Plasma additive manufacturing parameter pre-optimization system based on digital twinning
By optimizing plasma additive manufacturing parameters using digital twin technology, the problem of inaccurate process parameter optimization has been solved, achieving high-precision porosity control and robust guidance, adapting to the rapid development of new materials and components, and improving the processing quality of molded parts.
Patent Information
- Application Number
- CN202611548986.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-08
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for optimizing plasma additive manufacturing process parameters cannot accurately match actual processing results, leading to inaccurate porosity control and a lack of robust range guidance, making it difficult to meet the needs of rapid development of new materials and components.
A plasma additive manufacturing parameter pre-optimization system based on digital twins is adopted. By generating multiple sets of process parameter combinations for simulation, a porosity prediction model is constructed. The model is trained by combining virtual and real deviation sample sets, and the optimal process parameters and porosity risk distribution are output, realizing high-precision mapping and robust control of parameters.
It improves the accuracy and robustness of process parameter optimization, reduces trial and error experiments, lowers material and time costs, adapts to the rapid development needs of new materials and components, and improves the processing qualification rate of molded parts.
Smart Images

Figure CN123120154A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of digital twin technology, specifically to a pre-optimization system for plasma additive manufacturing parameters based on digital twins. Background Technology
[0002] Plasma powder bed melting technology, with its advantages of high energy input and fast forming efficiency, has been increasingly widely used in the fabrication of large-size complex metal components. Porosity is a core indicator for evaluating the mechanical properties and service reliability of plasma additive manufacturing parts. Excessive porosity can lead to substandard strength and fatigue performance of the formed parts, directly affecting the safety of the final components. In actual production, obtaining low-porosity formed parts that meet the requirements often requires adjusting process parameters through numerous trial-and-error experiments. This not only consumes a lot of materials and time but also makes it difficult to adapt to the rapid development needs of new materials and components.
[0003] However, most existing process parameter optimization methods are based on empirical formulas or offline experimental data to build predictive models, which cannot reflect the systematic deviation between digital simulation and actual processing. Affected by individual differences in equipment, fluctuations in environmental conditions, and the dispersion of material properties, the optimal parameters obtained through simulation often fail to achieve the expected porosity effect in actual processing, and the accuracy and robustness of parameter optimization are difficult to guarantee.
[0004] Furthermore, existing methods rarely provide a corresponding robustness range for given optimal parameters. When process parameters or material properties fluctuate slightly, the porosity of the molded parts can easily exceed the acceptable range, further increasing the difficulty of quality control in actual production. Summary of the Invention
[0005] This application provides a plasma additive manufacturing parameter pre-optimization system based on digital twins, which solves the technical problem that existing simulation-based optimal parameters cannot match actual processing results, resulting in poor parameter optimization accuracy and a lack of robust range guidance.
[0006] The technical solution to the above-mentioned technical problems in this application is as follows: This application provides a pre-optimization system for plasma additive manufacturing parameters based on digital twins, including: The initial process parameter combination acquisition module is used to generate multiple sets of process parameter combinations in the process parameter space of the plasma powder bed melting equipment, drive the digital twin model to perform powder bed melting process simulation on the multiple sets of process parameter combinations respectively, and obtain parameter-porosity simulation sample set; The process parameter extraction module is used to construct a porosity prediction model based on the parameter-porosity simulation sample set, use the porosity prediction model to perform iterative search in the process parameter space to obtain a virtual optimal parameter candidate set, and extract several sets of process parameters to be processed from it. The virtual-real deviation sample set construction module is used to drive the plasma powder bed melting equipment to print test samples according to the several sets of processing parameters, collect the real physical quantity data of the test sample printing process, compare it with the simulation prediction value output by the digital twin model, and combine the residual obtained by comparison with the corresponding process parameter combination and equipment operating status parameter combination to form a sample to construct the virtual-real deviation sample set. The optimal process parameter combination output module is used to train a virtual-real deviation prediction model based on the virtual-real deviation sample set, compensate and calibrate the simulated predicted porosity corresponding to each group of process parameters in the virtual optimal parameter candidate set, and output the optimal process parameter combination after compensation and calibration, as well as the corresponding porosity risk distribution and parameter robustness boundary.
[0007] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides a parameter pre-optimization system for plasma additive manufacturing based on digital twins. First, a high-precision mapping between process parameters and forming porosity is achieved through a digital twin model. Combined with a porosity prediction model based on Gaussian process regression, rapid parameter optimization in virtual space is achieved, avoiding numerous initial trial-and-error experiments. Second, a virtual-to-real deviation prediction model is constructed by collecting real processing data from a small number of test samples. This model compensates for deviations in the prediction results of the digital twin simulation, effectively reducing the systematic deviation between simulation predictions and actual processing, and improving the accuracy of parameter optimization. Finally, for the output optimal combination of process parameters, the robustness boundaries of the corresponding process parameters and material thermophysical parameters, as well as the global porosity risk distribution, are simultaneously output. This provides guidance for quality control in actual production. When small fluctuations in parameters occur during processing, production personnel can quickly determine whether the fluctuation range is within the acceptable range, effectively reducing the risk of exceeding porosity limits.
[0008] Through the above technical solution, this application solves the technical problem that the optimal parameters obtained by simulation cannot match the actual processing results, resulting in poor accuracy of parameter optimization and lack of robust range guidance. It can quickly complete the pre-optimization of process parameters before the production of new materials, and obtain the optimal process parameters that fit the actual processing results with only a small number of trial prototypes. At the same time, it effectively reduces material and time costs, adapts to the rapid development needs of new materials and new components, and improves the first-pass yield of molded parts. Attached Figure Description
[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0010] Figure 1 This is a schematic diagram of the structure of the plasma additive manufacturing parameter pre-optimization system based on digital twin provided in the embodiments of this application; Figure 2 This is a schematic diagram of the process for obtaining a virtual optimal parameter candidate set in the plasma additive manufacturing parameter pre-optimization system based on digital twin provided in the embodiments of this application.
[0011] The components represented by each number in the attached diagram are explained below: The module 11 is for obtaining the initial process parameter combination, the module 12 is for extracting the process parameters to be processed, the module 13 is for constructing the virtual and real deviation sample set, and the module 14 is for outputting the optimal process parameter combination. Detailed Implementation
[0012] This application provides a plasma additive manufacturing parameter pre-optimization system based on digital twins to address the technical problem that existing simulation-based optimal parameters cannot match actual processing results, resulting in poor parameter optimization accuracy and a lack of robust range guidance.
[0013] Examples, such as Figure 1 As shown in the embodiments of this application, a plasma additive manufacturing parameter pre-optimization system based on digital twins is provided, including: The initial process parameter combination acquisition module 11 is used to generate multiple sets of process parameter combinations in the process parameter space of the plasma powder bed melting equipment, drive the digital twin model to perform powder bed melting process simulation on the multiple sets of process parameter combinations respectively, and obtain parameter-porosity simulation sample set; Specifically, the Latin hypercube sampling method is used to generate multiple sets of initial process parameter combinations in a multidimensional process parameter space consisting of plasma arc power, scanning speed, layer thickness and scanning spacing. The Latin hypercube sampling method divides the value range of each process parameter into intervals equal to the number of sampling points, randomly selects a sampling value in each interval, and combines the sampling values of each process parameter according to the principle of non-repetition pairing.
[0014] In this embodiment, firstly, multiple sets of process parameter combinations are generated within the process parameter space of the plasma powder bed melting equipment. The process parameter space consists of plasma arc power, scanning speed, layer thickness, and scanning spacing, and is a four-dimensional continuous parameter space. The Latin hypercube sampling method is used to generate initial sampling points within the parameter space. The Latin hypercube sampling method divides the value range of each process parameter into intervals equal to the number of sampling points. A sampling value is randomly selected from each interval, and the sampling values of each process parameter are combined according to the principle of non-repetition pairing. This ensures that the sampling points are evenly distributed within the parameter space and that there are no repeated combinations. Compared with simple random sampling, this method can cover the entire parameter space with fewer sampling points, avoid the clustering of sampling points in local intervals, and improve the training efficiency and accuracy of the subsequent porosity prediction model.
[0015] For example, taking a certain titanium alloy material as an example, the plasma arc power ranges from 500W to 1500W, the scanning speed ranges from 200mm / s to 800mm / s, the layer thickness ranges from 50μm to 150μm, and the scanning spacing ranges from 0.5mm to 1.5mm. In this embodiment, the number of sampling points is set to 50 groups, covering all areas of the process parameter space.
[0016] Furthermore, after generating multiple sets of process parameter combinations, the digital twin model is driven to perform powder bed melting process simulation on the multiple sets of process parameter combinations to obtain a parameter-porosity simulation sample set. The digital twin model is a multiphysics simulation model, which can output the simulation predicted temperature field, simulation predicted layer height and simulation predicted porosity for each set of input process parameter combinations. After pairing each set of process parameters with the corresponding output simulation predicted porosity, a parameter-porosity simulation sample set containing multiple sets of paired data can be obtained.
[0017] The construction steps of the digital twin model include: A geometric model of the plasma powder bed melting equipment is established, taking the forming cavity, powder bed, and substrate as objects. Based on the geometric model, a heat conduction model, a molten pool dynamics model, and a porosity calculation model for the powder bed melting process are established and integrated into a digital twin model. The heat conduction model is used to calculate the spatial distribution of the temperature field of the powder bed under the action of the plasma arc. The molten pool dynamics model is used to calculate the width and depth of the molten pool. The porosity calculation model calculates the porosity based on the solidification state of the molten pool. Material property vectors are constructed based on the melting point, thermal conductivity, powder absorption rate, and apparent density of the powder bed of the new material. The material property vectors are then input into the digital twin model to complete the material parameter configuration. The digital twin model receives a combination of process parameters as simulation input and outputs the corresponding simulation predicted temperature field, simulation predicted layer height, and simulation predicted porosity.
[0018] In this embodiment, firstly, a geometric model of the plasma powder bed melting equipment is established, taking the forming cavity, powder bed and substrate as objects. Based on the geometric model, a heat conduction model, a molten pool dynamics model and a porosity calculation model of the powder bed melting process are established and integrated into a digital twin model. Among them, the heat conduction model uses the Eagar-Tsai moving Gaussian distribution heat source model to calculate the spatial distribution of the temperature field of the powder bed under plasma arc. Compared with the point heat source model, this model avoids the temperature field singularity problem and can more accurately predict the temperature distribution of the high-temperature zone and the heat-affected zone. The molten pool dynamics model calculates the molten pool width and depth based on the temperature field output by the heat conduction model. The porosity calculation model calculates the porosity based on the solidification state of the molten pool. For example, when the molten pool depth-to-width ratio is less than 1.62 and the ratio of the scanning interval to the molten pool width is greater than 0.76, it is determined that there are interlayer unfused pores. The molten pool depth-to-width ratio of less than 1.62 indicates that the molten pool depth is insufficient, and the ratio of the scanning interval to the molten pool width is greater than 0.76 indicates that the overlap rate between adjacent molten channels is insufficient. When both conditions are met, it is determined to be an unfused pore.
[0019] The weld pool depth-to-width ratio criterion of 1.62 originates from the critical relationship between weld pool cross-sectional morphology and incomplete fusion defects in welding metallurgy. Based on the analytical solution of Rosenthal's moving heat source and the weld pool geometry model, when the weld pool depth-to-width ratio exceeds 1.62, the weld pool enters a deep-melting mode, with sufficient but narrow weld root penetration. In this case, sufficiently high heat input and a suitable scanning interval are required to avoid incomplete fusion. When the depth-to-width ratio is less than 1.62, the weld pool is in a conduction mode. If the ratio of the scanning interval to the weld pool width is greater than 0.76, the overlap between adjacent weld passes is insufficient, failing to completely cover the edge of the previous weld pass, thus forming incomplete fusion porosity. These two criteria together constitute the complete criterion for incomplete fusion defects: a depth-to-width ratio less than 1.62 indicates insufficient weld pool depth, and a scanning interval to weld pool width ratio greater than 0.76 indicates insufficient overlap between passes. When both conditions are met simultaneously, it is determined to be an incomplete fusion porosity.
[0020] It should be noted that the aforementioned critical values of 1.62 and 0.76 are theoretically derived based on the geometric relationship of the molten pool cross-section under ideal heat conduction conditions. In practical applications, these values can be corrected according to the thermal properties of the target material and the characteristics of the equipment. The correction method involves calibrating the experimental data using a digital twin model. Using a set of known acceptable process parameters for the molten pool aspect ratio and scanning distance ratio as a benchmark, a material property correction coefficient is calculated. The theoretical critical value is then multiplied by this correction coefficient to serve as the criterion threshold for practical applications, ensuring that the criterion matches the actual molten pool morphology of the target material and equipment. The correction coefficient typically ranges from 0.85 to 1.15. For example, for materials with high thermal conductivity, such as aluminum alloys, the tendency for molten pool widening increases, and the critical value for the aspect ratio can be appropriately lowered; for materials with low thermal conductivity, such as titanium alloys, molten pool widening is limited, and the critical value for the aspect ratio can be appropriately increased.
[0021] Specifically, the digital twin model receives a combination of process parameters as simulation input and outputs the corresponding simulation prediction temperature field, simulation prediction layer height, and simulation prediction porosity. Before simulation, a material property vector needs to be constructed based on the melting point, thermal conductivity, powder absorption rate, and apparent density of the powder bed of the new material.
[0022] For example, the digital twin model training uses the combination of process parameters and the plasma arc stability index as input features, and uses the porosity output by the digital twin model simulation as the training label for supervised training. The training dataset comes from the simulation results of the digital twin model on the combination of Latin hypercube sampling process parameters. The digital twin model training uses the Adam optimizer, with an initial learning rate of 0.001, a batch size of 64, a maximum training epoch of 200 epochs, and an early stopping mechanism with a patience of 20. Training is terminated when the validation set loss does not decrease for 20 consecutive epochs, resulting in a trained digital twin model.
[0023] The process parameter extraction module 12 is used to construct a porosity prediction model based on the parameter-porosity simulation sample set, use the porosity prediction model to perform iterative search in the process parameter space to obtain a virtual optimal parameter candidate set, and extract several sets of process parameters to be processed from it. In this embodiment of the application, a porosity prediction model based on Gaussian process regression is constructed based on the parameter-porosity simulation sample set. This model can output the predicted porosity value and prediction confidence interval of the porosity under given process parameters. Then, iterative search is performed in the process parameter space to obtain a virtual optimal parameter candidate set, and several groups of parameters with the highest expected improvement rate are selected from the candidate set as the process parameters to be processed.
[0024] The porosity prediction model is constructed based on the parameter-porosity simulation sample set, including: During the simulation process of the digital twin model, the plasma arc current fluctuation sequence and the plasma arc voltage fluctuation sequence are collected in real time. The ratio of the standard deviation to the mean of the plasma arc current fluctuation sequence and the ratio of the standard deviation to the mean of the plasma arc voltage fluctuation sequence are calculated respectively. The two are then weighted and summed according to a pre-calibrated weighting coefficient to obtain the plasma arc stability index. The weighting coefficient is pre-calibrated according to the degree of influence of plasma arc current and voltage on the uniformity of heat input. Using the combination of process parameters and the plasma arc stability index as input, and the porosity corresponding to the parameter-porosity simulation sample set as the training label, a Gaussian process regression model is trained. The plasma arc stability index is introduced as a covariate into the kernel function of the Gaussian process regression model. The trained Gaussian process regression model is used as a porosity prediction model. The Gaussian process regression model outputs the predicted porosity value and prediction confidence interval for each combination of process parameters. The prediction confidence interval is calculated from the posterior variance of the Gaussian process regression model.
[0025] In this embodiment, firstly, a porosity prediction model is constructed based on the parameter-porosity simulation sample set. During the simulation process of the digital twin model, the plasma arc current fluctuation sequence and the plasma arc voltage fluctuation sequence are collected in real time. The ratio of the standard deviation to the mean of the current fluctuation sequence and the ratio of the standard deviation to the mean of the voltage fluctuation sequence are calculated respectively. The two are weighted and summed according to the pre-calibrated weight coefficients to obtain the plasma arc stability index. The weight coefficients are pre-calibrated according to the degree of influence of plasma arc current and voltage on the uniformity of heat input. Specifically, the contribution ratio of current fluctuation and voltage fluctuation to heat input fluctuation is statistically analyzed through historical processing data, and the contribution ratio is directly set as the corresponding weight coefficient. For example, in this embodiment, the current stability weight is 0.6 and the voltage stability weight is 0.4. The stability index reflects the energy input stability of the plasma arc. The closer the index is to 0, the more stable the arc is. For example, the plasma arc stability index = (current standard deviation / current mean) × 0.6 + (voltage standard deviation / voltage mean) × 0.4. The calculated index ranges from 0 to 1. The closer the index is to 0, the smaller the energy input fluctuation of the arc and the more stable the forming process is.
[0026] Secondly, using the combination of process parameters and the plasma arc stability index as inputs, and the porosity corresponding to the parameter-porosity simulation sample set as the training label, a Gaussian process regression model is trained. The kernel function of the Gaussian process regression model adopts a combination of a squared exponential kernel and a white noise kernel; for example, the kernel function adopts the ARD squared exponential kernel. Where M is the length-scale diagonal matrix, σf is the signal variance, the hyperparameters of the kernel function are obtained by maximizing the logarithmic marginal likelihood estimation, and the plasma arc stability index is introduced into the kernel function as a covariate so that the porosity prediction model can dynamically adjust the prediction uncertainty according to the arc stability. The trained Gaussian process regression model outputs the porosity prediction value and prediction confidence interval for each combination of process parameters. The prediction confidence interval is calculated from the posterior variance of the model.
[0027] For example, the porosity prediction model uses the L-BFGS-B optimizer for hyperparameter optimization, with a maximum number of iterations set to 500 rounds and a convergence tolerance set to 1e-5. After each iteration, the covariance matrix and its inverse matrix are recalculated, and the hyperparameter values are updated. Hyperparameter boundary constraints are added during training to prevent optimization divergence. The signal variance is limited to 0.01 to 10, each length scale is limited to 0.01 to 5, and the noise standard deviation is limited to 0.001 to 1. When the improvement of the minimum porosity prediction value in 10 consecutive iterations is less than the convergence threshold of 0.005%, the porosity prediction model is obtained.
[0028] Furthermore, the porosity prediction model is used to iteratively search within the process parameter space to obtain a virtual optimal parameter candidate set, including: Before performing the iterative search, the width and depth of the molten pool corresponding to each combination of process parameters are simulated and recorded through the digital twin model, and the ratio of the width to the depth of the molten pool is calculated to obtain the width-to-depth ratio of the molten pool. Determine the lower critical value and the upper critical value of volumetric energy density when the width-to-depth ratio of the molten pool is within a preset reasonable range. Use the interval between the lower critical value and the upper critical value of volumetric energy density as the volumetric energy density constraint interval and set it as the boundary condition for iterative search. Based on the porosity prediction value and prediction confidence interval output by the porosity prediction model, a confidence upper bound acquisition function is constructed that introduces a plasma arc stability index penalty term. In each iteration, within the volumetric energy density constraint range, a combination of process parameters that maximizes the value of the confidence upper bound acquisition function is selected, driving the digital twin model to perform a new round of simulation, and updating the simulation results to the porosity prediction model; When the improvement of the predicted porosity value in consecutive iterations is lower than the preset convergence threshold, the iteration is terminated, and the combination of process parameters with the lowest current predicted porosity value is output as a virtual optimal parameter candidate set.
[0029] In the embodiments of this application, such as Figure 2 As shown, the porosity prediction model is first used to perform an iterative search in the process parameter space to obtain a virtual optimal parameter candidate set. Before performing the iterative search, the width and depth of the molten pool corresponding to each combination of process parameters are recorded by simulation using a digital twin model. The ratio of the molten pool width to the molten pool depth is calculated to obtain the molten pool width-to-depth ratio. It is determined that the molten pool width-to-depth ratio is within a preset reasonable range. For example, in this embodiment, the lower and upper critical values of the volumetric energy density corresponding to 1.0 to 1.8 are taken. The lower limit of 1.0 is obtained by converting the non-fusion criterion of a molten pool width-to-depth ratio of 1.62, and the upper limit of 1.8 is determined by the empirical upper limit of the width-to-depth ratio of the molten pool in the stable conduction mode in the powder bed melting process.
[0030] Specifically, the formula for calculating volumetric energy density is: Where P is the plasma arc power, v is the scanning speed, t is the layer thickness, and h is the scanning interval. The interval between the lower and upper critical values of the volumetric energy density is taken as the volumetric energy density constraint interval and set as the boundary condition for the iterative search.
[0031] Secondly, based on the predicted porosity value μ(x) and the prediction confidence interval σ(x) output by the porosity prediction model, a confidence upper bound acquisition function incorporating a plasmonic arc stability index penalty term is constructed, for example... ,in To explore the coefficients, a balanced setting is adopted based on search requirements. In this embodiment, the coefficient is taken as... , The penalty coefficient is set according to the process stability requirements. In this embodiment, λ=0.001 is used. Sarc is the plasma arc stability index. The penalty term can prioritize the parameter combination with higher arc stability during the acquisition process to improve the robustness of the candidate parameters.
[0032] Subsequently, in each iteration, the combination of process parameters that minimizes the value of the upper confidence bound acquisition function is selected within the volumetric energy density constraint range. This drives the digital twin model to perform a new round of simulation, and the simulation results are updated to the porosity prediction model. When the improvement of the porosity prediction value is lower than the preset convergence threshold in several consecutive iterations, the iteration is terminated. The combination of process parameters with the lowest current porosity prediction value is output as a virtual optimal parameter candidate set. The number of iterations is set according to the parameter space size and optimization accuracy requirements. For example, in this embodiment, the maximum number of iterations is set to 30. The preset convergence threshold is set based on the porosity detection accuracy of the equipment, for example, it can be set to 0.005%. The iteration is terminated when the improvement of the minimum porosity prediction value is lower than this threshold in three consecutive iterations. Finally, 10 combinations of process parameters with the lowest porosity prediction values are obtained as a virtual optimal parameter candidate set to be processed.
[0033] It should be noted that during the virtual optimization phase, since actual printing has not yet started, the plasma arc stability index and equipment operating status parameters corresponding to each candidate process parameter combination are still unknown. Therefore, the aforementioned dynamic characteristics cannot be obtained at this stage. To address this issue, this embodiment adopts a conservative strategy in the virtual optimization phase: the selection of each group of process parameters in the virtual optimal parameter candidate set is determined solely based on the simulated porosity prediction value and its confidence interval. In the subsequent trial printing phase, a virtual-to-real deviation sample set is constructed based on the plasma arc stability index and equipment operating status parameters collected during the actual printing process. Finally, the top 5 process parameter combinations with the highest expected improvement rate are extracted from the candidate set as the process parameters to be processed in the actual process, thus completing the parameter pre-optimization process.
[0034] The virtual-real deviation sample set construction module 13 is used to drive the plasma powder bed melting equipment to print test samples according to the several sets of processing parameters, collect the real physical quantity data of the test sample printing process, compare it with the simulation prediction value output by the digital twin model, and combine the residual obtained by comparison with the corresponding process parameter combination and equipment operating status parameter combination to form a sample to construct the virtual-real deviation sample set. In this embodiment, the plasma powder bed melting equipment is driven to print test samples according to several sets of extracted process parameters. During the printing process, the actual plasma arc current and voltage fluctuation sequence are collected simultaneously. After printing, the actual porosity of each test sample is obtained by industrial CT scanning, and the equipment operating status parameters are recorded. Then, the actual measured porosity of each test sample is compared with the simulated predicted porosity output by the digital twin model under the same process parameter combination to obtain the virtual-real porosity residual. The residual is then combined with the corresponding process parameter combination and the actual collected equipment operating status parameters to form a single sample, thus obtaining the virtual-real deviation sample set.
[0035] Specifically, the actual physical quantity data during the printing process of the test sample are collected and compared with the simulation prediction values output by the digital twin model. The residuals obtained from the comparison are combined with the corresponding process parameter combinations and equipment operating status parameters to form a sample, constructing a virtual-real deviation sample set, including: During the printing process of the test sample, the plasma arc current fluctuation sequence and the plasma arc voltage fluctuation sequence were collected, and the plasma arc stability index was calculated. The actual spatial distribution of the temperature field, the layer height of the test sample, and the cross-sectional porosity of the test sample were collected during the printing process. The actual temperature field spatial distribution, the height of the test sample layer, and the porosity of the test sample cross section are compared with the simulated predicted temperature field, simulated predicted layer height, and simulated predicted porosity output by the digital twin model, and the residual between the simulated predicted value and the actual measured value of each physical quantity is calculated. Using process parameter combinations, plasma arc stability index, and equipment operating status parameters as feature dimensions, and the residuals of each physical quantity as labels, the feature dimensions and corresponding labels are combined to form a sample, thus constructing a virtual-real deviation sample set.
[0036] In this embodiment, firstly, several sets are extracted from the virtual optimal parameter candidate set. For example, in this embodiment, 300 sets of processing parameters are selected. The plasma powder bed melting equipment is driven to print test samples according to three sets of parameters. During the printing process of the test samples, the plasma arc current fluctuation sequence and voltage fluctuation sequence are collected, and the plasma arc stability index is calculated according to the aforementioned method.
[0037] Simultaneously, the actual spatial distribution of the temperature field, the layer height of the test sample, and the cross-sectional porosity of the test sample were collected during the printing process. Specifically, the actual spatial distribution of the temperature field was obtained through a thermal imager, the layer height of the test sample was obtained through a laser displacement sensor, and the cross-sectional porosity of the test sample was obtained through metallographic microscope image analysis.
[0038] Furthermore, the actual spatial distribution of the temperature field, the layer height of the test sample, and the cross-sectional porosity of the test sample are compared item by item with the simulated predicted temperature field, simulated predicted layer height, and simulated predicted porosity output by the digital twin model. The residual between the simulated predicted value and the actual measured value of each physical quantity is calculated. Using the combination of process parameters, the plasma arc stability index, and equipment operating status parameters, such as the cumulative operating time of the equipment, as feature dimensions, and the residual of each physical quantity as labels, the feature dimensions and corresponding labels are combined to form a sample, thus constructing a virtual-real deviation sample set.
[0039] For example, the residual between the simulated predicted value and the actual measured value of each physical quantity can be calculated by the mean absolute error function. The average value of the absolute difference between the simulated predicted temperature and the actual temperature at each spatial location is taken as the overall residual of the temperature field. The floor height residual is directly taken as the difference between the simulated predicted floor height and the actual measured floor height. The porosity residual is taken as the difference between the simulated predicted porosity and the actual measured cross-sectional porosity. Finally, each sample corresponds to three residual labels for different physical quantities, which are used for training the subsequent deviation prediction model.
[0040] The optimal process parameter combination output module 14 is used to train a virtual-real deviation prediction model based on the virtual-real deviation sample set, compensate and calibrate the simulation prediction porosity corresponding to each group of process parameters in the virtual optimal parameter candidate set, and output the optimal process parameter combination after compensation and calibration, as well as the corresponding porosity risk distribution and parameter robustness boundary.
[0041] In this embodiment, a virtual-real deviation prediction model is trained based on the constructed virtual-real deviation sample set. This model is used to compensate and calibrate the simulated predicted porosity of each set of process parameters in the virtual optimal parameter candidate set. The calibrated porosity is equal to the original simulated predicted porosity plus the predicted deviation compensation. After calibrating all candidate parameters, the set of process parameters with the lowest calibrated porosity is selected as the final compensated and calibrated optimal process parameter combination output. At the same time, the distribution curve of porosity risk with process parameter changes is fitted by combining the calibration results of all parameters in the candidate set, as well as the parameter robustness boundary that meets the preset maximum porosity requirement, providing a reference range for parameter adjustment in actual processing.
[0042] The virtual-real deviation prediction model, trained based on the virtual-real deviation sample set, includes: Calculate the difference between the volumetric energy density corresponding to each combination of process parameters in the virtual optimal parameter candidate set and the nearest boundary in the volumetric energy density constraint interval, and use it as the volumetric energy density deviation distance feature. Using process parameter combinations, plasma arc stability index, cross-features of plasma arc stability index and volumetric energy density, volumetric energy density deviation distance features, and equipment operating status parameters as input features, and using the residuals of each physical quantity in the virtual-real deviation sample set as training labels, a gradient boosting decision tree model is trained until convergence, generating a virtual-real deviation prediction model.
[0043] In this embodiment of the application, firstly, the difference between the volumetric energy density corresponding to each combination of process parameters in the virtual optimal parameter candidate set and the nearest boundary in the volumetric energy density constraint interval is calculated as the volumetric energy density deviation distance feature. The process parameter combination, plasma arc stability index, the cross feature between plasma arc stability index and volumetric energy density, the volumetric energy density deviation distance feature, and the equipment operating status parameters are used as input features, and the residuals of each physical quantity in the virtual-real deviation sample set are used as training labels.
[0044] For example, a virtual-to-real bias prediction model is trained based on a gradient boosting decision tree model. The number of trees is set to 100, the learning rate to 0.1, the maximum tree depth to 6 layers, the minimum leaf node sample weight to 3, the column sampling ratio to 0.8, the row sampling ratio to 0.8, the L1 regularization coefficient to 0.1, and the L2 regularization coefficient to 1.0. The model training uses mean squared error as the loss function. In each iteration, the residual of the current model on the training set is calculated, and a new decision tree is trained to fit this residual. The leaf node weights of the tree are determined by minimizing the objective function. Five-fold cross-validation is used to monitor the validation set loss during training. After adding a new tree in each iteration, the root mean square error of the validation set is calculated. If the validation set loss does not decrease for 20 consecutive iterations, an early stopping mechanism is triggered to terminate training.
[0045] Furthermore, the simulated predicted porosity corresponding to each group of process parameters in the virtual optimal parameter candidate set is compensated and calibrated, and the optimal combination of process parameters after compensation and calibration is output, including: The deviation compensation amount output by the virtual-real deviation prediction model for each group of process parameters in the virtual optimal parameter candidate set is superimposed on the simulation prediction porosity output by the digital twin model for the same combination of process parameters to obtain the compensated and calibrated porosity prediction value corresponding to each combination of process parameters. The process parameter combinations in the virtual optimal parameter candidate set are sorted in ascending order according to the predicted porosity after compensation and calibration, and the process parameter combination with the lowest predicted porosity after compensation and calibration is taken as the optimal process parameter combination.
[0046] In this embodiment, the deviation compensation amount output by the virtual-real deviation prediction model for each group of process parameters in the virtual optimal parameter candidate set is superimposed on the simulated predicted porosity output by the digital twin model for the same combination of process parameters to obtain the compensated and calibrated porosity prediction value corresponding to each combination of process parameters. Then, the combination of process parameters is sorted in ascending order according to the compensated and calibrated porosity prediction value, and the combination of process parameters with the lowest compensated and calibrated porosity prediction value is taken as the optimal combination of process parameters.
[0047] For example, if the original simulation predicts a porosity of 2.3% and the porosity residual prediction value output by the virtual-real deviation prediction model is 0.21%, then the calibrated porosity prediction value is 2.51%, which is closer to the actual porosity level that will be obtained under this parameter in actual processing.
[0048] Specifically, the output porosity risk distribution includes: Based on the posterior variance of the porosity prediction model at the optimal process parameter combination, the prediction confidence interval at the optimal process parameter combination is calculated, and the porosity value corresponding to the upper bound of the prediction confidence interval is taken as the upper limit of porosity risk. In the process parameter space, the compensated and calibrated porosity prediction values corresponding to each combination of process parameters are combined with the corresponding prediction confidence intervals to form a porosity risk distribution.
[0049] In this embodiment, based on the posterior variance of the porosity prediction model at the optimal process parameter combination, the prediction confidence interval at the optimal process parameter combination is calculated. Then, the porosity value corresponding to the upper bound of the prediction confidence interval is used as the upper limit of porosity risk. In the process parameter space, the compensated and calibrated porosity prediction values corresponding to each set of process parameter combinations are combined with the corresponding prediction confidence intervals to form a porosity risk distribution.
[0050] Specifically, for any combination of process parameters, the confidence interval for porosity prediction after compensation and calibration is equal to the compensated point estimate plus the posterior standard deviation given by the original porosity prediction model multiplied by the quantile of the corresponding confidence level, thus obtaining the porosity risk interval that includes prediction uncertainty, which more intuitively reflects the risk of porosity exceeding the standard under different parameters.
[0051] Specifically, the robustness boundaries of the output parameters include: In the process parameter space, find all process parameter combinations that ensure the predicted porosity value after compensation calibration does not exceed the upper limit of porosity risk. Calculate the minimum and maximum values of the process parameter combinations in each dimension of plasma arc power, scanning speed, layer thickness, and scanning spacing. Use the interval formed by the minimum and maximum values in each dimension as the robustness boundary of the corresponding process parameter. The robustness boundaries in each dimension together constitute the robustness boundary output of the process parameter. With the optimal process parameter combination fixed, in the material thermal property parameter space composed of melting point, thermal conductivity, powder absorptivity, and powder bed apparent density, find all material thermal property parameter combinations that ensure the predicted porosity value after compensation calibration does not exceed the upper limit of porosity risk. Statistically calculate the minimum and maximum values of each material thermal property parameter combination in each dimension (melting point, thermal conductivity, powder absorptivity, and powder bed apparent density). Use the intervals formed by the minimum and maximum values in each dimension as the robustness boundaries of the corresponding material thermal property parameters. The robustness boundaries of each dimension together constitute the robustness boundary output of the material thermal property.
[0052] In this embodiment, firstly, in the process parameter space, the minimum and maximum values of the process parameter combinations that ensure the predicted porosity value after compensation calibration does not exceed the upper limit of porosity risk are statistically analyzed in each dimension of plasma arc power, scanning speed, layer thickness, and scanning spacing. The interval formed by the minimum and maximum values in each dimension is then used as the robustness boundary of the corresponding process parameter.
[0053] Furthermore, with the optimal combination of process parameters fixed, in the material thermal property parameter space consisting of melting point, thermal conductivity, powder absorptivity, and apparent density of powder bed, all combinations of material thermal property parameters that ensure the predicted porosity after compensation calibration does not exceed the upper limit of porosity risk are identified. The minimum and maximum values in each dimension are statistically analyzed, and the interval formed by the minimum and maximum values in each dimension is taken as the robustness boundary of the corresponding material thermal property parameter. Finally, the robustness boundary of process parameters and the robustness boundary of material thermal properties are output to provide process window guidance for actual production.
[0054] For example, if the robustness boundary of plasma arc power corresponding to the optimal process parameters is 1200W-1500W, it means that the porosity of the final molded part can be controlled within the allowable risk limit when processing within this power range. If the power fluctuates slightly during actual processing, the molding quality can still be guaranteed to meet the requirements without the need to readjust the parameters.
[0055] In summary, the embodiments of this application have at least the following technical effects: This application provides a parameter pre-optimization system for plasma additive manufacturing based on digital twins. First, a high-precision mapping between process parameters and forming porosity is achieved through a digital twin model. Combined with a porosity prediction model based on Gaussian process regression, rapid parameter optimization in virtual space is achieved, avoiding numerous initial trial-and-error experiments. Second, a virtual-to-real deviation prediction model is constructed by collecting real processing data from a small number of test samples. This model compensates for deviations in the prediction results of the digital twin simulation, effectively reducing the systematic deviation between simulation predictions and actual processing, and improving the accuracy of parameter optimization. Finally, for the output optimal combination of process parameters, the robustness boundaries of the corresponding process parameters and material thermophysical parameters, as well as the global porosity risk distribution, are simultaneously output. This provides guidance for quality control in actual production. When small fluctuations in parameters occur during processing, production personnel can quickly determine whether the fluctuation range is within the acceptable range, effectively reducing the risk of exceeding porosity limits.
[0056] Through the above technical solution, this application solves the technical problem that the optimal parameters obtained by simulation cannot match the actual processing results, resulting in poor accuracy of parameter optimization and lack of robust range guidance. It can quickly complete the pre-optimization of process parameters before the production of new materials, and obtain the optimal process parameters that fit the actual processing results with only a small number of trial prototypes. At the same time, it effectively reduces material and time costs, adapts to the rapid development needs of new materials and new components, and improves the first-pass yield of molded parts.
Claims
1. A plasma additive manufacturing parameter pre-optimization system based on digital twin, characterized in that, include: The initial process parameter combination acquisition module is used to generate multiple sets of process parameter combinations in the process parameter space of the plasma powder bed melting equipment, drive the digital twin model to perform powder bed melting process simulation on the multiple sets of process parameter combinations respectively, and obtain parameter-porosity simulation sample set; The process parameter extraction module is used to construct a porosity prediction model based on the parameter-porosity simulation sample set, use the porosity prediction model to perform iterative search in the process parameter space to obtain a virtual optimal parameter candidate set, and extract several sets of process parameters to be processed from it. The virtual-real deviation sample set construction module is used to drive the plasma powder bed melting equipment to print test samples according to the several sets of processing parameters, collect the real physical quantity data of the test sample printing process, compare it with the simulation prediction value output by the digital twin model, and combine the residual obtained by comparison with the corresponding process parameter combination and equipment operating status parameter combination to form a sample to construct the virtual-real deviation sample set. The optimal process parameter combination output module is used to train a virtual-real deviation prediction model based on the virtual-real deviation sample set, compensate and calibrate the simulated predicted porosity corresponding to each group of process parameters in the virtual optimal parameter candidate set, and output the optimal process parameter combination after compensation and calibration, as well as the corresponding porosity risk distribution and parameter robustness boundary.
2. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 1, characterized in that, The Latin hypercube sampling method is used to generate multiple sets of initial process parameter combinations in a multidimensional process parameter space consisting of plasma arc power, scanning speed, layer thickness and scanning spacing. The Latin hypercube sampling method divides the value range of each process parameter into intervals equal to the number of sampling points, randomly selects a sampling value in each interval, and combines the sampling values of each process parameter according to the principle of non-repetition pairing.
3. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 1, characterized in that, The steps for constructing the digital twin model include: A geometric model of the plasma powder bed melting equipment is established, taking the forming cavity, powder bed, and substrate as objects. Based on the geometric model, a heat conduction model, a molten pool dynamics model, and a porosity calculation model for the powder bed melting process are established and integrated into a digital twin model. The heat conduction model is used to calculate the spatial distribution of the temperature field of the powder bed under the action of the plasma arc. The molten pool dynamics model is used to calculate the width and depth of the molten pool. The porosity calculation model calculates the porosity based on the solidification state of the molten pool. Material property vectors are constructed based on the melting point, thermal conductivity, powder absorption rate, and apparent density of the powder bed of the new material. The material property vectors are then input into the digital twin model to complete the material parameter configuration. The digital twin model receives a combination of process parameters as simulation input and outputs the corresponding simulation predicted temperature field, simulation predicted layer height, and simulation predicted porosity.
4. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 1, characterized in that, A porosity prediction model is constructed based on the aforementioned parameter-porosity simulation sample set, including: During the simulation process of the digital twin model, the plasma arc current fluctuation sequence and the plasma arc voltage fluctuation sequence are collected in real time. The ratio of the standard deviation to the mean of the plasma arc current fluctuation sequence and the ratio of the standard deviation to the mean of the plasma arc voltage fluctuation sequence are calculated respectively. The two are then weighted and summed according to a pre-calibrated weighting coefficient to obtain the plasma arc stability index. The weighting coefficient is pre-calibrated according to the degree of influence of plasma arc current and voltage on the uniformity of heat input. Using the combination of process parameters and the plasma arc stability index as input, and the porosity corresponding to the parameter-porosity simulation sample set as the training label, a Gaussian process regression model is trained. The plasma arc stability index is introduced as a covariate into the kernel function of the Gaussian process regression model. The trained Gaussian process regression model is used as a porosity prediction model. The Gaussian process regression model outputs the predicted porosity value and prediction confidence interval for each combination of process parameters. The prediction confidence interval is calculated from the posterior variance of the Gaussian process regression model.
5. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 4, characterized in that, The porosity prediction model is used to iteratively search within the process parameter space to obtain a virtual optimal parameter candidate set, including: Before performing the iterative search, the width and depth of the molten pool corresponding to each combination of process parameters are simulated and recorded through the digital twin model, and the ratio of the width to the depth of the molten pool is calculated to obtain the width-to-depth ratio of the molten pool. Determine the lower critical value and the upper critical value of volumetric energy density when the width-to-depth ratio of the molten pool is within a preset reasonable range. Use the interval between the lower critical value and the upper critical value of volumetric energy density as the volumetric energy density constraint interval and set it as the boundary condition for iterative search. Based on the porosity prediction value and prediction confidence interval output by the porosity prediction model, a confidence upper bound acquisition function is constructed that introduces a plasma arc stability index penalty term. In each iteration, within the volumetric energy density constraint range, a combination of process parameters that maximizes the value of the confidence upper bound acquisition function is selected, driving the digital twin model to perform a new round of simulation, and updating the simulation results to the porosity prediction model; When the improvement of the predicted porosity value in consecutive iterations is lower than the preset convergence threshold, the iteration is terminated, and the combination of process parameters with the lowest current predicted porosity value is output as a virtual optimal parameter candidate set.
6. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 5, characterized in that, The actual physical quantity data during the printing process of the test sample are collected and compared with the simulation prediction values output by the digital twin model. The residuals obtained from the comparison are combined with the corresponding process parameter combinations and equipment operating status parameters to form a sample, constructing a virtual-real deviation sample set, including: During the printing process of the test sample, the plasma arc current fluctuation sequence and the plasma arc voltage fluctuation sequence were collected, and the plasma arc stability index was calculated. The actual spatial distribution of the temperature field, the layer height of the test sample, and the cross-sectional porosity of the test sample were collected during the printing process. The actual temperature field spatial distribution, the height of the test sample layer, and the porosity of the test sample cross section are compared with the simulated predicted temperature field, simulated predicted layer height, and simulated predicted porosity output by the digital twin model, and the residual between the simulated predicted value and the actual measured value of each physical quantity is calculated. Using process parameter combinations, plasma arc stability index, and equipment operating status parameters as feature dimensions, and the residuals of each physical quantity as labels, the feature dimensions and corresponding labels are combined to form a sample, thus constructing a virtual-real deviation sample set.
7. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 6, characterized in that, A virtual-real deviation prediction model is trained based on the aforementioned virtual-real deviation sample set, including: Calculate the difference between the volumetric energy density corresponding to each combination of process parameters in the virtual optimal parameter candidate set and the nearest boundary in the volumetric energy density constraint interval, and use it as the volumetric energy density deviation distance feature. Using process parameter combinations, plasma arc stability index, cross-features of plasma arc stability index and volumetric energy density, volumetric energy density deviation distance features, and equipment operating status parameters as input features, and using the residuals of each physical quantity in the virtual-real deviation sample set as training labels, a gradient boosting decision tree model is trained until convergence, generating a virtual-real deviation prediction model.
8. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 1, characterized in that, The simulated predicted porosity corresponding to each group of process parameters in the virtual optimal parameter candidate set is compensated and calibrated, and the optimal combination of process parameters after compensation and calibration is output, including: The deviation compensation amount output by the virtual-real deviation prediction model for each group of process parameters in the virtual optimal parameter candidate set is superimposed on the simulation prediction porosity output by the digital twin model for the same combination of process parameters to obtain the compensated and calibrated porosity prediction value corresponding to each combination of process parameters. The process parameter combinations in the virtual optimal parameter candidate set are sorted in ascending order according to the predicted porosity after compensation and calibration, and the process parameter combination with the lowest predicted porosity after compensation and calibration is taken as the optimal process parameter combination.
9. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 8, characterized in that, Output porosity risk distribution, including: Based on the posterior variance of the porosity prediction model at the optimal process parameter combination, the prediction confidence interval at the optimal process parameter combination is calculated, and the porosity value corresponding to the upper bound of the prediction confidence interval is taken as the upper limit of porosity risk. In the process parameter space, the compensated and calibrated porosity prediction values corresponding to each combination of process parameters are combined with the corresponding prediction confidence intervals to form a porosity risk distribution.
10. The plasma additive manufacturing parameter pre-optimization system based on digital twin according to claim 9, characterized in that, Output parameter robustness boundaries include: In the process parameter space, find all process parameter combinations that ensure the predicted porosity value after compensation calibration does not exceed the upper limit of porosity risk. Calculate the minimum and maximum values of the process parameter combinations in each dimension of plasma arc power, scanning speed, layer thickness, and scanning spacing. Use the interval formed by the minimum and maximum values in each dimension as the robustness boundary of the corresponding process parameter. The robustness boundaries in each dimension together constitute the robustness boundary output of the process parameter. With the optimal process parameter combination fixed, in the material thermal property parameter space composed of melting point, thermal conductivity, powder absorptivity, and powder bed apparent density, find all material thermal property parameter combinations that ensure the predicted porosity value after compensation calibration does not exceed the upper limit of porosity risk. Statistically calculate the minimum and maximum values of each material thermal property parameter combination in each dimension (melting point, thermal conductivity, powder absorptivity, and powder bed apparent density). Use the intervals formed by the minimum and maximum values in each dimension as the robustness boundaries of the corresponding material thermal property parameters. The robustness boundaries of each dimension together constitute the robustness boundary output of the material thermal property.
