A metal 3D printing forming accuracy monitoring method, device and electronic equipment thereof

Abstract

The present invention discloses a metal 3D printing molding accuracy monitoring method, device and electronic equipment thereof, which relates to the field of metal 3D printing technology, and includes the following steps: obtaining multimodal sensor data, collecting multidimensional data in the metal 3D printing process based on the multimodal sensor data, determining that the optical sensor records the molten pool morphology and powder distribution, determining that the thermal infrared camera monitors the temperature field distribution, determining that the acoustic sensor collects the printing medium and high frequency sound wave signal, and identifies the defect signal; according to the edge computing device, the edge computing device pre-processes the data locally, extracts the boundary and geometric features from the optical image based on the convolutional neural network, and extracts the key temperature change area in response to the thermal field gradient analysis; extracts the spectrum features from the acoustic data based on the time-frequency analysis. The present invention realizes data fusion and defect detection by synchronously collecting multimodal data such as optics, thermal and acoustics, and using advanced deep learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of metal 3D printing, and in particular to a metal 3D printing molding accuracy monitoring method, device and electronic equipment thereof. Background Art

[0002] Metal 3D printing is a method of creating complex metal parts and components by accumulating metal materials layer by layer. The principle is to use computer-aided design software to decompose the metal product model to be manufactured into layers of slices, and then heat the metal powder material to a high temperature and melt it through the laser beam, arc, electron beam or inkjet inside the metal 3D printer. After the metal powder material is melted, the 3D printer accurately superimposes the molten metal powder material on top of the previous layer by controlling the movement trajectory of the mobile platform, nozzle or laser beam according to the instructions of the CAD model. In this way, layers of metal powder gradually melt and accumulate, and finally form the desired metal product.

[0003] The technology is used to detect and evaluate the geometric shape, size, surface quality and other molding accuracy indicators of the workpiece during or after the metal 3D printing process. Its purpose is to ensure that the quality of the printed part meets the design requirements, optimize the printing process, and improve production efficiency.

[0004] Ensure print quality: Verify that prints meet design specifications, such as dimensional tolerance, shape accuracy, surface roughness, etc., through accuracy monitoring.

[0005] Process control: Real-time monitoring and feedback during the printing process to detect problems and adjust parameters in time.

[0006] Defect prevention and detection: Identify printing defects (such as deformation, cracking, porosity) and analyze their causes.

[0007] Data-driven optimization: By analyzing precision data, optimizing process parameters, and improving printing efficiency and stability, metal 3D printing molding precision monitoring technology is an important support in the field of advanced manufacturing. Its continuous development will further promote the popularization and maturity of metal 3D printing technology.

[0008] The main problems of existing metal 3D printing molding accuracy monitoring technology:

[0009] Lack of real-time performance: Current online monitoring technologies are usually unable to process complex data (such as melt pool dynamics, powder deposition status, etc.) in real time, resulting in delayed feedback of printing defects, which in turn affects control accuracy.

[0010] Lack of multimodal fusion monitoring: Existing technologies usually rely on only a single monitoring method (such as optical monitoring or thermal monitoring), which makes it difficult to fully capture the complex multi-physical field phenomena (heat, stress, deformation, etc.) in the metal 3D printing process.

[0011] Limited data processing capabilities: Massive amounts of data are generated during the printing process. Existing data collection and processing methods cannot efficiently extract useful information, and lack advanced data analysis and prediction models.

[0012] Error compensation lag: Printing error compensation is mostly done offline rather than online closed-loop control, and it is impossible to adjust process parameters in time during printing to optimize molding quality.

[0013] Online detection of internal defects is difficult: Current technologies still face challenges in online detection of internal defects (such as pores and cracks) in printed parts, and mostly rely on offline detection methods, which limits the timeliness of process optimization.

[0014] Interference from environmental influences: There are high temperatures, dust, vibrations, etc. in the metal 3D printing environment. These conditions will affect the performance of the sensor and reduce the accuracy and stability of monitoring;

[0015] In the process of metal 3D printing, the monitoring of molding accuracy is a key link to ensure the printing quality. At present, optical sensors are often used to monitor the size of parts. By real-time monitoring of dimensional defects and anomalies during the printing process, it is determined whether the part size meets the requirements. However, the existing detection devices have certain limitations and cannot perform all-round accuracy monitoring of parts.

[0016] To this end, a metal 3D printing forming accuracy monitoring method, device and electronic equipment are proposed. Summary of the invention

[0017] The purpose of the present invention is to provide a metal 3D printing molding accuracy monitoring method, device and electronic equipment thereof, so as to solve the problem of building an efficient data processing and real-time control system by integrating multimodal monitoring data (optical, thermal, acoustic, etc.) and deep learning technology. The printing parameters are optimized by using a reinforcement learning algorithm to achieve dynamic closed-loop control of the printing process, ensuring that the molding accuracy is significantly improved, so as to solve one of the problems raised in the above background technology.

[0018] First aspect: To solve the above technical problems, a technical solution adopted in this application is: a method for monitoring metal 3D printing forming accuracy, comprising the following steps:

[0019] Acquire multimodal sensor data. Based on the multimodal sensor data, collect multi-dimensional data in the metal 3D printing process, determine the optical sensor to record the molten pool morphology and powder distribution, determine the thermal infrared camera to monitor the temperature field distribution, determine the acoustic sensor to collect high-frequency sound wave signals in printing, and identify defect signals;

[0020] According to the local preprocessing data of the edge computing device, the boundary and geometric features are extracted from the optical image based on the convolutional neural network, and the key temperature change area is extracted in response to the thermal field gradient analysis; the spectrum features are extracted from the acoustic data based on the time-frequency analysis;

[0021] Construct a multimodal convolutional neural network, input optical, thermal and acoustic data into the model for feature fusion and classification, analyze the dynamic changes in the printing process based on the Transformer time series prediction model, and simulate future printing quality;

[0022] According to the analysis results, the printing parameters are adjusted dynamically. The printing parameters are dynamically optimized through environmental feedback based on the reinforcement learning algorithm. The printing parameters are connected to the printing device based on the printing parameters, and the printing parameters are adjusted in real time to achieve closed-loop control.

[0023] Build a web-based visualization interface to display key indicators and optimize process parameters through closed-loop control.

[0024] As a further preferred embodiment of the present technical solution, the method further comprises:

[0025] Based on the holographic microscope, the 3D structural changes of the printed melt pool are collected in real time to analyze the surface defects and depth information during printing;

[0026] According to the holographic microscope and the multi-dimensional data collection, the data collection method is adjusted to determine the interference pattern; based on the Fourier transform algorithm, the interference pattern is converted into a visualized three-dimensional structure, and the surface roughness and crack depth are calculated;

[0027] Use deep convolutional neural networks to clean up noise and artifacts in holographic data, and determine the spatiotemporal encoder to achieve time series alignment and embedding conversion of inter-modal features;

[0028] Build a multi-task deep learning model for multi-dimensional data, map the features of multi-dimensional data to quantum states, and realize feature representation in high-dimensional space through quantum circuits;

[0029] Build a digital twin model of the printing equipment and process, simulate the physical and chemical behavior of the printing process in real time, and dynamically update the status of the digital twin model based on monitoring data and historical databases.

[0030] As a further preferred embodiment of the present technical solution, the multi-dimensional data includes the molten pool temperature field, the powder deposition state and the laser reflection signal. The formula of the multi-dimensional data is expressed as follows: :

[0031]

[0032] in, , and They are optical, thermal and acoustic data respectively;

[0033] The output of the multimodal convolutional neural network F is the defect prediction result :

[0034] .

[0035] As a further preferred embodiment of the present technical solution, the dynamically adjusted printing parameters include laser power, scanning speed and powder supply amount, and the formula of the reinforcement learning algorithm is expressed as follows: The optimal strategy is determined by maximizing the expected reward, as follows:

[0036] ;

[0037] in, is the optimal strategy, It is a mathematical operation, which means finding the parameter that makes the following expression take the maximum value. ; is the mathematical expectation, is an infinite series, representing the discounted sum of all future rewards starting from time t=0, is the discount factor, is the reward obtained at time t.

[0038] As a further preferred embodiment of the present technical solution, the optimized process parameters include molten pool shape, temperature field distribution and defect probability.

[0039] Second aspect: To solve the above technical problems, another technical solution adopted by the present application is: a metal 3D printing molding accuracy monitoring device, further comprising a part molding cover, wherein a monitoring component is installed inside the part molding cover, and the monitoring component comprises a lifting seat, two limit plates, two limit rails, two sliders, a tooth plate, a mounting seat, a laser ranging sensor, a slide slot, a driving gear and a servo motor;

[0040] The two limit rails are symmetrically opened on the upper surface and the lower surface of the lifting seat, the two sliders are symmetrically fixedly connected to the two ends of the lifting seat, the slide groove is opened inside the lifting seat, the tooth plate is fixedly connected to the inner bottom wall of the slide groove, the driving gear is coaxially fixedly connected to the output shaft of the servo motor, one side of the two limit plates is symmetrically fixedly connected to the upper surface and the lower surface of the mounting seat, the laser ranging sensor is installed on the front surface of the mounting seat, a display screen is installed on one side of the part forming cover, the other side of the two limit plates away from the mounting seat is slidably connected to the two limit rails one by one, the driving gear is located inside the slide groove, and the driving gear is meshed with the tooth plate.

[0041] As a further preferred embodiment of the present technical solution: the servo motor is installed inside the mounting seat, and the monitoring assembly further includes a drive motor, a screw rod, two guide rods and two guide seats;

[0042] The two guide seats are symmetrically installed at two adjacent corners in the part forming cover, the top end of the screw rod is coaxially fixedly connected to the output shaft of the driving motor, and the two ends of the two guide rods are symmetrically fixedly connected to the inner top wall and the inner bottom wall of the guide seat.

[0043] As a further preferred embodiment of the present technical solution: the bottom end of the screw rod is rotatably connected to the inner bottom wall of the guide seat, the slider is threadedly connected to the outer side wall of the screw rod, the slider is slidably connected to the outer side walls of the two guide rods, the slider is slidably connected to the inside of the guide seat, and the drive motor is installed on the upper surface of the guide seat.

[0044] The third aspect: In order to solve the above-mentioned technical problems, another technical solution adopted in this application is: an electronic device, comprising a processor, a memory and a communication interface, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of any of the above-mentioned metal 3D printing forming accuracy monitoring methods are implemented.

[0045] Fourth aspect: In order to solve the above-mentioned technical problems, another technical solution adopted in this application is: a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a metal 3D printing forming accuracy monitoring method as described in any one of the above are implemented.

[0046] Advantages of the present invention:

[0047] 1. The present invention realizes data fusion and defect detection by synchronously collecting multimodal data such as optics, thermal and acoustics and using advanced deep learning technology;

[0048] 2. The present invention adopts holographic imaging to provide three-dimensional structural information of the microscopic morphology of the printing process, which is a major supplement to the current mainstream monitoring technology. It realizes high-dimensional feature fusion optimization and has forward-looking advantages in complex data processing. Through real-time virtual simulation and control, the monitoring technology is deeply integrated with intelligent manufacturing;

[0049] 3. The monitoring component of the present invention can realize the precision monitoring of parts through the laser ranging sensor, and the monitoring data are displayed on the display screen. The printing error can be calculated by referring to the monitoring data and the designed printing data. The servo motor drives the driving gear, and the driving gear meshes with the gear plate through the teeth. The servo motor drives the mounting seat, which can make the mounting seat produce horizontal displacement, and then monitor the molding accuracy of each horizontal position of the part. The driving motor drives the lead screw, and the lead screw drives the slider through the thread. The lifting seat drives the laser ranging sensor through the mounting seat, which can make the laser ranging sensor produce vertical displacement, and then realize the all-round monitoring of the precision of the molded parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A schematic diagram of a process of monitoring metal 3D printing forming accuracy according to the present invention;

[0052] Figure 2 Schematic diagram of the process of the metal 3D printing molding accuracy monitoring method in the third embodiment of the present invention;

[0053] Figure 3 It is a structural schematic diagram of an electronic device of the present invention;

[0054] Figure 4 It is a structural schematic diagram of the present invention;

[0055] Figure 5 It is a schematic diagram of the structure of the monitoring component of the present invention;

[0056] Figure 6 It is a schematic diagram of the connection between the lifting seat and the limiting plate of the present invention;

[0057] Figure 7 It is a schematic diagram of the structure of the limiting plate of the present invention;

[0058] Figure 8 It is a schematic diagram of the lifting seat structure of the present invention.

[0059] In the figure: 101, monitoring component; 11, lifting seat; 12, limit plate; 13, limit track; 14, slider; 15, tooth plate; 16, mounting seat; 17, laser ranging sensor; 18, slide groove; 19, driving gear; 21, servo motor; 22, driving motor; 23, screw rod; 24, guide rod; 25, guide seat; 26, display screen; 30, part forming cover. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Figure 1 is a flow chart of a method for monitoring metal 3D printing forming accuracy according to an embodiment of the present invention. It should be noted that if there are substantially the same results, the method of the present application is not limited to Figure 1 The process sequence shown is limited.

[0062] Embodiment 1

[0063] like Figure 1 A method for monitoring metal 3D printing forming accuracy is shown, comprising the following steps:

[0064] A method for monitoring metal 3D printing forming accuracy comprises the following steps:

[0065] S10, acquiring multimodal sensor data, collecting multi-dimensional data in the metal 3D printing process based on the multimodal sensor data, determining that the optical sensor records the molten pool morphology and powder distribution, determining that the thermal infrared camera monitors the temperature field distribution, determining that the acoustic sensor collects the high-frequency sound wave signals in the printing, and identifying the defect signals;

[0066] Specifically, the integrated design of multimodal sensors provides comprehensive process data and makes up for the shortcomings of a single monitoring method.

[0067] S20, extracting boundaries and geometric features from the optical image based on a convolutional neural network according to the local preprocessing data of the edge computing device, extracting key temperature change areas in response to thermal field gradient analysis; extracting spectral features from the acoustic data based on time-frequency analysis;

[0068] Use edge computing devices to process pre-processed data locally to reduce transmission delays; use high-bandwidth communication protocols (such as 5G or Wi-Fi6) to support high-speed data transmission; use the low-latency protobuf protocol as the data transmission format to solve the problem of data transmission delay and achieve real-time data processing;

[0069] S30, build a multimodal convolutional neural network, input optical, thermal and acoustic data into the model for feature fusion and classification, analyze the dynamic change results in the printing process based on the Transformer time series prediction model, and simulate future printing quality;

[0070] The multimodal deep learning method is adopted to significantly improve the accuracy and real-time performance of data processing. The model is run in real time on GPU acceleration hardware to ensure millisecond-level inference speed, extract abnormal features from massive monitoring data in real time, and predict potential defects.

[0071] S40. According to the analysis results, the printing parameters are adjusted dynamically. The printing parameters are dynamically optimized through environmental feedback based on the reinforcement learning algorithm. The printing device is connected based on the printing parameters, and the printing parameters are adjusted in real time to achieve closed-loop control.

[0072] Reinforcement learning algorithms (such as Proximal Policy Optimization, PPO) are used to dynamically optimize printing strategies through environmental feedback. The control system is connected to the printing device to adjust printing parameters in real time to achieve closed-loop control. According to the analysis results, the printing parameters (such as laser power, scanning speed, and powder supply) are dynamically adjusted. The dynamic optimization capabilities of reinforcement learning are used to significantly improve printing accuracy and stability.

[0073] S50. Build a Web-based visualization interface to display key indicators and optimize process parameters through closed-loop control.

[0074] Provide engineers with real-time feedback and data visualization interface, and store data in the cloud for long-term use in subsequent analysis and process optimization.

[0075] Specifically, the multi-dimensional data includes the molten pool temperature field, the powder deposition state and the laser reflection signal. The formula of the multi-dimensional data is expressed as follows: :

[0076]

[0077] in, , and They are optical, thermal and acoustic data respectively;

[0078] The output of the multimodal convolutional neural network F is the defect prediction result :

[0079] .

[0080] in, Represents the optical opt data Perform convolutional neural network (CNN) processing;

[0081] Indicates thermal imaging therm data Perform convolutional neural network processing; Represents acoustic acou data Perform convolutional neural network processing.

[0082] The concat function connects the above three different modal data after CNN processing according to the feature dimension;

[0083] The concatenated features are further processed by MLP (Multi-layer Perceptron) to integrate information from different modalities;

[0084] The softmax function converts the output of MLP into a probability distribution to obtain the final prediction result. .

[0085] Specifically, the dynamically adjusted printing parameters include laser power, scanning speed and powder supply amount. The formula of the reinforcement learning algorithm is expressed as: The optimal strategy is determined by maximizing the expected reward, as follows:

[0086] ;

[0087] in, is the optimal strategy, It is a mathematical operation, which means finding the parameter that makes the following expression take the maximum value. ; is the mathematical expectation, is an infinite series, representing the discounted sum of all future rewards starting from time t=0, is the discount factor, is the reward obtained at time t.

[0088] Specifically, the optimized process parameters include molten pool shape, temperature field distribution and defect probability.

[0089] Specifically, the combination of edge computing and high-bandwidth communications enables millisecond-level data collection and analysis, breaking through the time bottleneck of traditional technologies, integrating multimodal data such as optics, thermals, and acoustics, capturing more information, and significantly improving the accuracy and coverage of defect detection;

[0090] The reinforcement learning algorithm dynamically adjusts printing parameters, realizing intelligent and dynamic closed-loop control of the printing process for the first time; the combination of multimodal CNN and Transformer models makes defect prediction more accurate and effectively reduces false alarm and missed alarm rates.

[0091] Embodiment 2

[0092] In order to solve the above technical problems, based on the first embodiment, another technical solution adopted by the present application is:

[0093] A method for monitoring metal 3D printing forming accuracy comprises the following steps:

[0094] Step 1: Obtain multimodal data. Use a high-speed industrial camera to capture the dynamics of the molten pool, obtain surface morphology changes and powder deposition information, configure an infrared thermal imager to obtain the temperature field distribution of the molten pool and its surroundings in real time, identify potential thermal instability areas, and use high-frequency acoustic sensors to monitor internal defects such as pores or cracks that occur during the printing process. A unified timestamp system is used to ensure the time alignment of multimodal data acquisition.

[0095] Specifically, the optical monitoring module:

[0096] Equipment: Equipped with a high-speed industrial camera (e.g., 2000fps or more) and an adjustable focus lens, and a filter suitable for the high reflective properties of metal surfaces.

[0097] Function: Capture print pool dynamics (size, shape, boundary changes), powder deposition quality and scanning path trajectory.

[0098] Data types: grayscale image, RGB image, pixel-level coordinate points.

[0099] Thermal monitoring module:

[0100] Equipment: Use high-resolution infrared thermal imager (wavelength 3-5μm) combined with temperature correction module. Function: Record the temperature field distribution of the molten pool and the surrounding environment, track thermal gradients and heat accumulation areas. Data type: Matrix of pixel-by-pixel temperature values, frame rate synchronized with optical monitoring.

[0101] Acoustic Monitoring Module:

[0102] Equipment: Install a broadband high-sensitivity acoustic sensor (frequency range to ), and reduce the interference of mechanical vibration through seismic isolation devices. Function: Monitor the acoustic wave characteristics generated by delamination, microcracks or pores inside the material during printing. Data type: time domain and frequency domain signals.

[0103] Synchronous control and data transmission: Use the master control unit (FPGA or edge computing node) to unify the timestamp, collect and align multimodal data; transmit data through high-speed interfaces (such as EtherCAT or optical fiber) to ensure low transmission latency;

[0104] Specifically, it realizes the synchronous and high-precision collection of multimodal data, provides a basis for subsequent data fusion, and realizes the high-precision and synchronous collection of three types of data: optical, thermal and acoustic, providing multi-dimensional data support for subsequent fusion analysis.

[0105] Step 2: Multimodal data preprocessing and feature extraction;

[0106] Data cleaning: remove noise and blurry areas in optical images; smooth fluctuations in thermal imaging data;

[0107] Filter out ambient noise from acoustic signals.

[0108] Data alignment: Ensure multimodal data synchronization through time alignment algorithms.

[0109] Feature extraction: Use convolutional neural networks (CNNs) to extract boundary and geometric features from optical images;

[0110] Thermal field gradient analysis is used to extract key temperature change areas; spectral features are extracted from acoustic data based on time-frequency analysis.

[0111] Specifically, cleaning, alignment and preprocessing feature extraction provide high-quality input for subsequent data fusion.

[0112] Optical data processing uses Gaussian filtering to remove random noise in the image, enhance the boundary of the printed melt pool, and extract geometric features such as melt pool shape and powder coverage through edge detection algorithms (such as Canny).

[0113] Thermal data processing, based on bilateral filtering, reduces thermal imaging noise while retaining thermal gradient information, calculates temperature distribution histogram and thermal gradient change rate, and forms a feature matrix.

[0114] Acoustic data processing, segmented Fourier transform to obtain spectral features, identify abnormal frequency bands (such as microcrack acoustic wave characteristics), extract short-time Fourier transform (STFT) feature maps, and input them into the deep learning model.

[0115] Data alignment and standardization, based on unified timestamps, uses interpolation to complete uneven time series and normalizes all modal data to Scope to ensure that multimodal features are consistent, clean multimodal data, extract key features, and provide high-quality data for deep learning model input.

[0116] Step 3: Build a multimodal data fusion and deep learning model, implement direct splicing at the data layer (such as combining optical, thermal, and acoustic features into a unified feature vector), use an attention mechanism at the feature layer, and assign weights to the mode that best explains the defect. Multimodal data fusion and deep learning model structure design:

[0117] Transformer-based multimodal neural network processes sequence and spatial features; multi-head attention mechanism extracts inter-modal interaction features; predicts possible defect location, type and severity.

[0118] Specifically, the data fusion method directly splices the optical, thermal, and acoustic feature vectors to form a multimodal feature set, adopts a multi-head attention mechanism, assigns dynamic weights to different modal features, and adjusts the modal priority according to the melt pool state.

[0119] Input layer: receives pre-processed optical, thermal and acoustic features;

[0120] Backbone network: A Transformer-based multimodal deep learning network that uses a self-attention mechanism to learn the association between modalities;

[0121] Output layer: predicts defect type (e.g., pores, cracks, surface roughness anomalies) and location.

[0122] The formula of the Transformer-based multimodal neural network is as follows: assuming that the optical, thermal, and acoustic characteristics are , , , the multimodal features are ;

[0123] ;

[0124] in, The operation is to concatenate (connect) the data of these three modes.

[0125] The mechanism is used to process the spliced ​​data in order to perform weighted fusion of data from different modalities.

[0126] Final defect prediction for:

[0127] ;

[0128] in, is the output of the first formula, is the weight matrix, is the bias vector, The function is used to convert the result of linear transformation into probability distribution;

[0129] Through deep learning models, efficient fusion of multimodal data and identification of complex defects can be achieved.

[0130] Step 4: Based on the defect location and type output by the model, evaluate its impact on the performance of the printed part in real time:

[0131] Use simulation tools based on finite element analysis to predict stress concentration or thermal instability caused by defects. When defects threaten the molding quality, the control system triggers an alarm and pauses printing. The detected potential defects are evaluated in real time and an alarm or correction mechanism is triggered.

[0132] Step 5: Dynamically adjust key process parameters such as laser power and scanning speed based on the defect prediction results.

[0133] Reinforcement learning controller: Use the Proximal Policy Optimization (PPO) algorithm to optimize the printing strategy and ensure the system's adaptability in dynamic environments.

[0134] Defect classification assessment: Combined with the output of the deep learning model, defects are identified as geometric deformation, thermal instability, or internal porosity, and finite element analysis (FEA) is used to evaluate the impact of defects on mechanical properties (such as stress concentration).

[0135] Real-time alarm: set defect thresholds (such as maximum allowable pore size, temperature gradient) and trigger alarms when they exceed the range;

[0136] Dynamic Response: Pauses printing and marks affected areas.

[0137] The reinforcement learning algorithm is formulated as follows: The optimal strategy is determined by maximizing the expected reward, as follows:

[0138] ;

[0139] in, is the optimal strategy, It is a mathematical operation, which means finding the parameter that makes the following expression take the maximum value. ; is the mathematical expectation, is an infinite series, representing the discounted sum of all future rewards starting from time t=0, is the discount factor, is the reward obtained at time t;

[0140] Optimize printing parameters through closed-loop control, minimize defect rate, evaluate printing defects in real time, promptly issue alarms and take measures to correct problems.

[0141] Specifically, multimodal data fusion deep learning technology is used to achieve comprehensive monitoring of the printing status from the multi-dimensional perspectives of optics, thermals, and acoustics; a deep learning model with an attention mechanism is integrated to capture the complex correlation between modalities. Data fusion improves the accuracy and coverage of defect detection and reduces the false negative rate of a single sensor; closed-loop feedback and optimization strategies adjust printing parameters in real time, greatly improving molding accuracy and printing stability. The present invention improves the comprehensiveness and real-time nature of monitoring, can provide deep insights into complex, multi-physical field coupled metal 3D printing processes, and has significant foresight and practical application value.

[0142] Embodiment 3

[0143] like Figure 2As shown, in order to solve the above technical problems, based on the first embodiment, another technical solution adopted by the present application is: a metal 3D printing forming accuracy monitoring method, further comprising:

[0144] Step 1: Based on the holographic microscope, the three-dimensional structural changes of the printed molten pool are collected in real time to analyze the surface defects and depth information during printing;

[0145] Step 2: According to the holographic microscope and the multi-dimensional data collection, the data collection method is adjusted to determine the interference pattern; based on the Fourier transform algorithm, the interference pattern is converted into a visualized three-dimensional structure, and the surface roughness and crack depth are calculated;

[0146] Step 3: Use a deep convolutional neural network to clean up the noise and artifacts in the holographic data, and determine the spatiotemporal encoder to achieve time series alignment and embedding conversion of inter-modal features;

[0147] Step 4: Build a multi-task deep learning model for multi-dimensional data, map the multi-dimensional data features to quantum states, and realize feature representation in high-dimensional space through quantum circuits;

[0148] Step 5: Build a digital twin model of the printing equipment and process, simulate the physical and chemical behavior of the printing process in real time, and dynamically update the status of the digital twin model based on monitoring data and historical databases.

[0149] Specifically, a high-resolution holographic microscope (with nanometer-level resolution) is used to collect the three-dimensional structural changes of the printed melt pool in real time.

[0150] Data type: Holographic interference pattern (light field intensity + phase), which can analyze subtle surface defects and depth information during the printing process.

[0151] Specifically, it supplements the lack of optical and thermal data and enables comprehensive monitoring of micro-geometry and internal material properties;

[0152] By unifying the clock to synchronize the collection of holographic imaging and other modal data, the time accuracy is ensured to be within the microsecond range.

[0153] Dynamically adjust the sampling frequency of each sensor, adjust the data collection strategy in time according to the complexity of the printing process, and reduce redundancy.

[0154] Specifically, the Fourier transform algorithm is used to transform the interference pattern into a visual three-dimensional structure, and parameters such as surface roughness and microcrack depth are calculated.

[0155] Clean up the noise and artifacts in the holographic data through deep convolutional neural network (DCNN) to improve the reconstruction quality;

[0156] SpatiotemporalEncoder is used to achieve time series alignment and embedding conversion of inter-modal features. They are unified into embedding vectors for easy subsequent model processing.

[0157] Model architecture, input layer: input feature vectors of four modes: optical, thermal, acoustic and holographic.

[0158] Shared encoder: Extracts cross-modal global features based on the Transformer structure and dynamically adjusts the weights of each modality.

[0159] Task Decoder: Decodes three main tasks:

[0160] Defect classification (e.g. pores, cracks, etc.).

[0161] Accurately locate defects (output 3D coordinates).

[0162] Process parameter prediction (such as laser power, scanning speed, etc.).

[0163] Total loss function :

[0164] ;

[0165] in , , is a weight factor, which is dynamically adjusted according to the importance of the task; is the classification loss, is the positioning loss, is the parameter loss.

[0166] Step 4: Quantum feature embedding and optimization of quantum feature embedding;

[0167] Map multimodal features to quantum states and realize feature representation in high-dimensional space through quantum circuits.

[0168] Quantum embedding formula:

[0169] ;

[0170] in, is the modal feature, is the corresponding quantum state quantum optimization process;

[0171] The variational quantum algorithm (VQA) is used to optimize the data fusion weights and find the optimal feature combination.

[0172] Specifically: Utilize the characteristics of quantum superposition and entanglement to greatly improve the efficiency of feature interaction and fusion.

[0173] Build a digital twin model of the printing equipment and process to simulate the physical and chemical behavior of the printing process in real time.

[0174] Data input: Combine monitoring data with historical database to dynamically update the twin model status.

[0175] Based on the prediction results of the twin system, key process parameters such as laser power and scanning speed are optimized in real time to prevent defects from expanding.

[0176] The adaptive control algorithm formula is as follows:

[0177] ;

[0178] in, To monitor the error, is the process parameter adjustment amount; is the proportionality coefficient, is the integral term, is the differential term.

[0179] Specifically, it provides three-dimensional structural information of the microscopic morphology of the printing process, which is a major supplement to the current mainstream monitoring technology.

[0180] It realizes high-dimensional feature fusion optimization and has forward-looking advantages in complex data processing.

[0181] Through real-time virtual simulation and control, monitoring technology and intelligent manufacturing are deeply integrated.

[0182] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0183] Embodiment 4

[0184] like Figure 3 The present invention provides a schematic diagram of the structure of an electronic device according to an embodiment of the present invention, which is suitable for implementing the electronic device in the embodiment of the present invention. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0185] like Figure 3 As shown, an electronic device includes a processor, a memory and a communication interface. The memory stores a computer program. When the processor executes the computer program, a metal 3D printing forming accuracy monitoring method of each embodiment of the present disclosure is implemented. The electronic device can exchange data with other devices or systems through the communication interface to realize real-time update and sharing of drug information.

[0186] The processor in the above-mentioned electronic device serves as the core of the electronic device. The processor is responsible for executing the computer program stored in the memory to realize various functions of the paperless conference terminal intelligent interaction method. The processor can use a high-performance multi-core CPU or a dedicated chip to meet the needs of complex calculations and real-time processing. The memory is used to store operating systems, applications, data, computer programs, etc. In this embodiment, the memory stores a computer program for realizing the paperless conference terminal intelligent interaction method. The memory can be RAM, ROM, Flash memory or other types of non-volatile memory. The communication interface is used to connect the electronic device with other devices or networks to realize data transmission and exchange. In this embodiment, the communication interface supports a variety of communication protocols and interface standards, such as Wi-Fi, Bluetooth, USB, Ethernet, etc., to meet the communication needs in different scenarios.

[0187] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0188] Embodiment 5

[0189] According to a computer-readable storage medium of an embodiment of the present disclosure, a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the various functions of the metal 3D printing forming accuracy monitoring method of the aforementioned embodiments of the present disclosure are implemented.

[0190] The above-mentioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or mobile hard disk), media with built-in rewritable non-volatile memory (e.g., memory card) and media with built-in ROM (e.g., ROM box).

[0191] For detailed description of this embodiment, reference may be made to the corresponding descriptions in the aforementioned embodiments, which will not be repeated here.

[0192] Embodiment six:

[0193] See also Figure 4-Figure 8 The present invention provides a technical solution: a metal 3D printing molding accuracy monitoring device, further comprising a part molding cover 30, wherein a monitoring component 101 is installed inside the part molding cover 30, and the monitoring component 101 comprises a lifting seat 11, two limit plates 12, two limit rails 13, two sliders 14, a tooth plate 15, a mounting seat 16, a laser ranging sensor 17, a slide 18, a driving gear 19 and a servo motor 21;

[0194] The two limiting rails 13 are symmetrically opened on the upper surface and the lower surface of the lifting seat 11, the two sliders 14 are symmetrically fixedly connected to the two ends of the lifting seat 11, the slide groove 18 is opened inside the lifting seat 11, the tooth plate 15 is fixedly connected to the inner bottom wall of the slide groove 18, the driving gear 19 is coaxially fixedly connected to the output shaft of the servo motor 21, one side of the two limiting plates 12 is symmetrically fixedly connected to the upper surface and the lower surface of the mounting seat 16, the laser ranging sensor 17 is installed on the front surface of the mounting seat 16, and a display screen 26 is installed on one side of the part forming cover 30. The other side of the two limiting plates 12 away from the mounting seat 16 is slidably connected to the two limiting rails 13 one by one, and the driving gear 19 is located inside the slide groove 18, and the driving gear 19 is meshed with the tooth plate 15;

[0195] The model of the laser distance sensor 17 is: HC-J030;

[0196] The monitoring component 101 is used to monitor the accuracy of the 3D printed parts. During the printing process, the 3D printed parts are located in the part molding cover 30. The monitoring component 101 can monitor the accuracy of the parts through the laser ranging sensor 17. The monitoring data are displayed on the display screen 26. By referring to the monitoring data and the design printing data, the printing error can be calculated;

[0197] By cooperating with structures such as the driving gear 19, the servo motor 21, the mounting seat 16, the toothed plate 15, etc., the laser ranging sensor 17 can be moved along the slide groove 18 to monitor the molding accuracy of different positions of the parts and realize all-round monitoring. The adjacent surfaces of the two limit plates 12 are slidably connected to the outer wall of the lifting seat 11, and the end of the limit plate 12 away from the mounting seat 16 is slidably connected to the inner wall of the limit track 13. The position of the mounting seat 16 can be limited by the limit plate 12 so that the mounting seat 16 can move horizontally along the limit track 13.

[0198] In this embodiment, specifically: the driving gear 19 is located inside the slide groove 18, the outer side wall of the driving gear 19 is meshed and connected to the upper surface of the tooth plate 15, the servo motor 21 is installed inside the mounting seat 16, the driving gear 19 is driven by the servo motor 21, the driving gear 19 is meshed with the tooth plate 15 through the teeth, and the servo motor 21 drives the mounting seat 16, so that the mounting seat 16 can be horizontally displaced, thereby monitoring the molding accuracy of different positions of the parts, and the monitoring component 101 also includes a driving motor 22, a screw 23, two guide rods 24 and two guide seats 25;

[0199] The two guide seats 25 are symmetrically installed at two adjacent corners in the part forming cover 30, the top end of the screw rod 23 is coaxially fixedly connected to the output shaft of the driving motor 22, the two ends of the two guide rods 24 are symmetrically fixedly connected to the inner top wall and the inner bottom wall of the guide seat 25, and the slider 14 is threadedly connected to the outer side wall of the screw rod 23;

[0200] The lead screw 23 is driven by the driving motor 22, and the lead screw 23 drives the slider 14 through the thread, and the slider 14 drives the lifting seat 11, and the lifting seat 11 drives the laser ranging sensor 17 through the mounting seat 16, so that the laser ranging sensor 17 can produce a vertical displacement, thereby realizing all-round monitoring of the precision of the formed parts. The slider 14 is slidably connected to the outer side walls of the two guide rods 24, and the slider 14 is slidably connected to the inside of the guide seat 25. The driving motor 22 is installed on the upper surface of the guide seat 25, and the lead screw 23 is driven by the driving motor 22, and the lead screw 23 drives the slider 14 to slide in the guide seat 25, and the slider 14 moves along the guide rod 24, and the guide rod 24 can play a guiding role.

[0201] The working principle or structural principle is that when the 3D printer is in use, during the printing process, the 3D printed part is located in the part forming cover 30, and the accuracy of the part is monitored by the laser ranging sensor 17. The monitoring data is displayed on the display screen 26. By referring to the monitoring data and the design printing data, the printing error can be calculated. The servo motor 21 drives the driving gear 19, and the driving gear 19 is meshed with the tooth plate 15 through the teeth. The servo motor 21 drives the mounting seat 16, which can cause the mounting seat 16 to produce horizontal displacement, thereby monitoring the molding accuracy of each horizontal position of the part. The driving motor 22 drives the lead screw 23, and the lead screw 23 drives the slider 14 through the thread. The slider 14 drives the lifting seat 11, and the lifting seat 11 drives the laser ranging sensor 17 through the mounting seat 16, which can cause the laser ranging sensor 17 to produce vertical displacement, thereby realizing all-round monitoring of the accuracy of the molded parts.

[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring metal 3D printing molding accuracy, characterized in that: The following steps are involved: Acquire multimodal sensor data. Based on the multimodal sensor data, collect multi-dimensional data in the metal 3D printing process, determine the optical sensor to record the molten pool morphology and powder distribution, determine the thermal infrared camera to monitor the temperature field distribution, determine the acoustic sensor to collect high-frequency sound wave signals in printing, and identify defect signals; According to the local preprocessing data of the edge computing device, the boundary and geometric features are extracted from the optical image based on the convolutional neural network, and the key temperature change area is extracted in response to the thermal field gradient analysis; Extract spectral features from acoustic data based on time-frequency analysis; Construct a multimodal convolutional neural network, input optical, thermal and acoustic data into the model for feature fusion and classification, analyze the dynamic changes in the printing process based on the Transformer time series prediction model, and simulate future printing quality; According to the analysis results, the printing parameters are adjusted dynamically. The printing parameters are dynamically optimized through environmental feedback based on the reinforcement learning algorithm. The printing parameters are connected to the printing device based on the printing parameters, and the printing parameters are adjusted in real time to achieve closed-loop control. Build a web-based visual interface to display key indicators and optimize process parameters through closed-loop control; Based on the holographic microscope, the 3D structural changes of the printed melt pool are collected in real time to analyze the surface defects and depth information during printing; According to the holographic microscope and the multi-dimensional data collection, the data collection method is adjusted to determine the interference pattern; based on the Fourier transform algorithm, the interference pattern is converted into a visualized three-dimensional structure, and the surface roughness and crack depth are calculated; Use deep convolutional neural networks to clean up noise and artifacts in holographic data, and determine the spatiotemporal encoder to achieve time series alignment and embedding conversion of inter-modal features; Build a multi-task deep learning model for multi-dimensional data, map the features of multi-dimensional data to quantum states, and realize feature representation in high-dimensional space through quantum circuits; Build a digital twin model of the printing equipment and process, simulate the physical and chemical behavior of the printing process in real time, and dynamically update the status of the digital twin model based on monitoring data and historical databases.

2. A metal 3D printing forming accuracy monitoring method according to claim 1, characterized in that: The multi-dimensional data includes the molten pool temperature field, the powder deposition state and the laser reflection signal. The formula of the multi-dimensional data is expressed as follows: : in, , and They are optical, thermal and acoustic data respectively; The output of the multimodal convolutional neural network F is the defect prediction result ; 。 3. A metal 3D printing forming accuracy monitoring method according to claim 1, characterized in that: The dynamically adjusted printing parameters include laser power, scanning speed and powder supply, and the formula of the reinforcement learning algorithm is expressed as follows: The optimal strategy is determined by maximizing the expected reward, as follows: ； in, is the optimal strategy, It is a mathematical operation, which means finding the parameter that makes the following expression take the maximum value. ; is the mathematical expectation, is an infinite series, representing the discounted sum of all future rewards starting from time t=0, is the discount factor, is the reward obtained at time t.

4. A metal 3D printing forming accuracy monitoring method according to claim 1, characterized in that: The optimized process parameters include molten pool shape, temperature field distribution and defect probability.

5. A metal 3D printing molding accuracy monitoring device, applied to the metal 3D printing molding accuracy monitoring method according to any one of claims 1 to 4, characterized in that: It also includes a part forming cover (30), wherein a monitoring assembly (101) is installed inside the part forming cover (30), and the monitoring assembly (101) includes a lifting seat (11), two limit plates (12), two limit rails (13), two sliders (14), a tooth plate (15), a mounting seat (16), a laser distance sensor (17), a slide groove (18), a driving gear (19) and a servo motor (21); The two limit rails (13) are symmetrically opened on the upper surface and the lower surface of the lifting seat (11); the two sliders (14) are symmetrically fixedly connected to the two ends of the lifting seat (11); the slide groove (18) is opened inside the lifting seat (11); the tooth plate (15) is fixedly connected to the inner bottom wall of the slide groove (18); the driving gear (19) is coaxially fixedly connected to the output shaft of the servo motor (21); one side of the two limit plates (12) is symmetrically fixedly connected to the upper surface and the lower surface of the mounting seat (16); the laser distance sensor (17) is installed on the front surface of the mounting seat (16); a display screen (26) is installed on one side of the part forming cover (30); the other side of the two limit plates (12) away from the mounting seat (16) is slidably connected to the two limit rails (13) in a one-to-one correspondence; the driving gear (19) is located inside the slide groove (18); and the driving gear (19) is meshedly connected to the tooth plate (15).

6. A metal 3D printing forming accuracy monitoring device according to claim 5, characterized in that: The servo motor (21) is installed inside the mounting seat (16), and the monitoring assembly (101) further comprises a drive motor (22), a screw rod (23), two guide rods (24) and two guide seats (25); The two guide seats (25) are symmetrically mounted at two adjacent corners in the part forming cover (30); the top end of the screw rod (23) is coaxially fixedly connected to the output shaft of the drive motor (22); and the two ends of the two guide rods (24) are symmetrically fixedly connected to the inner top wall and the inner bottom wall of the guide seat (25).

7. A metal 3D printing forming accuracy monitoring device according to claim 6, characterized in that: The bottom end of the screw rod (23) is rotatably connected to the inner bottom wall of the guide seat (25), the slider (14) is threadedly connected to the outer side wall of the screw rod (23), the slider (14) is slidably connected to the outer side walls of the two guide rods (24), the slider (14) is slidably connected to the inside of the guide seat (25), and the drive motor (22) is mounted on the upper surface of the guide seat (25).

8. An electronic device, characterized in that: It comprises a processor, a memory and a communication interface, wherein a computer program is stored in the memory, and when the processor executes the computer program, a metal 3D printing forming accuracy monitoring method as described in any one of claims 1 to 4 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, a metal 3D printing forming accuracy monitoring method according to any one of claims 1 to 4 is implemented.

Images