visually demonstrate the overall temperature distribution and dynamic changes of the powder bed surface.

[0089] Production environment data includes factors affecting 3D printing quality such as ambient temperature, humidity, oxygen concentration, and dust concentration. This data can be collected in real time using environmental sensors. By mapping the sampled values ​​of each indicator to the sampling time and arranging them into a time series, we obtain the production environment time series. This series reflects the dynamic changes in the printing environment, providing important clues for subsequent anomaly detection and quality prediction.

[0090] After preprocessing the three time series data points described above, such as time alignment and interpolation smoothing, they can be used for subsequent feature extraction and data fusion. This real-time acquisition method based on multi-source heterogeneous data has the following advantages:

[0091] Comprehensiveness: Simultaneously acquiring information on localized powder bed temperature, overall heat distribution, production environment, and other aspects lays a data foundation for a comprehensive understanding of the printing process status.

[0092] Real-time performance: Real-time sampling by the sensor ensures the timeliness of the data, enabling timely response to temperature monitoring and control.

[0093] Dynamics: Time series data reflects the dynamic evolution of the entire printing process, which helps to characterize the time-varying characteristics of the system and grasp its changing patterns.

[0094] Computability: Standardized time series data formats facilitate storage, retrieval, and computational analysis, providing convenience for data-driven process optimization.

[0095] For example, suppose a 3D printing task lasts 30 minutes. During this time, a temperature sensor array collects temperature data from nine control areas every second, an infrared thermal imager captures a thermal image every minute, and environmental sensors collect temperature and humidity data every 10 seconds. At the end of the printing process, a first temperature time series containing 1800 data points, 30 thermal images, and 180 environmental parameter sampling points will be generated. This time series data serves as both a basis for quality control and material for process optimization, and is of significant value in improving the intelligence level of 3D printing.

[0096] Step S1300: Generate a first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series;

[0097] Furthermore, such as Figure 2 As shown, step S1300 includes:

[0098] Step S1310: Extract features from the first temperature time series to obtain a temperature feature series;

[0099] Furthermore, such as Figure 3 As shown, step S1310 includes:

[0100] Step S1311: Divide the first temperature time series into n2 time windows, each time window containing a fixed number of temperature sampling points; n2 is a positive integer;

[0101] Step S1312: Calculate the mean, variance, maximum and minimum temperature for each time window to generate a temperature feature vector;

[0102] Step S1313: Arrange the temperature feature vectors of each time window in chronological order to form a temperature feature sequence.

[0103] Specifically, the purpose of temperature feature extraction is to extract key indicators that characterize temperature changes from raw temperature-time series data, replacing redundant raw data with relatively concise feature representations, thereby reducing the complexity of subsequent calculations. A commonly used feature extraction method is sliding window segmented statistics.

[0104] First, a fixed-width time window (e.g., 30 seconds) is defined, and this window is gradually slid across the temperature time series, moving by half the window length each time to form a series of overlapping time segments. Then, statistical analysis is performed on the temperature data within each segment, calculating its mean (reflecting the average temperature level), variance (reflecting the degree of temperature fluctuation), maximum and minimum values ​​(reflecting the temperature range), and other characteristic indicators. These are then combined into a temperature feature vector, which is associated with the starting timestamp of the corresponding segment.

[0105] Arranging the feature vectors of all segments sequentially in sliding order yields the temperature feature sequence. Compared to the original temperature time series, this feature sequence not only significantly reduces the amount of data but also aggregates temperature information from its neighborhood at each time point, containing local dynamic features and making it more conducive to characterizing the evolution of temperature.

[0106] Taking the example in step S1200, if the first temperature time series contains 1800 temperature sampling points, and the time window is set to 30 seconds (30 sampling points), then sliding through it once yields 60 time segments. Assuming that 4 statistical features are extracted from each segment, the final temperature feature sequence consists of 60 4-dimensional feature vectors arranged in chronological order. This compresses the original 1800×1 temperature sequence into a 60×4 feature sequence, significantly reducing the data dimensionality.

[0107] Step S1310 smooths the data by setting a time window, effectively reducing the impact of instantaneous temperature fluctuations and sensor noise. Statistical features are used to represent the temperature change pattern of each segment, summarizing the main attributes of the original data with a limited amount of information, which is beneficial for subsequent analysis and modeling. The sequential information of the time segments is preserved, and the feature sequence in the time dimension reflects the dynamic process of temperature change, capturing its temporal correlation. Compared with the original time series, the data size of the feature sequence is significantly reduced, which can greatly save subsequent computing resources and time.

[0108] Temperature feature extraction technology has been widely used in industry, such as fault diagnosis, process monitoring, and quality prediction. Introducing it into 3D printing temperature analysis can lay a solid data foundation for intelligent quality control and process optimization. Subsequent temperature prediction and anomaly detection will be based on temperature feature sequences.

[0109] Step S1320: Extract features from the thermal image time series to obtain the thermal image feature sequence;

[0110] Furthermore, such as Figure 4 As shown, step S1320 includes:

[0111] Step S1321: Divide each thermal image into sub-regions corresponding to n1 actual temperature control areas, and extract the grayscale histogram, gradient histogram and texture features of each sub-region to form an image feature vector.

[0112] Step S1322: Concatenate the image feature vectors of all sub-regions at the same time into a complete thermal image feature vector;

[0113] Step S1323: Arrange the feature vectors of the thermal images at all times in chronological order to obtain the thermal image feature sequence.

[0114] Specifically, the purpose of thermal image feature extraction is to transform two-dimensional infrared thermal images into image feature descriptions represented in vector form, which are used to characterize the dynamic changes in the temperature distribution on the surface of the powder bed. Since thermal images contain far more information than a finite number of discrete temperature sampling points, relevant methods of image processing and pattern recognition are needed for feature engineering.

[0115] Typically, each thermal image is first divided into sub-regions corresponding to n1 actual temperature control areas based on their physical spatial locations, ensuring a one-to-one spatial correspondence between these sub-regions and the actual temperature control areas (i.e., the coverage areas of each sensor in the temperature sensor array). Then, a series of visual features are extracted from each sub-region. Commonly used features include:

[0116] Gray-scale histogram: Reflects the statistical characteristics of the distribution of pixel gray-scale values ​​(corresponding to temperature levels) within a sub-region, such as average gray-scale and the ratio of the brightest / darkest pixels.

[0117] Gradient histogram: Reflects the statistical distribution characteristics of the intensity of pixel gray-level changes (corresponding to temperature gradient) within a sub-region, such as average gradient, maximum gradient, etc.

[0118] Texture features: Features that reflect the local spatial relationships of pixels within a sub-region (corresponding to temperature distribution patterns), such as gray-level co-occurrence matrix and local binary patterns.

[0119] Each type of feature can be extracted into several numerical indicators. Combining all the numerical indicators of the same sub-region into a feature vector yields the image feature representation of that sub-region.

[0120] Next, the feature vectors of all sub-regions at the same time are concatenated in a certain order (such as from left to right or from top to bottom) to form a long feature vector, which serves as the feature description of the entire thermal image at that time, reflecting the complete thermal distribution state of the powder bed surface at the current time point.

[0121] Finally, the feature vectors of the thermal images at different times are sorted according to the acquisition time, resulting in a dynamic sequence of thermal image features over time. This sequence characterizes the spatiotemporal evolution of the surface temperature distribution of the powder bed at the image level.

[0122] Using the example from step S1200, assume the powder bed is divided into 9 actual temperature control zones (3×3), and each thermal image is also divided into 9 corresponding sub-regions. If 10 visual features are extracted from each sub-region, a single thermal image will generate a 90-dimensional feature vector. If 30 thermal images are acquired within 30 minutes, the final thermal image feature sequence will be a 30×90 two-dimensional array. Compared to the first temperature sequence, although the thermal image feature sequence has a larger data scale, it is richer in information and more comprehensive in its representation, providing more knowledge sources for subsequent process optimization.

[0123] The main advantages of this method include:

[0124] Comprehensive Information: The thermal image covers every pixel on the powder bed surface, comprehensively recording complete temperature distribution information. Image feature extraction maximizes the extraction of useful information, laying the foundation for global analysis.

[0125] High resolution: Thermal images typically have a much higher spatial resolution than temperature sensor arrays, depicting finer details of temperature distribution and helping to detect anomalies in local areas.

[0126] Spatial correlation: By dividing into sub-regions and integrating them at the feature level, the intrinsic relationship of temperature between different regions can be established, which helps to analyze the transfer and impact of heat between regions.

[0127] Multi-scale representation: Features such as grayscale, gradient, and texture depict temperature distribution patterns from different scales, complementing each other and forming a more accurate and robust temperature field estimate.

[0128] Thermal image feature extraction, a typical application combining infrared thermal imaging technology with advanced visual analysis methods, has been widely used in industries such as manufacturing and medicine. Extending it to online monitoring of 3D printing enables visualized analysis and quality prediction of the production process, improving process efficiency. Subsequent functions such as anomaly detection and temperature field reconstruction will rely on high-quality thermal image feature sequences.

[0129] Step S1330: Extract features from the production environment time series to obtain the production environment feature sequence;

[0130] The feature extraction of the production environment time series includes a first level and a second level; the first level is to extract features from the time series of a single environmental parameter to obtain a feature sequence of the single environmental parameter; the second level is to combine the feature sequences of different environmental parameters.

[0131] Specifically, production environment data reflects the external conditions of the 3D printing process, such as ambient temperature, humidity, oxygen concentration, and dust concentration, all of which can affect the quality of the printed parts. Therefore, it is necessary to extract features from this data for quality correlation analysis.

[0132] The basic idea for extracting production environment features is similar to that of temperature feature extraction in step S1310, also employing a sliding time window segmented statistical method. The difference lies in the fact that production environment parameters are typically multidimensional, meaning multiple parameter values ​​(such as temperature and humidity) exist simultaneously. Therefore, feature extraction needs to be performed at two levels:

[0133] The first level involves feature extraction from the time series of individual parameters. Taking ambient temperature as an example, a time window can be set, and statistical features such as mean and variance can be extracted from the temperature series using a sliding window approach to obtain the ambient temperature feature series. Similar processing is performed on parameters such as humidity and airflow to obtain their respective feature series.

[0134] The second level involves combining the feature sequences of different parameters to form a multi-dimensional comprehensive feature sequence of the production environment. The simplest combination method is to concatenate the feature vectors of different parameters at the same time to form a long vector. Alternatively, one can consider extracting the cross-features between different parameters, such as the correlation between temperature and humidity, to uncover deeper environmental impact patterns.

[0135] Taking step S1200 as an example, suppose environmental temperature and humidity data are collected at 180 time points. Extracting features (mean and variance) within a 30-second time window yields an environmental temperature feature sequence (6×2-dimensional) and a humidity feature sequence (6×2-dimensional). Aligning and concatenating these sequences along the time dimension results in a 6×4-dimensional production environment feature sequence. If we consider adding temperature and humidity correlation features (such as covariance), the feature dimensions will be further expanded.

[0136] The significance of extracting production environment features lies in:

[0137] External condition characterization: Production environment data is the external driving force for the evolution of temperature field. Feature extraction from it can quantitatively characterize the environmental conditions in which the printing process takes place.

[0138] Quality correlation analysis: Environmental factors are often closely related to printing quality, and environmental characteristics provide important variables for the analysis of quality impact mechanisms and the construction of prediction models.

[0139] Anomaly Diagnosis: Sudden changes in environmental parameters usually indicate some kind of abnormality. Analyzing the characteristic changes can help diagnose problems in the printing process.

[0140] Controllability: Compared to temperature itself, environmental conditions are generally easier to intervene in and control externally. Extracting environmental characteristics can provide new ideas for optimizing process parameters.

[0141] Step S1340: Align the temperature feature sequence, thermal image feature sequence, and production environment feature sequence in the time dimension to form the first comprehensive feature sequence.

[0142] Specifically, generating the comprehensive feature sequence requires fusing feature sequences from different sources along the time axis. Since the sampling frequencies of the three types of features (temperature, thermal image, and environment) may differ, timestamp matching and alignment are necessary first. Using the thermal image as the benchmark with the lowest sampling frequency, the other two types of feature sequences are interpolated to unify them to the same sampling time as the thermal image.

[0143] After alignment, the feature vectors corresponding to the three types of features at the same time are concatenated to obtain the comprehensive feature vector at that time. This vector integrates the internal and external state information of the system at the current time point, including temperature distribution, thermal field image, environmental parameters, etc., and can be represented by a very long real-valued vector.

[0144] Arranging the comprehensive feature vectors of all sampling moments sequentially forms the first comprehensive feature sequence. This organically unifies the originally scattered multi-source data onto a single time axis, creating a trajectory of the entire printing process's state evolution. This representation preserves dynamic temporal information while also taking into account the inherent relationships between multiple physical quantities, providing a richer information source for tasks such as state monitoring, anomaly diagnosis, and process optimization.

[0145] Let's take step S1200 as an example. Assume the printing process lasts 30 minutes, during which the infrared thermal imager acquires an image every minute; the thermal image feature sequence length is 30. Interpolating both the temperature feature sequence and the environmental feature sequence to the same 30 time points yields 30×... and 30× The characteristic matrix ( , (This refers to the corresponding eigenvector dimension). Then, these three types of matrices are concatenated column-wise to obtain a 30×(90+) matrix. + The ultra-long matrix of ) is the first comprehensive feature sequence (here it is assumed that the thermal image features are 90-dimensional).

[0146] This comprehensive feature extraction method has the following advantages:

[0147] Multi-source data fusion: Temperature, thermal imaging, and environmental data contain rich and complementary process status information, and the comprehensive features can reflect the working status of the system from different perspectives.

[0148] Temporal information preservation: By introducing the time dimension, the comprehensive feature sequence describes the evolution of the state over time, which helps to understand the dynamic behavior of the system.

[0149] Globally consistent representation: Unifying different physical quantities into a characteristic space can establish the correlation between internal and external states and reveal their interaction mechanism.

[0150] Dimensionality reduction alleviates redundancy: Original multi-source data contains a large amount of data redundancy and noise. Through feature transformation, the dimensionality of the data can be significantly reduced while retaining the main information, which is beneficial for subsequent calculations.

[0151] Comprehensive feature sequences are crucial for condition monitoring and quality optimization. Based on these comprehensive features, intelligent applications such as visualized analysis of the printing process, real-time anomaly detection, and dynamic optimization of process parameters can be achieved, ultimately increasing yield, shortening production cycles, and saving costs. It is a key step in transforming multi-source sensor information into usable knowledge.

[0152] In summary, the core of step S1300 is to extract a comprehensive feature sequence from the original multi-source time series data that can fully reflect the state of the 3D printing system. It is a complex process consisting of sub-steps such as local feature extraction, heterogeneous feature fusion, and time information integration, involving multiple technical fields such as signal processing, image analysis, and multivariate statistics.

[0153] First, features are extracted from three types of time series: temperature, thermal images, and environment, resulting in corresponding feature sequences. These features must summarize the main patterns of the original data while highlighting the unique characteristics of different physical quantities. This requires designing different feature engineering strategies for different types of data, such as sliding window statistics and image block extraction. Next, the extracted features are aligned along the time dimension to form a unified timeline representation. This requires matching the sampling frequencies of different physical quantities, potentially involving data preprocessing operations such as interpolation and downsampling. The aligned feature sequence should reflect the synchronous changes in the system state. Finally, the aligned multi-source feature sequences are concatenated along the feature dimensions to obtain the first comprehensive feature sequence. This ultra-long time series comprehensively depicts the overall state of the 3D printing system at different times, integrating multi-source information reflecting different physical levels, providing an information foundation for quality monitoring and optimization.

[0154] In summary, comprehensive feature sequences are a crucial cornerstone for the intelligentization of the 3D printing process. They integrate multi-source information reflecting the internal and external states of the system and fully exploit its temporal dynamics, thus enabling a more comprehensive and accurate characterization of the complex behavior of the printing process. This lays a solid data representation foundation for achieving object-oriented and digital quality management.

[0155] Step S2000: Extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database; obtain the temperature prediction smoothing curve for each actual temperature control area based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model; obtain the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task based on the optimal temperature distribution model; and conduct graded early warning for temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range.

[0156] Further, step S2000 includes:

[0157] Step S2100: Extract the key process parameters of the current printing task to form a process parameter tuple. Use the process parameter tuple to retrieve the optimal powder bed temperature sequence prediction model from the process database. Input the first comprehensive feature sequence into the optimal powder bed temperature sequence prediction model to generate the first temperature prediction sequence for each actual temperature control area in the future. Perform smoothing and noise reduction processing on the first temperature prediction sequence to obtain the temperature prediction smooth curve for each actual temperature control area.

[0158] Further, step S2100 includes:

[0159] Step S2110: Extract key process parameters from the process configuration file of the current printing task to form a process parameter tuple, and use the process parameter tuple to retrieve the optimal powder bed temperature sequence prediction model in the process database.

[0160] Furthermore, such as Figure 5 As shown, step S2110 includes:

[0161] Step S2111: Extract key process parameters from the process configuration file of the current printing task to form a process parameter tuple; the key process parameters include powder bed size, material type, layer thickness and printing speed.

[0162] Specifically, the process requirements and temperature characteristics of different printing tasks vary greatly, therefore, it is necessary to select an appropriate prediction model for specific process parameters. Typically, the key process parameters affecting the evolution of the temperature field include:

[0163] Toner bed size: Reflects the spatial span of the printed part and determines the distance of heat conduction. The larger the size, the more uneven the temperature field distribution, and the higher the requirements for the spatial adaptability of the model.

[0164] Material type: Different materials, such as titanium alloys, copper alloys, and cemented carbides, have significantly different physical properties such as thermal conductivity and heat capacity, resulting in variations in their heating rates, temperature stability, and thermal deformation characteristics. The model needs to be tailored to these material properties.

[0165] Layer thickness: The thickness of the printed layer affects the accumulation and dissipation of heat in the vertical direction, thus affecting the period and amplitude of temperature fluctuations over time. Layer thickness is a key factor affecting the temperature distribution along the z-axis.

[0166] Printing speed: reflects the rate at which heat is applied per unit of time. The faster the speed, the faster the heat accumulates, the faster the temperature rises, and the greater the temperature fluctuations. A slower speed results in more gradual temperature changes.

[0167] The above parameters are read from the printing task's process configuration file and arranged into a parameter tuple in the order of powder bed size, material type, layer thickness, and printing speed. This tuple serves as an index for the subsequent retrieval and prediction model. This approach avoids using all process parameters and focuses only on the primary parameters, thus reducing the search space.

[0168] Taking the common Fused Deposition Modeling (FDM) printing as an example, the typical range of values ​​for key parameters is as follows:

[0169] Powder bed dimensions: 150x150, 200x200, 300x300 (mm);

[0170] Material types: Polylactic acid (PLA), ABS, Nylon, Polyethylene terephthalate (PETG).

[0171] Layer thickness: 0.1, 0.2, 0.3 (mm);

[0172] Printing speed: 40, 60, 80 (mm / s);

[0173] An example of a process parameter tuple could be (200x200, PLA, 0.2, 60), representing printing with PLA material at a layer thickness of 0.2 mm and a speed of 60 mm / s on a 200 mm square powder bed. Clearly, different parameter combinations correspond to different temperature characteristics, requiring different models to fit them.

[0174] Step S2112: Use the process parameter tuple to retrieve the optimal powder bed temperature sequence prediction model from the process database;

[0175] The process parameter tuples are compared item by item with the process parameter combinations corresponding to the pulverized bed temperature sequence prediction model in the process database. If all key process parameters are equal, they are considered to be a perfect match. The pulverized bed temperature sequence prediction model that is a perfect match is taken as the optimal pulverized bed temperature sequence prediction model.

[0176] If no perfectly matching pulverized bed temperature sequence prediction model is found, the similarity between the process parameter tuple and the process parameter combination corresponding to the pulverized bed temperature sequence prediction model in the process database is calculated, and the one with the highest similarity is selected as the optimal pulverized bed temperature sequence prediction model.

[0177] Specifically, step S2112 involves identifying the powder bed temperature sequence prediction model that best matches the process parameters of the current printing task from the existing process database, and using this model as the optimal model for prediction. The key to this process is comparing the similarity of the process parameters.

[0178] Process parameters refer to a series of settings that affect the quality of 3D printing, such as powder material type, particle size distribution, powder thickness, laser power, scanning speed, and scanning spacing. These parameters are closely related to the formation and evolution of the temperature field. For example, different materials will exhibit different temperature rise curves under the same laser irradiation due to their different thermal properties; while excessively high laser power or excessively low scanning speed may lead to localized overheating of the powder bed, causing defects such as warping and deformation.

[0179] The process parameter tuple is an ordered array consisting of the values ​​of all process parameters for the current printing task. It represents a specific printing scheme. This tuple is compared item by item with the process parameter combination corresponding to each prediction model in the process database to determine whether they are completely consistent in each parameter. If all parameters are completely equal, the two are considered to be a perfect match.

[0180] In the case of a perfect match, the model corresponds to a historical printing scheme that is completely equivalent to the current scheme, and its prediction results can be directly applied to this printing. Therefore, it is determined as the optimal powder bed temperature sequence prediction model. This matching method ensures the model's relevance and applicability.

[0181] However, in practical applications, due to the limitations of the parameter combination space, it is often difficult to find a perfect match. In this case, approximate matching is required. A common strategy is to use a similarity metric.

[0182] Similarity is a metric function that characterizes the degree of proximity between two objects in a certain feature space. Common similarity metrics include Euclidean distance, cosine similarity, and KL divergence. For process parameters, a suitable similarity function can be designed based on their physical meaning and numerical characteristics. A simple approach is to assign different weights to the differences of each parameter and then linearly combine the weighted differences. The weights can be preset based on expert experience or dynamically generated through data analysis.

[0183] The prediction model with the highest similarity to the current process parameter tuple in the process database is considered a suboptimal match and is determined as the optimal model for this prediction. This flexible matching strategy relaxes the applicable conditions to some extent and expands the scope of model applicability. As the database capacity continues to expand, the accuracy of approximate matching will also improve.

[0184] For example, suppose the process parameters for the current printing task are: material = PA12, powder thickness = 0.12mm, laser power = 50W, and scanning speed = 2500mm / s. A search in the process database reveals a model with the most similar parameter combination: material = PA12, powder thickness = 0.1mm, laser power = 48W, and scanning speed = 2400mm / s. Despite the differences, these two schemes are quite close in key parameters such as material and energy density. It is reasonable to assume that the prediction model corresponding to the latter is also applicable to the former.

[0185] This model selection method based on similarity comparison has the following advantages:

[0186] Highly targeted: By matching process parameters, the most relevant prediction model for the current printing task can be identified, ensuring the consistency between temperature prediction and actual process.

[0187] Rapid response: No need to repeatedly train the model; existing model knowledge can be directly called, which can greatly shorten the prediction response time and improve real-time performance.

[0188] Experience reuse: By accumulating process databases, a map of predictive models can be formed in the parameter space for reference and reuse in subsequent printing tasks, continuously enriching printing experience.

[0189] Self-improvement: As the database is continuously expanded and updated, the matching models will become more and more accurate, the probability of matching failure will gradually decrease, and the predictive ability of the entire system will continue to improve.

[0190] This data-driven intelligent scheduling method represents a significant development direction for 3D printing quality control. By continuously collecting and analyzing data on processes, materials, and equipment, an experiential knowledge base is formed. Patterns are then extracted and models are built from this knowledge base, which are then applied to actual production, creating a virtuous cycle of knowledge reuse and intelligent decision-making. This comprehensively improves the intelligence level and industrial efficiency of 3D printing. Step S2112 specifically matches computing resources, avoiding the resource waste that may result from blind experimentation. This embodies a demand-driven computing philosophy, offering important insights for process modeling and quality optimization in 3D printing.

[0191] Step S2120: Input the first comprehensive feature sequence into the optimal powder bed temperature sequence prediction model to generate the first temperature prediction sequence for each actual temperature control area in the future.

[0192] Further, step S2120 includes:

[0193] Step S2121: Match and align the first comprehensive feature sequence according to the input format required by the optimal powder bed temperature sequence prediction model;

[0194] Step S2122: Set the future prediction time length and sampling interval, and gradually predict the temperature prediction value at each future moment;

[0195] Step S2123: Arrange the temperature prediction values ​​in chronological order to form the first temperature prediction sequence.

[0196] Specifically, step S2120 uses the optimal pulverized bed temperature sequence prediction model obtained in step S2110 to predict the temperature change trend over a future period. First, the first comprehensive feature sequence needs to be matched and aligned with the input format required by the prediction model. This may involve preprocessing operations such as adjusting the feature vector dimensions and data normalization to meet the model's input requirements. Then, a future prediction time length (e.g., 30 minutes) and a sampling interval (e.g., sampling once every 10 seconds) are set, and these two parameters, along with the aligned feature input, are passed to the prediction model. The model advances forward step by step at 10-second intervals, predicting the temperature value 10 seconds after the current moment at each step. After 180 consecutive iterations, the model outputs a temperature prediction sequence of one value every 10 seconds over the next 30 minutes, totaling 180 predicted values. Arranging these discrete predicted values ​​in chronological order forms the first temperature prediction sequence. This sequence, with a 10-second time granularity, depicts the expected trajectory of pulverized bed temperature changes over the next 30 minutes and can be used for early warning and control. Data-driven sequence forecasting avoids the subjectivity of relying solely on experience, thus improving the scientific rigor and reliability of predictions. Furthermore, by setting a time span for predicting the future, the foresight and practicality of the forecasts are ensured.

[0197] Step S2130: The first temperature prediction sequence is smoothed and denoised to eliminate high-frequency oscillations in the predicted values, and a temperature prediction smooth curve for each actual temperature control region is obtained.

[0198] Specifically, due to various factors, the temperature prediction sequence provided by the model is often not a smooth curve, but rather a broken line with spikes and noise. To more robustly depict the temperature change trend, the prediction sequence needs to be smoothed and denoised. Commonly used smoothing algorithms include simple moving average, weighted moving average, and Savitzky-Golay filtering. Moving average uses a sliding window to filter the local mean of the sequence, while Savitzky-Golay considers local high-order polynomial fitting, which can better preserve data features while smoothing. After smoothing, the temperature prediction broken line is free of high-frequency noise and transformed into a smooth curve, presenting the temperature change trend in a clearer and more stable form.

[0199] Step S2200: Based on the key process parameters of the current printing task, retrieve the optimal temperature distribution model from the process database, generate the target temperature distribution matrix based on the optimal temperature distribution model, calculate the upper and lower thresholds of the target temperature distribution matrix, and obtain the allowable temperature variation range of each actual temperature control area.

[0200] Further, step S2200 includes:

[0201] Step S2210: Based on the key process parameters of the current printing task, retrieve the optimal temperature distribution model from the process database; the optimal temperature distribution model includes the division of temperature control areas and the target temperature value for each temperature control area.

[0202] Specifically, the purpose of step S2210 is to find an optimal temperature distribution model that best matches the current printing task, serving as a reference for determining the target temperature range. Key process parameters include the type of printing material (e.g., ABS, PLA) and printing process parameters (e.g., printing speed, layer height). Using these key process parameters as search criteria, the optimal temperature distribution model that perfectly matches the target temperature distribution model is searched in the process database. The optimal temperature distribution model represents the most ideal temperature distribution scheme. "Optimal" means that the printing scheme corresponding to this temperature distribution model can achieve the best balance between production efficiency and energy consumption while ensuring product quality. However, a perfect match in ideal circumstances is not always achievable in practice. If a model that precisely matches both the material and process parameters cannot be found, a less ideal model is selected, assuming the material type is the same, choosing the model with the closest process parameters. This approximate matching largely ensures the applicability of the temperature distribution model. Whether it's an exact match or an approximate match, the finally selected optimal model will be loaded into memory, providing a reference for subsequent determination of the target temperature range. Reusing knowledge from the process database avoids redundant work and combines prior experience with the current task.

[0203] The optimal temperature distribution model is an ideal powder bed temperature distribution scheme. It is the best temperature distribution obtained for specific printing materials (such as ABS, PLA) and printing process parameters (such as printing speed, layer height).

[0204] The optimal temperature distribution model consists of two parts:

[0205] Temperature control zone division: The model divides the powder bed into several temperature control zones, and each zone can be heated independently.

[0206] Target temperature values ​​for each temperature control zone: The model provides the ideal temperature value that each temperature zone should achieve, which is crucial for ensuring print quality.

[0207] The optimal temperature distribution model embodies the best balance between ensuring product quality and balancing production efficiency and energy consumption. It is the culmination of knowledge and experience from numerous past printing practices. In practical application, the optimal temperature distribution model needs to be geometrically transformed based on the current toner bed size, and the target temperature value needs to be fine-tuned to generate a target temperature distribution matrix that fits the current printing task.

[0208] Step S2220: Combining the powder bed size and the actual temperature control area, the optimal temperature distribution model is adaptively modified to generate the target temperature distribution matrix;

[0209] Furthermore, such as Figure 6 As shown, step S2220 includes:

[0210] Step S2221: Align the temperature control region of the optimal temperature distribution model with the actual temperature control region of the powder bed. If the particle size of the temperature control region of the optimal temperature distribution model is inconsistent with that of the actual temperature control region, then perform interpolation or merging processing on the optimal temperature distribution model to make the temperature control region of the optimal temperature distribution model match the actual temperature control region.

[0211] Methods for aligning the temperature control region of the optimal temperature distribution model with the actual temperature control region of the powder bed include:

[0212] ;

[0213] ;

[0214] in:

[0215] : The temperature control region matrix of the optimal temperature distribution model, used to describe the ideal temperature distribution of the powder bed during the printing process;

[0216] The actual temperature control region matrix of the toner bed is used to describe the actual temperature distribution of the toner bed during the printing process.

[0217] Aligned temperature control region matrix;

[0218] : The Middle Line number The temperature values ​​of the temperature control zone are listed;

[0219] : The temperature value of the temperature control zone in the m-th row and n-th column;

[0220] M: The total number of rows in the temperature control region matrix of the optimal temperature distribution model;

[0221] N: The total number of columns in the temperature control region matrix of the optimal temperature distribution model;

[0222] : Nonlinear adjustment factor 1 (usually a positive number) is used to adjust the degree of influence of the weights;

[0223] Nonlinear adjustment factor 2 (usually a positive number) is used to further adjust the degree of nonlinearity of the interpolation results;

[0224] I: The row number of the aligned temperature control region matrix;

[0225] J: The number of columns in the aligned temperature control region matrix;

[0226] Weighting factor, representing Temperature value in row m and column n The Middle Line number The contribution of the temperature values; It can be determined based on distance and the area of ​​regional overlap;

[0227] A small constant to prevent division by zero errors;

[0228] This formula uses a dual nonlinear adjustment factor. and It allows for more flexible adjustment of weight allocation during the interpolation process, adapting to different practical needs and complex situations; ensures consistency between the optimal temperature distribution model and the actual temperature control area division, reducing errors; avoids abrupt changes in temperature values, ensuring the smoothness of the temperature field; and is applicable to matrices of different sizes and resolutions, as well as different temperature control requirements.

[0229] Step S2222: Perform a geometric transformation on the optimal temperature distribution model based on the powder bed size to obtain an optimal temperature distribution transformation model that is adapted to the powder bed size.

[0230] Step S2223: Fine-tune the target temperature value in the optimal temperature distribution transformation model to generate the adjusted optimal temperature distribution transformation model;

[0231] Methods for fine-tuning the target temperature value in the optimal temperature distribution transformation model include:

[0232] ;

[0233] in:

[0234] : Represents the target temperature value after adjustment of the temperature control area in row i and column j;

[0235] The target temperature value of the temperature control region in the i-th row and j-th column of the optimal temperature distribution model;

[0236] Actual temperature deviation, i.e. ,in, The actual temperature value of the temperature control zone in the i-th row and j-th column;

[0237] Predicted temperature deviation, i.e. ,in, The predicted temperature value of the temperature control region in the i-th row and j-th column of the powder bed output by the optimal powder bed temperature sequence prediction model;

[0238] : Factors affecting the production environment, representing the influence of the production environment parameters of the temperature control area in the i-th row and j-th column on the temperature;

[0239] : Temperature change rate, representing the rate of change of temperature in the i-th row and j-th column of the optimal temperature distribution model over time, which can be calculated by recording the target temperature values ​​at multiple time points;

[0240] Weighting coefficient for actual temperature deviation;

[0241] Weighting coefficients for predicted temperature deviations;

[0242] Weighting coefficients of factors influencing the production environment;

[0243] Weighting coefficient for the rate of temperature change;

[0244] : Indicates time;

[0245] This formula comprehensively considers factors such as actual temperature deviation, predicted temperature deviation, and the influence of the production environment, and introduces the temperature change rate for fine-tuning of temperature, thereby ensuring the accuracy and stability of temperature control; through reasonable temperature control, the interlayer adhesion strength and surface quality of printed parts are improved; and through precise temperature control, energy consumption is minimized while ensuring printing quality.

[0246] Step S2224: Discretize the adjusted optimal temperature distribution transformation model to generate a target temperature distribution matrix corresponding to the actual temperature control area.

[0247] Specifically, firstly, the temperature control zoning scheme in the optimal temperature distribution model is aligned and compared with the actual temperature control area layout of the powder bed. Ideally, the optimal temperature distribution model and the actual powder bed zoning should completely overlap. However, in reality, there are often differences in particle size or shape between the two. In this case, mathematical transformations such as interpolation or merging are needed to perform on the optimal model while maintaining the thermal distribution characteristics.

[0248] For example, if the optimal temperature distribution model divides the powder bed into a 6×6 grid, but the actual powder bed only has 3×3 temperature control zones, then the optimal temperature distribution model needs to be merged. Specifically, the average temperature values ​​of several small grids within the same actual zone in the optimal temperature distribution model are taken as the optimized temperature for that zone. Conversely, if the actual zone granularity is finer, the optimal model needs to be interpolated, dividing the large grid into several small regions and interpolating the target temperature values ​​for each small region.

[0249] This step enables spatial registration between the temperature control zones of the optimal temperature distribution model and the actual powder bed, laying the foundation for subsequent model applications. This partitioning mapping concept draws on mesh subdivision and merging techniques in computer graphics, enabling the conversion between different granularity representations of thermal distribution.

[0250] The size of the powder bed is a crucial geometric parameter affecting temperature distribution. Even with identical materials and processes, powder beds of different sizes will exhibit different optimal temperature fields. Therefore, it is necessary to geometrically transform the optimal temperature distribution model according to the actual powder bed size. This can be achieved using scaling transformations in computer graphics to proportionally enlarge or reduce the optimal temperature distribution model, making it spatially compatible with the actual powder bed. During the transformation, the original heat distribution pattern and temperature gradient remain unchanged, but a linear scaling adjustment is performed numerically.

[0251] For example, if the optimal temperature distribution model corresponds to a 300mm × 300mm powder bed, while the actual powder bed is 200mm × 200mm, then the model needs to be scaled down by a factor of 3 / 2. After the transformation, the target temperature value of each zone in the original model is also scaled down by a factor of 3 / 2. Conversely, if the actual powder bed is larger, a scaling-up transformation is performed.

[0252] With appropriate geometric transformations, the optimal temperature distribution model can be well adapted to powder beds of different sizes. This method of achieving scale matching through geometric transformation is simple and effective, avoiding the enormous workload of repeatedly training the model, and has great engineering application value.

[0253] Due to the specific characteristics of actual production environments, the optimal temperature distribution model corrected according to the above steps may still deviate somewhat from actual requirements. To further improve the applicability of the model, it is necessary to fine-tune the key parameter, namely the target temperature value.

[0254] Analysis of actual production data revealed that the preset temperature values ​​in the model were generally too low, leading to increased surface roughness in the printed parts. Therefore, it was decided to increase the target temperature values ​​for each zone by 5°C based on the modified model. The increased temperature is closer to the material's optimal melting point and better matches the mold surface temperature, thus improving thermal deformation.

[0255] On the other hand, considering the special requirements of the powder bed temperature for the first layer printing, the target temperature of the bottom heating zone should be appropriately increased at the beginning of the printing stage to ensure sufficient bonding between the powder and the forming platform. Experiments have shown that increasing the target temperature of the bottom zone by 8°C during the first layer printing can significantly improve the bonding strength of the first layer and reduce warping deformation.

[0256] The degree of fine-tuning should be determined based on actual process requirements and material properties. Parameters are continuously optimized and adjusted through repeated process experiments and data analysis. Compared to complex parameter optimization algorithms in machine learning, this application-oriented fine-tuning method is more intuitive and easier to implement. The optimization process also incorporates the experience of engineering experts, balancing theoretical guidance with practical verification.

[0257] The fine-tuned optimal temperature distribution model is ready for practical application. To facilitate execution by the temperature control system, the model needs to be discretized to generate a target temperature matrix that corresponds one-to-one with the actual temperature control region.

[0258] Discretization refers to dividing a continuous temperature field into several discrete small regions in space while preserving the shape of the heat distribution. Each small region corresponds to a constant target temperature value. This process can be achieved using the mesh generation method of finite element analysis.

[0259] The specific approach is as follows: using the actual powder bed heating unit as the basic unit, the optimal temperature distribution model is meshed. The temperature field falling within each mesh is discretized into a constant value, which is the target temperature of that mesh. The target temperatures of all meshes are arranged in matrix form, thus obtaining the discretized target temperature distribution matrix. The number of rows and columns of the matrix corresponds to the number of rows and columns of the actual temperature control zones.

[0260] For example, suppose a certain type of powder bed heating platform consists of 20×20 heating units. The fine-tuned optimal temperature distribution model is discretized into 400 grid cells, each corresponding to a temperature control unit. The target temperature values ​​of these grid cells are arranged in a 20x20 matrix, generating a 400-dimensional temperature vector. This vector specifies the temperature setpoint for each temperature control unit and can be directly used to guide the temperature control of the powder bed.

[0261] While discretization simplifies the model to some extent, it also significantly improves its operability. The generated temperature matrix matches the powder bed hardware system and can be seamlessly integrated into the CNC system. The matrix size is also moderate, facilitating real-time calculation and storage. This engineering-application-oriented model transformation and encapsulation method is widely used in the 3D printing field, possessing both theoretical basis and practical value.

[0262] Step S2220, following the above steps, perfectly adapts the optimal temperature distribution model to the target powder bed. The modified model overcomes the gap between theoretical research and engineering practice, becoming more realistic. Simultaneously, through fine-tuning and discretization of the model, the operability and robustness of the solution are further improved, providing effective theoretical guidance for intelligent temperature control. This model-engineering approach can be extended to the optimization control of other industrial systems, showing broad application prospects.

[0263] (1) Spatial registration was used to achieve a consistent match between the optimal temperature model and the target powder bed in the temperature control zone, thus solving the compatibility problem between the theoretical model and the actual system. Interpolation / merging operations can achieve smooth conversion between temperature representations of different particle sizes.

[0264] (2) Geometric transformation achieves the matching of model size with powder bed size, so that the trained model can be applied to powder beds of different sizes, avoiding the cost of repeated modeling and improving the versatility of the model.

[0265] (3) Parameter fine-tuning further enhances the fit between the model and the process based on model transformation. Through the analysis and feedback of actual production data, key temperature parameters are adjusted in a targeted manner, making the optimization scheme closer to actual needs.

[0266] (4) Discretization is a key step in model engineering. By meshing and discretizing the temperature field, a directly usable temperature control matrix is ​​generated, allowing the optimization scheme to be seamlessly embedded into the CNC system. At the same time, appropriate discretization reduces model complexity and improves computational efficiency while ensuring the characteristics of heat distribution.

[0267] Step S2230: Calculate the upper and lower thresholds of the target temperature distribution matrix to obtain the allowable temperature variation range for each actual temperature control zone.

[0268] Methods for calculating the upper and lower thresholds of the target temperature distribution matrix to obtain the allowable temperature variation range for each actual temperature control zone include:

[0269] ;

[0270] ;

[0271] ;

[0272] in:

[0273] The first element in the target temperature distribution matrix Line number List the actual temperature values ​​of the temperature control zone;

[0274] : No. Line number List the allowable temperature variation range of the actual temperature control area;

[0275] : No. Line number The lower limit threshold of the actual temperature control area;

[0276] : No. Line number List the upper temperature limit threshold of the actual temperature control area;

[0277] : The average temperature value of the target temperature distribution matrix;

[0278] Nonlinear adjustment factor three is used to adjust the temperature variation range;

[0279] : Environmental factor adjustment coefficient, which takes into account the influence of environmental factors such as temperature and humidity on the temperature variation range;

[0280] The sensitivity coefficient of the printing material reflects the degree to which the material is sensitive to temperature changes;

[0281] The standard deviation of production environment data reflects the volatility of environmental conditions.

[0282] This formula takes into account nonlinear adjustment factors, material sensitivity, and environmental factors, making the calculation results closer to reality. The formula can be adjusted according to different materials and environmental conditions, and has wide applicability. By smoothing the temperature change range, it reduces the impact of temperature fluctuations on printing quality and improves the surface quality and structural strength of printed parts.

[0283] Specifically, the purpose of step S2230 is to determine the reasonable range of temperature regulation for each actual temperature control zone based on the target temperature. Material thermal properties are the primary consideration; only by controlling temperature fluctuations within the acceptable range for the material can the forming quality of the printed parts be guaranteed from the source. Based on this, the specific determination of the temperature threshold also needs to balance printing efficiency and energy consumption. A lenient threshold helps reduce the number of heating power adjustments and lower energy consumption, but may come at the cost of finished product accuracy. A strict threshold, while helping to improve temperature constantness and printing quality, will lead to frequent heating adjustments and increased energy consumption. The optimal selection of the upper and lower temperature thresholds is a "three-body problem" between quality, efficiency, and energy consumption, and needs to be flexibly determined according to the specific application and requirements. Integrating the target temperature and threshold temperature into a ternary set provides a concise and unified data representation, facilitating the programmed implementation of the temperature control unit. The dynamic temperature control scheme based on quantized thresholds overcomes the limitations of constant temperature control, providing greater flexibility and room for imagination for adaptive optimization of the printing process.

[0284] In step S2300, the temperature prediction smoothing curve is compared with the allowable temperature change range to determine whether a temperature anomaly occurs within the prediction time. If there is no temperature anomaly, the process returns to step S1200 to continue temperature detection and real-time data acquisition. If a temperature anomaly occurs, the over-limit amplitude and duration of the predicted temperature value in the temperature prediction smoothing curve are calculated, and a graded warning for the temperature anomaly is generated based on the over-limit amplitude and duration.

[0285] The temperature anomaly warning levels include: Level 1 warning, Level 2 warning, Level 3 warning, and Level 4 warning;

[0286] Level 1 warning: Exceeding limit amplitude AM < 5%; Duration DU < 1 min;

[0287] Level II Warning: 5% ≤ Exceedance Amplitude AM < 10%; 1 min ≤ Duration DU < 5 min;

[0288] Level 3 warning: 10% ≤ Exceedance amplitude AM < 20%; 5min ≤ Duration DU < 10min;

[0289] Level 4 warning: Exceeding the limit AM ≥ 10%; Duration DU ≥ 10 min.

[0290] Specifically, step S2300 determines whether the system will experience temperature anomalies based on the predicted smoothing curves of the actual temperature control zones of the toner bed within the future time window. Temperature anomalies refer to situations where the printing process deviates significantly from the set process temperature, such as inadequate heating leading to poor powder melting, or excessively high temperatures causing material thermal degradation. Temperature anomalies significantly affect print quality, making it crucial to predict and warn of them before they occur.

[0291] The core of this step lies in comparing the predicted temperature with the allowable range. In practice, a moving window method can be used, sliding the prediction window along the time axis with a fixed step size to obtain a series of predicted temperature values ​​for future moments. Then, each predicted value is checked to see if it falls within the allowable range. If all temperature values ​​within the predicted time fluctuate within the range, the process can be considered under control and the temperature is normal. At this point, return to the data acquisition step and continue monitoring.

[0292] If the predicted temperature exceeds the range at any given time, it indicates that the system may be about to enter an abnormal state, requiring timely warning. To avoid false alarms, observation can continue for a period of time. If subsequent consecutive data points remain outside the limits, the warning is valid. Based on the duration and magnitude of the deviation, temperature anomalies can be classified into different severity levels. Generally, the greater the deviation and the longer the duration, the more severe the deviation from normal process values.

[0293] This step uses a four-level early warning system:

[0294] Level 1 (Mild Anomaly): Exceeding the allowable range by less than 5% and lasting for less than 1 minute. This situation is usually caused by transient disturbances, has a minor impact, and requires no special handling.

[0295] Level 2 (Moderate Abnormality): Exceeding limits by 5%-10%, lasting 1-5 minutes. At this point, close monitoring of the abnormal area is necessary, and the power of the relevant heating units can be fine-tuned if required.

[0296] Level 3 (Severe Abnormality): Exceeding limits by 10%-20%, lasting 5-10 minutes. The system has clearly deviated from normal operating conditions, requiring the activation of the emergency plan, adjustment of process parameters, or suspension of printing.

[0297] Level 4 (Extremely Abnormal): Exceeding limits by more than 20% for more than 10 minutes. This indicates a system malfunction, requiring immediate halting of printing and investigation of the cause of the failure. The entire batch of products must be scrapped.

[0298] Accurately determining the level of anomaly requires extensive practical production experience. The aforementioned thresholds can be derived from statistical patterns through statistical analysis of historical data. In practical applications, the early warning grading standards can be dynamically adjusted based on product quality feedback.

[0299] In summary, this step, based on predictive models and statistical analysis methods, enables early warning of temperature anomalies. By differentiating anomaly levels, it provides a reference for subsequent control decisions and manual intervention. Compared to traditional post-incident inspection, predictive maintenance can significantly reduce rework and defect rates. Integrating this method into a 3D printing intelligent monitoring platform can effectively improve the consistency and reliability of print quality.

[0300] (1) By using the moving time window method to obtain the predicted temperature values ​​at various future times, and comparing them with the allowable range, it is possible to determine in real time whether the system will enter an abnormal state. Compared with the traditional post-event temperature check, this forward-looking prediction can buy valuable time for emergency response.

[0301] (2) By continuously tracking the extent and duration of exceeding limits, temperature anomalies are classified into different levels, providing quantitative indicators for process adjustments. This avoids large-scale waste due to anomalies.

[0302] (3) By combining process experience and historical data analysis, statistical inferences are made about the anomaly level thresholds, making the early warning classification more convincing and operable. At the same time, the classification scheme can be dynamically optimized in production practice in order to achieve the best balance between early warning accuracy, real-time performance and cost.

[0303] Step S3000: Obtain the second temperature distribution matrix according to the temperature anomaly warning level; collect the current temperature of each actual temperature control area in real time to form a second temperature data matrix; compare the second temperature data matrix with the second temperature distribution matrix to generate a temperature deviation matrix; adjust the temperature of each actual temperature control area according to the temperature deviation matrix; evaluate the adjustment effect after the temperature adjustment is completed.

[0304] Further, step S3000 includes:

[0305] Step S3100: Obtain the second temperature distribution matrix based on the temperature anomaly warning level;

[0306] Further, step S3100 includes:

[0307] Step S3110: If the temperature anomaly warning level is Level 1 or Level 2, then the initial temperature distribution matrix is ​​adaptively fine-tuned according to the temperature anomaly warning level to generate a second temperature distribution matrix.

[0308] Furthermore, such as Figure 7 As shown, step S3110 includes:

[0309] Step S3111: Map the Level 1 warning and Level 2 warning to adjustment intensity coefficients, where the adjustment intensity coefficient for Level 2 warning is greater than that for Level 1 warning;

[0310] Step S3112: Identify the actual temperature control area where the temperature prediction smoothing curve exceeds the allowable temperature change range, and calculate the temperature correction value based on the adjustment intensity coefficient.

[0311] Step S3113: Substitute the temperature correction value into the corresponding position of the initial temperature distribution matrix to adaptively fine-tune the temperature value and generate the second temperature distribution matrix.

[0312] Specifically, step S3110 involves adaptively adjusting the initial temperature matrix in place when the warning level is low, resulting in a second temperature distribution matrix. Compared to a full-area reset, this local fine-tuning method optimizes the temperature distribution with minimal cost. The correlation between the warning level and the adjustment intensity allows the control scheme to adaptively adjust the optimization intensity according to the degree of deviation. High-level warnings trigger significant adjustments to quickly alleviate the crisis; low-level warnings only require small adjustments, optimizing local areas while maintaining stability. After clarifying the adjustment area and intensity, the specific calculation of the temperature correction value needs to balance multiple objectives such as printing efficiency, energy consumption, and quality. The correction value must be sufficient to alleviate temperature anomalies while minimizing new temperature fluctuations caused by excessive adjustment. Substituting the local correction value into the initial temperature distribution matrix yields the second temperature distribution matrix after local optimization, while maintaining the overall distribution pattern. Compared to the initial temperature distribution matrix, the second temperature distribution matrix, while inheriting the existing reasonable layout, achieves dynamic self-optimization of the distribution pattern through local fine-tuning, enabling the temperature distribution scheme to continuously adjust in response to changes in the printing status.

[0313] Step S3120: If the temperature anomaly warning level is Level 3 or Level 4, generate a second temperature distribution matrix according to the preset safety mode and trigger an alarm to prompt manual inspection.

[0314] Further, step S3120 includes:

[0315] Step S3121: Based on the first comprehensive feature sequence, the temperature anomaly warning level, and the key process parameters of the current printing task, form a temperature distribution replanning information set;

[0316] Step S3122: Retrieve the preset safety mode temperature distribution scheme from the process database;

[0317] Step S3123: Input the temperature distribution replanning information set into the safety mode temperature distribution scheme, adjust the key temperature parameters in the safety mode temperature distribution scheme, and generate the adjusted safety mode temperature distribution scheme.

[0318] Step S3124: Discretize the adjusted safety mode temperature distribution scheme to form a second temperature distribution matrix.

[0319] The safety mode temperature distribution scheme is a conservative temperature distribution plan pre-designed and stored based on the physicochemical properties of the printing materials and the performance limits of the equipment. Its characteristics include a gentle temperature gradient, small temperature fluctuations, low thermal stress levels, and an overall lower temperature level. While this may reduce printing speed and efficiency, it is more conducive to ensuring the stability and safety of the printing process. The temperature distribution replanning information set is input into the safety mode temperature distribution scheme, adjusting key temperature parameters such as target temperature values ​​and temperature change rates in each temperature zone to adapt to the current specific printing environment. The adjustment range follows the principle of "better low than high," that is, reducing the temperature level as much as possible while maintaining safety and controllability. Compared with the initial temperature distribution matrix, the second temperature distribution matrix generated in this step generally lowers the target temperature values ​​of each actual temperature control zone, reduces the temperature difference between adjacent actual temperature control zones, and extends the time scale of temperature changes, thereby minimizing the risk of high-temperature runaway. Given the high warning level, the system's automatic adjustment may not be able to fully cope; experienced operators are required to analyze and judge based on the actual situation. Operators can focus on checking for malfunctions in hardware such as heaters and temperature sensors, ensuring proper supply and application of printing material, and verifying the shape and size of printed parts. If necessary, some temperature parameters in the safety mode matrix can be manually adjusted. After manually confirming the safety mode matrix's feasibility, it is output as the second temperature distribution matrix to guide subsequent temperature control. Simultaneously, the printing process continues to be monitored; if another warning occurs, the above steps are repeated; if the warning level decreases, the system can switch to automatic control mode.

[0320] In summary, generating a second temperature distribution matrix using a preset safety mode scheme when the warning level is high serves as a risk avoidance and emergency response mechanism. It quickly generates a conservative temperature control scheme, which, while potentially sacrificing some printing efficiency, minimizes and controls the risk of high-temperature runaway, preventing more serious quality incidents and equipment damage. Simultaneously, alarm prompts and manual intervention further enhance the reliability and flexibility of the response. This semi-automated handling mode, characterized by "machine-predicted plans and manual review and adjustment," fully leverages the advantages of human-machine collaboration, improving the system's rapid response capability and robustness in dealing with printing anomalies.

[0321] Step S3200: Real-time acquisition of the current temperature of each actual temperature control area to form a second temperature data matrix; element-wise subtraction operation between the second temperature data matrix and the second temperature distribution matrix to obtain the temperature deviation matrix between the two matrices; temperature regulation of the actual temperature control area based on the temperature deviation matrix.

[0322] Further, step S3200 includes:

[0323] Step S3210: Real-time acquisition of the current temperature of each actual temperature control zone to form a second temperature data matrix; element-by-element subtraction operation between the second temperature data matrix and the second temperature distribution matrix to calculate the temperature deviation value, and obtain the temperature deviation matrix between the two matrices.

[0324] Specifically, the temperature deviation matrix reveals the degree of deviation of the current temperature distribution from the predetermined scheme. A second temperature data matrix is ​​formed by real-time acquisition of current temperature readings from each actual temperature control zone using a temperature sensor array. The rationality of the sensor array layout and the sufficiency of the sampling frequency are crucial for obtaining high-quality temperature feedback data. By directly subtracting the feedback data from the target scheme, the deviation information becomes readily apparent. On one hand, the sign of the deviation indicates whether the controlled area is underheated or overheated; on the other hand, the magnitude of the deviation directly relates to the urgency of adjustment: the larger the deviation, the higher the timeliness requirement for adjustment. A comprehensive understanding of the deviation data for each zone makes the temperature control results more intuitive. Based on the deviation matrix, it is possible to determine whether each area needs adjustment, as well as the direction and intensity of the adjustment, thereby achieving selective and targeted control. In short, deviation calculation is the master switch for dynamic temperature feedback control, providing a quantifiable and operable breakthrough for intelligent control.

[0325] Step S3220: Judge the temperature deviation value of each actual temperature control area. If the absolute value of the temperature deviation value is less than the tolerance threshold, the temperature of the actual temperature control area is considered to be normal and temperature regulation is not triggered. Otherwise, calculate the temperature regulation value of the actual temperature control area and send it to the corresponding temperature control actuator to perform temperature regulation. The temperature regulation value is proportional to the magnitude of the temperature deviation value and has the opposite sign to the temperature deviation value.

[0326] Specifically, a tolerance threshold for temperature deviation is set. The threshold setting must balance the impact of temperature fluctuations on print quality and the energy cost of temperature control. Too small a threshold leads to excessively frequent adjustments, while too large a threshold causes deviations to accumulate to an uncontrollable level. Based on material properties and process requirements, a reasonable tolerance range is set to avoid unnecessary adjustments and improve control efficiency. Each element of the temperature deviation matrix is ​​iterated through, and its absolute value is compared to the tolerance threshold. If the absolute deviation value is less than the tolerance threshold, the actual temperature control area is considered normal and is not included in the adjustment plan to save computational and execution costs. If the absolute deviation value is greater than or equal to the tolerance threshold, the actual temperature control area is considered abnormal and requires adjustment. For each actual temperature control area requiring adjustment, its temperature deviation value is multiplied by a proportional coefficient to obtain the temperature control value. The magnitude of the proportional coefficient determines the adjustment strength, requiring a balance between rapid correction and avoiding overshoot. The sign of the control value is opposite to the deviation value to offset the deviation and bring the actual temperature back to the target value. The temperature control value corresponding to each actual temperature control area is sent to the corresponding temperature controller actuator. The actuator adjusts the radiant heating power accordingly based on the sign and magnitude of the adjustment value. Because the adjustment is timely and targeted, temperature deviations can be eliminated in the shortest possible time, ensuring temperature stability throughout the printing process. This automatic feedback control enables real-time, selective, and accurate temperature regulation. By continuously and dynamically suppressing new temperature deviations, it stabilizes the temperature distribution at the target level, ensuring a consistent temperature environment throughout the printing process. Thanks to timely feedback and rapid adjustment, this solution can control random temperature fluctuations within the tolerance range of the printing process, achieving precise control while simplifying manual monitoring, thus truly automating and intelligently controlling temperature.

[0327] Step S3300: After the temperature control is completed, evaluate the control effect;

[0328] Further, step S3300 includes:

[0329] Step S3310: Real-time acquisition of the second temperature time series after temperature control, the second thermal image time series of the entire powder bed surface, and the second production environment time series during the 3D printing process; extraction of the second temperature feature sequence of the second temperature time series, the second thermal image feature sequence of the second thermal image time series, and the second production environment feature sequence of the second production environment time series; and generation of the second comprehensive feature sequence.

[0330] Step S3320: Input the second comprehensive feature sequence into the optimal powder bed temperature sequence prediction model to generate the second temperature prediction sequence after temperature regulation;

[0331] Step S3330: Compare the second temperature prediction sequence with the first temperature prediction sequence to assess the difference in temperature change trends before and after temperature regulation; calculate the temperature distribution deviation value between the second temperature prediction sequence and the target temperature distribution matrix to assess the degree of closeness between the regulated temperature distribution and the ideal distribution.

[0332] Step S3340: Based on the differences in temperature change trends and temperature distribution deviations, obtain the comprehensive control effect evaluation score.

[0333] Methods for obtaining a comprehensive control effect evaluation score based on differences in temperature change trends and temperature distribution deviations include:

[0334] Differences in temperature change trends are measured by the standard deviation of temperature changes:

[0335] ;

[0336] The average temperature change for:

[0337] ;

[0338] in:

[0339] Standard deviation of temperature change;

[0340] : The actual temperature of the k-th actual temperature control zone;

[0341] The total number of actual temperature control zones.

[0342] Temperature distribution deviation can be calculated by measuring the temperature deviation in each actual temperature control zone. Sum of squares, and weighted summation:

[0343] ;

[0344] in:

[0345] : Temperature deviation of the k-th actual temperature control zone;

[0346] Temperature distribution deviation value;

[0347] : The weight of the k-th actual temperature control region;

[0348] The calculation methods for the comprehensive regulation effect evaluation score include:

[0349] ;

[0350] in:

[0351] : Comprehensive regulation effect evaluation score;

[0352] : The maximum allowable value for temperature deviation;

[0353] Smoothing factor: Used to adjust the sensitivity of temperature change trends;

[0354] Deviation factor: Used to adjust the sensitivity of temperature distribution deviation.

[0355] The formula uses standard deviation This measures the dispersion of temperature changes in different regions, reflecting the stability of the temperature field. It is achieved through a weighted sum of squared deviations. The deviation of each region from the target temperature is measured to reflect the accuracy of temperature control. A comprehensive control effect evaluation score S is used, combining differences in temperature change trends and temperature distribution deviations, to fully assess the temperature control effectiveness. Weight normalization and deviation normalization ensure the comparability and consistency of the evaluation scores.

[0356] Step S3350: Compare the comprehensive control effect evaluation score with the control effect threshold. If the comprehensive control effect evaluation score is greater than the control effect threshold, the control effect is considered to be up to standard, and the next temperature control cycle begins. Otherwise, return to step S3100, and re-regulate the temperature of each actual temperature control area based on the latest obtained second temperature data matrix and second temperature distribution matrix until the evaluation result meets the standard or the number of adjustments reaches the upper limit.

[0357] Specifically, the core of step S3300 is to objectively evaluate the temperature control effect through a closed-loop feedback mechanism and dynamically optimize the control strategy. By comparing the two temperature prediction sequences before and after adjustment, the impact of the adjustment on the temperature change trend can be intuitively judged: if the two sequences have similar shapes, it indicates that the adjustment has a limited impact; if the second temperature prediction sequence is more stable and the value is closer to the target, the adjustment is effective. In addition, by calculating the deviation between the second temperature prediction sequence and the ideal target temperature matrix, the degree to which the adjustment approximates the actual temperature distribution can be quantitatively evaluated: the smaller the deviation, the better the control effect. Taking into account both trend and deviation indicators, and using a weighted scoring mechanism, the control effect can be comprehensively quantified, and the control strategy can be optimized in a fine-grained manner; the evaluation results are used for automatic judgment and dynamic feedback to build a real-time closed-loop control system, and the temperature control can be dynamically adjusted according to the evaluation results, exhibiting adaptability; a scientifically reasonable scoring threshold and maximum number of adjustments are preset to prevent divergence in the adjustment process and to quickly converge to the ideal state within a limited time. This dynamic evaluation and feedback method makes temperature control more intelligent and refined, safeguarding high-quality and high-efficiency 3D printing production.

[0358] Example 2

[0359] This embodiment, based on Embodiment 1, provides a real-time monitoring and control system for powder bed temperature, such as... Figure 8 As shown, it includes:

[0360] Feature generation module: used to divide the powder bed into n1 actual temperature control zones, set target temperature values ​​and allowable deviations for each actual temperature control zone, form an initial temperature distribution matrix, and store the initial temperature distribution matrix into the process database; acquire the first temperature time series, thermal image time series, and production environment time series, and generate the first comprehensive feature sequence based on the first temperature time series, thermal image time series, and production environment time series; n1 is a positive integer;

[0361] The graded early warning module is used to extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database. Based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model, it obtains the temperature prediction smoothing curve for each actual temperature control area. Based on the optimal temperature distribution model, it obtains the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task. Based on the temperature prediction smoothing curve and the allowable temperature variation range, it provides graded early warnings for temperature anomalies.

[0362] Control module: Based on the temperature anomaly warning level, obtain the second temperature distribution matrix; collect the current temperature of each actual temperature control area in real time to form a second temperature data matrix; compare the second temperature data matrix with the second temperature distribution matrix to generate a temperature deviation matrix; adjust the temperature of each actual temperature control area according to the temperature deviation matrix; evaluate the control effect after the temperature control is completed.

[0363] In the feature generation module, acquiring the first temperature time series, the thermal image time series, and the production environment time series includes:

[0364] The temperature time series is formed by collecting the temperature change sequence of each actual temperature control area through a temperature sensor array; the thermal image time series is formed by collecting real-time thermal images of the entire powder bed surface at different time points through an infrared thermal imager; the production environment time series is formed by collecting the production environment data change sequence over time during the 3D printing process; the production environment time series includes the time series of n3 environmental parameters, including temperature, humidity, oxygen concentration and dust concentration.

[0365] In the feature generation module, generating the first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series includes:

[0366] Step S1310: Extract features from the first temperature time series to obtain a temperature feature series;

[0367] Step S1320: Extract features from the thermal image time series to obtain the thermal image feature sequence;

[0368] Step S1330: Extract features from the production environment time series to obtain the production environment feature sequence;

[0369] Step S1340: Align the temperature feature sequence, thermal image feature sequence, and production environment feature sequence in the time dimension to form the first comprehensive feature sequence.

[0370] Step S1310 includes:

[0371] Step S1311: Divide the first temperature time series into n2 time windows, each time window containing a fixed number of temperature sampling points; n2 is a positive integer;

[0372] Step S1312: Calculate the mean, variance, maximum and minimum temperature for each time window to generate a temperature feature vector;

[0373] Step S1313: Arrange the temperature feature vectors of each time window in chronological order to form a temperature feature sequence.

[0374] Step S1320 includes:

[0375] Step S1321: Divide each thermal image into sub-regions corresponding to n1 actual temperature control areas, and extract the grayscale histogram, gradient histogram and texture features of each sub-region to form an image feature vector.

[0376] Step S1322: Concatenate the image feature vectors of all sub-regions at the same time into a complete thermal image feature vector;

[0377] Step S1323: Arrange the feature vectors of the thermal images at all times in chronological order to obtain the thermal image feature sequence.

[0378] The feature extraction of the production environment time series includes a first level and a second level; the first level is to extract features from the time series of a single environmental parameter to obtain the feature sequence of the single environmental parameter; the second level is to combine the feature sequences of different environmental parameters.

[0379] In the graded early warning module, the optimal powder bed temperature sequence prediction model for the current printing task extracted from the process database includes:

[0380] Step S2111: Extract key process parameters from the process configuration file of the current printing task to form a process parameter tuple; the key process parameters include powder bed size, material type, layer thickness and printing speed.

[0381] Step S2112: Use the process parameter tuple to retrieve the optimal powder bed temperature sequence prediction model from the process database;

[0382] The method of using process parameter tuples to retrieve the optimal pulverized bed temperature sequence prediction model from the process database includes:

[0383] The process parameter tuples are compared item by item with the process parameter combinations corresponding to the pulverized bed temperature sequence prediction model in the process database. If all key process parameters are equal, they are considered to be a perfect match. The pulverized bed temperature sequence prediction model that is a perfect match is taken as the optimal pulverized bed temperature sequence prediction model.

[0384] In the graded early warning module, obtaining the temperature prediction smoothing curve for each actual temperature control area based on the first comprehensive feature sequence and the optimal powder bed temperature sequence prediction model includes:

[0385] Step S2120: Input the first comprehensive feature sequence into the optimal powder bed temperature sequence prediction model to generate the first temperature prediction sequence for each actual temperature control area in the future.

[0386] Step S2130: The first temperature prediction sequence is smoothed and denoised to eliminate high-frequency oscillations in the predicted values, and a smoothed temperature prediction curve for each actual temperature control region is obtained.

[0387] Step S2120 includes:

[0388] Step S2121: Match and align the first comprehensive feature sequence according to the input format required by the optimal powder bed temperature sequence prediction model;

[0389] Step S2122: Set the future prediction time length and sampling interval, and gradually predict the temperature prediction value at each future moment;

[0390] Step S2123: Arrange the temperature prediction values ​​in chronological order to form the first temperature prediction sequence.

[0391] In the graded early warning module, the allowable temperature variation range of each actual temperature control zone of the toner bed in the current printing task, obtained according to the optimal temperature distribution model, includes:

[0392] Step S2210: Based on the key process parameters of the current printing task, retrieve the optimal temperature distribution model from the process database; the optimal temperature distribution model includes the division of temperature control areas and the target temperature value for each temperature control area.

[0393] Step S2220: Combining the powder bed size and the actual temperature control area, the optimal temperature distribution model is adaptively modified to generate the target temperature distribution matrix;

[0394] Step S2230: Calculate the upper and lower thresholds of the target temperature distribution matrix to obtain the allowable temperature variation range for each actual temperature control zone.

[0395] Step S2220 includes:

[0396] Step S2221: Align the temperature control region of the optimal temperature distribution model with the actual temperature control region of the powder bed. If the particle size of the temperature control region of the optimal temperature distribution model is inconsistent with that of the actual temperature control region, then perform interpolation or merging processing on the optimal temperature distribution model to make the temperature control region of the optimal temperature distribution model match the actual temperature control region.

[0397] Step S2222: Perform a geometric transformation on the optimal temperature distribution model based on the powder bed size to obtain an optimal temperature distribution transformation model that is adapted to the powder bed size.

[0398] Step S2223: Fine-tune the target temperature value in the optimal temperature distribution transformation model to generate the adjusted optimal temperature distribution transformation model;

[0399] Step S2224: Discretize the adjusted optimal temperature distribution transformation model to generate a target temperature distribution matrix corresponding to the actual temperature control area.

[0400] In the graded early warning module, the step of providing graded early warnings for temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range includes:

[0401] In step S2300, the temperature prediction smoothing curve is compared with the allowable temperature change range to determine whether a temperature anomaly occurs within the prediction time. If there is no temperature anomaly, the process returns to step S1200 to continue temperature detection and real-time data acquisition. If a temperature anomaly occurs, the over-limit amplitude and duration of the predicted temperature value in the temperature prediction smoothing curve are calculated, and a graded warning for the temperature anomaly is generated based on the over-limit amplitude and duration.

[0402] The temperature anomaly warning levels include: Level 1 warning, Level 2 warning, Level 3 warning, and Level 4 warning;

[0403] Level 1 warning: Exceeding limit amplitude AM < 5%; Duration DU < 1 min;

[0404] Level II Warning: 5% ≤ Exceedance Amplitude AM < 10%; 1 min ≤ Duration DU < 5 min;

[0405] Level 3 warning: 10% ≤ Exceedance amplitude AM < 20%; 5min ≤ Duration DU < 10min;

[0406] Level 4 warning: Exceeding the limit AM ≥ 10%; Duration DU ≥ 10 min.

[0407] In the control module, obtaining the second temperature distribution matrix based on the temperature anomaly warning level includes:

[0408] Step S3110: If the temperature anomaly warning level is Level 1 or Level 2, then the initial temperature distribution matrix is ​​adaptively fine-tuned according to the temperature anomaly warning level to generate a second temperature distribution matrix.

[0409] Step S3120: If the temperature anomaly warning level is Level 3 or Level 4, then generate the second temperature distribution matrix according to the preset safety mode.

[0410] Step S3110 includes:

[0411] Step S3111: Map the Level 1 warning and Level 2 warning to adjustment intensity coefficients, where the adjustment intensity coefficient for Level 2 warning is greater than that for Level 1 warning;

[0412] Step S3112: Identify the actual temperature control area where the temperature prediction smoothing curve exceeds the allowable temperature change range, and calculate the temperature correction value based on the adjustment intensity coefficient.

[0413] Step S3113: Substitute the temperature correction value into the corresponding position of the initial temperature distribution matrix to adaptively fine-tune the temperature value and generate the second temperature distribution matrix.

[0414] Step S3120 includes:

[0415] Step S3121: Based on the first comprehensive feature sequence, the temperature anomaly warning level, and the key process parameters of the current printing task, form a temperature distribution replanning information set;

[0416] Step S3122: Retrieve the preset safety mode temperature distribution scheme from the process database;

[0417] Step S3123: Input the temperature distribution replanning information set into the safety mode temperature distribution scheme, adjust the key temperature parameters in the safety mode temperature distribution scheme, and generate the adjusted safety mode temperature distribution scheme.

[0418] Step S3124: Discretize the adjusted safety mode temperature distribution scheme to form a second temperature distribution matrix.

[0419] In the control module, the step of adjusting the temperature of each actual temperature control zone according to the temperature deviation matrix includes:

[0420] Step S3210: Real-time acquisition of the current temperature of each actual temperature control zone to form a second temperature data matrix; element-by-element subtraction operation between the second temperature data matrix and the second temperature distribution matrix to calculate the temperature deviation value, and obtain the temperature deviation matrix between the two matrices.

[0421] Step S3220: Judge the temperature deviation value of each actual temperature control area. If the absolute value of the temperature deviation value is less than the tolerance threshold, the temperature of the actual temperature control area is considered to be normal and temperature regulation is not triggered. Otherwise, calculate the temperature regulation value of the actual temperature control area and send it to the corresponding temperature control actuator to perform temperature regulation. The temperature regulation value is proportional to the magnitude of the temperature deviation value and has the opposite sign to the temperature deviation value.

[0422] In the control module, the evaluation of the control effect includes:

[0423] Step S3310: Real-time acquisition of the second temperature time series after temperature control, the second thermal image time series of the entire powder bed surface, and the second production environment time series during the 3D printing process; extraction of the second temperature feature sequence of the second temperature time series, the second thermal image feature sequence of the second thermal image time series, and the second production environment feature sequence of the second production environment time series; and generation of the second comprehensive feature sequence.

[0424] Step S3320: Input the second comprehensive feature sequence into the optimal powder bed temperature sequence prediction model to generate the second temperature prediction sequence after temperature regulation;

[0425] Step S3330: Compare the second temperature prediction sequence with the first temperature prediction sequence to assess the difference in temperature change trends before and after temperature regulation; calculate the temperature distribution deviation value between the second temperature prediction sequence and the target temperature distribution matrix to assess the degree of closeness between the regulated temperature distribution and the ideal distribution.

[0426] Step S3340: Based on the differences in temperature change trends and temperature distribution deviations, obtain the comprehensive control effect evaluation score.

[0427] Example 3

[0428] This embodiment discloses an electronic device that may include one or more processors and one or more memories. The memories store computer-readable code, which, when executed by the one or more processors, can perform the real-time monitoring and control method for powder bed temperature as described above.

[0429] The method or system according to the embodiments of this application can also be implemented using the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, read-only memory (ROM), random access memory (RAM), a communication port connected to a network, input / output components, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store the real-time monitoring and control method for powder bed temperature provided in this application. The real-time monitoring and control method for powder bed temperature may, for example, include: dividing the powder bed into n1 actual temperature control zones, setting a target temperature value and allowable deviation for each actual temperature control zone, forming an initial temperature distribution matrix, and storing the initial temperature distribution matrix in a process database; acquiring a first temperature time series, a thermal image time series, and a production environment time series, and generating a first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series; n1 is a positive integer; extracting the optimal powder bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database, and generating a first comprehensive feature sequence based on the first comprehensive feature sequence and the optimal powder bed temperature sequence prediction model. The system obtains the temperature prediction smoothing curve for each actual temperature control area; obtains the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task based on the optimal temperature distribution model; classifies and warns of temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range; obtains a second temperature distribution matrix based on the temperature anomaly warning level; collects the current temperature of each actual temperature control area in real time to form a second temperature data matrix; compares the second temperature data matrix with the second temperature distribution matrix to generate a temperature deviation matrix; adjusts the temperature of each actual temperature control area based on the temperature deviation matrix; and evaluates the adjustment effect after the temperature adjustment is completed.

[0430] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in this invention is merely exemplary; in implementing different devices, one or more components of the electronic device disclosed in this invention may be omitted according to actual needs.

[0431] Example 4

[0432] This embodiment discloses a computer-readable storage medium storing computer-readable instructions. When the computer-readable instructions are executed by a processor, the real-time monitoring and control method for powder bed temperature according to the embodiments of this application can be performed. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0433] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowchart can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as: dividing the powder bed into n1 actual temperature control regions, setting a target temperature value and allowable deviation for each actual temperature control region to form an initial temperature distribution matrix, and storing the initial temperature distribution matrix in a process database; acquiring a first temperature time series, a thermal image time series, and a production environment time series, and generating a first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series; n1 being a positive integer; and extracting the optimal powder bed temperature sequence prediction for the current printing task from the process database. The system employs a model and an optimal temperature distribution model to obtain a temperature prediction smoothing curve for each actual temperature control area based on a first comprehensive feature sequence and an optimal powder bed temperature sequence prediction model. It also obtains the allowable temperature variation range for each actual temperature control area of ​​the powder bed in the current printing task based on the optimal temperature distribution model. Furthermore, it provides graded early warnings for temperature anomalies based on the temperature prediction smoothing curves and the allowable temperature variation ranges. A second temperature distribution matrix is ​​obtained based on the temperature anomaly warning level. The system collects the current temperature of each actual temperature control area in real time to form a second temperature data matrix. This second temperature data matrix is ​​compared with the second temperature distribution matrix to generate a temperature deviation matrix. Temperature regulation is then applied to each actual temperature control area based on the temperature deviation matrix. Finally, the regulation effect is evaluated after the temperature regulation is completed. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0434] The methods, systems, and apparatus of this application may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the method is for illustrative purposes only, and the steps of the method of this application are not limited to the order specifically described above, unless otherwise specifically stated. Furthermore, in some embodiments, this application may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the method according to this application. Thus, this application also covers recording media storing programs for performing the method according to this application.

[0435] In addition, the parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of the corresponding technical solutions in the prior art have not been described in detail, so as to avoid excessive elaboration.

[0436] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring and control of powder bed temperature, characterized in that, The method includes: The powder bed is divided into n1 actual temperature control zones. A target temperature value and allowable deviation are set for each actual temperature control zone to form an initial temperature distribution matrix. The initial temperature distribution matrix is ​​stored in the process database. The first temperature time series, thermal image time series and production environment time series are obtained. Based on the first temperature time series, thermal image time series and production environment time series, a first comprehensive feature sequence is generated. n1 is a positive integer. Extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database. Obtain the temperature prediction smoothing curve for each actual temperature control area based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model. Obtain the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task based on the optimal temperature distribution model. Perform graded early warning for temperature anomalies based on the temperature prediction smoothing curve and the allowable temperature variation range. Based on the temperature anomaly warning level, a second temperature distribution matrix is ​​obtained; the current temperature of each actual temperature control area is collected in real time to form a second temperature data matrix; the second temperature data matrix is ​​compared with the second temperature distribution matrix to generate a temperature deviation matrix; the temperature of each actual temperature control area is adjusted according to the temperature deviation matrix; after the temperature adjustment is completed, the adjustment effect is evaluated.

2. The method for real-time monitoring and control of powder bed temperature according to claim 1, characterized in that, The acquisition of the first temperature time series, the thermal image time series, and the production environment time series includes: The temperature time series is formed by collecting the temperature change sequence of each actual temperature control area through a temperature sensor array; the real-time thermal images of the entire powder bed surface at different time points are collected by an infrared thermal imager to form a thermal image time series; the production environment data change sequence during the 3D printing process is collected to form a production environment time series; the production environment time series includes the time series of n3 environmental parameters, including temperature, humidity, oxygen concentration and dust concentration.

3. The method for real-time monitoring and control of powder bed temperature according to claim 1, characterized in that, The step of generating the first comprehensive feature sequence based on the first temperature time series, the thermal image time series, and the production environment time series includes: Feature extraction is performed on the first temperature time series to obtain the temperature feature series; Feature extraction is performed on the time series of thermal images to obtain the feature sequence of thermal images; Feature extraction is performed on the time series of the production environment to obtain the production environment feature sequence; The temperature feature sequence, thermal image feature sequence, and production environment feature sequence are aligned along the time dimension to form the first comprehensive feature sequence. The feature extraction of the first temperature time series includes: The first temperature time series is divided into n² time windows, each containing a fixed number of temperature sampling points; n² is a positive integer. For each time window, calculate its mean, variance, maximum, and minimum temperature values ​​to generate a temperature feature vector. The temperature feature vectors of each time window are arranged in chronological order to form a temperature feature sequence.

4. The method for real-time monitoring and control of powder bed temperature according to claim 3, characterized in that, The feature extraction of the thermal image time series includes: Each thermal image is divided into sub-regions corresponding to n1 actual temperature control areas. The grayscale histogram, gradient histogram and texture features of each sub-region are extracted to form an image feature vector. The image feature vectors of all sub-regions at the same time are concatenated into a complete thermal image feature vector; the thermal image feature vectors at all times are arranged in chronological order to obtain the thermal image feature sequence. The feature extraction of the production environment time series includes a first level and a second level; the first level is to extract features from the time series of a single environmental parameter to obtain the feature sequence of the single environmental parameter; the second level is to combine the feature sequences of different environmental parameters.

5. The method for real-time monitoring and control of powder bed temperature according to claim 1, characterized in that, The optimal powder bed temperature sequence prediction model for the current printing task extracted from the process database includes: Extract key process parameters from the process configuration file of the current printing task to form a process parameter tuple; the key process parameters include powder bed size, material type, layer thickness and printing speed; The process parameter tuples are compared item by item with the process parameter combinations corresponding to the pulverized bed temperature sequence prediction model in the process database. If all key process parameters are equal, they are considered to be a perfect match. The pulverized bed temperature sequence prediction model that is a perfect match is taken as the optimal pulverized bed temperature sequence prediction model. If no perfectly matching pulverized bed temperature sequence prediction model is found, the similarity between the process parameter tuple and the process parameter combination corresponding to the pulverized bed temperature sequence prediction model in the process database is calculated, and the one with the highest similarity is selected as the optimal pulverized bed temperature sequence prediction model.

6. The method for real-time monitoring and control of powder bed temperature according to claim 5, characterized in that, The allowable temperature variation range of each actual temperature control zone of the toner bed in the current printing task, obtained according to the optimal temperature distribution model, includes: By combining the powder bed size and the actual temperature control area, the optimal temperature distribution model is adaptively modified to generate the target temperature distribution matrix; the upper and lower thresholds of the target temperature distribution matrix are calculated to obtain the allowable temperature variation range of each actual temperature control area.

7. The method for real-time monitoring and control of powder bed temperature according to claim 6, characterized in that, The optimal temperature distribution model includes the division of temperature control zones and the target temperature value for each temperature control zone; The adaptive modification of the optimal temperature distribution model to generate the target temperature distribution matrix includes: Align the temperature control region of the optimal temperature distribution model with the actual temperature control region of the powder bed. If the particle size of the temperature control region of the optimal temperature distribution model is inconsistent with that of the actual temperature control region, then perform interpolation or merging processing on the optimal temperature distribution model to match the temperature control region of the optimal temperature distribution model with that of the actual temperature control region. Geometric transformation of the optimal temperature distribution model is performed based on the powder bed size to obtain an optimal temperature distribution transformation model adapted to the powder bed size; Fine-tune the target temperature value in the optimal temperature distribution transformation model to generate the adjusted optimal temperature distribution transformation model; The adjusted optimal temperature distribution transformation model is discretized to generate a target temperature distribution matrix corresponding to the actual temperature control area.

8. The method for real-time monitoring and control of powder bed temperature according to claim 6, characterized in that, The method of classifying and issuing early warnings for temperature anomalies based on temperature prediction smoothing curves and allowable temperature variation ranges includes: The temperature prediction smoothing curve is compared with the allowable temperature change range to determine whether a temperature anomaly will occur within the prediction time. If a temperature anomaly occurs, the magnitude and duration of the temperature exceeding the limit in the temperature prediction smoothing curve are calculated. Based on the magnitude and duration of the exceeding the limit, a graded warning for the temperature anomaly is issued, generating a temperature anomaly warning level. The temperature anomaly warning levels include: Level 1 warning, Level 2 warning, Level 3 warning, and Level 4 warning; Level 1 warning: Exceeding limit amplitude AM < 5%; Duration DU < 1 min; Level II Warning: 5% ≤ Exceedance Amount AM < 10%; 1 min ≤ Duration DU < 5 min; Level 3 warning: 10% ≤ Exceedance amplitude AM < 20%; 5min ≤ Duration DU < 10min; Level 4 warning: Exceeding the limit by AM ≥ 10%; Duration by DU ≥ 10 min.

9. The method for real-time monitoring and control of powder bed temperature according to claim 8, characterized in that, The evaluation of the control effect includes: Real-time acquisition of the second temperature time series after temperature control, the second thermal image time series of the entire powder bed surface, and the second production environment time series during the 3D printing process; extraction of the second temperature feature sequence of the second temperature time series, the second thermal image feature sequence of the second thermal image time series, and the second production environment feature sequence of the second production environment time series; and generation of the second comprehensive feature sequence. The second comprehensive feature sequence is input into the optimal powder bed temperature sequence prediction model to generate the second temperature prediction sequence after temperature regulation. Compare the second temperature prediction sequence with the first temperature prediction sequence to assess the difference in temperature change trends before and after temperature regulation; calculate the temperature distribution deviation between the second temperature prediction sequence and the target temperature distribution matrix. The comprehensive control effect evaluation score is obtained based on the differences in temperature change trends and temperature distribution deviations.

10. A real-time monitoring and control system for powder bed temperature, used to implement the real-time monitoring and control method for powder bed temperature as described in any one of claims 1-9, characterized in that, The system includes: Feature generation module: used to divide the powder bed into n1 actual temperature control zones, set target temperature values ​​and allowable deviations for each actual temperature control zone, form an initial temperature distribution matrix, and store the initial temperature distribution matrix into the process database; acquire the first temperature time series, thermal image time series, and production environment time series, and generate the first comprehensive feature sequence based on the first temperature time series, thermal image time series, and production environment time series; n1 is a positive integer; The graded early warning module is used to extract the optimal toner bed temperature sequence prediction model and the optimal temperature distribution model for the current printing task from the process database. Based on the first comprehensive feature sequence and the optimal toner bed temperature sequence prediction model, it obtains the temperature prediction smoothing curve for each actual temperature control area. Based on the optimal temperature distribution model, it obtains the allowable temperature variation range for each actual temperature control area of ​​the toner bed in the current printing task. Based on the temperature prediction smoothing curve and the allowable temperature variation range, it provides graded early warnings for temperature anomalies. Control module: Based on the temperature anomaly warning level, obtain the second temperature distribution matrix; collect the current temperature of each actual temperature control area in real time to form a second temperature data matrix; compare the second temperature data matrix with the second temperature distribution matrix to generate a temperature deviation matrix; adjust the temperature of each actual temperature control area according to the temperature deviation matrix; evaluate the control effect after the temperature control is completed.

Images