Search

CN-122021252-A - LPBF ventilation steel process parameter optimization method based on deep learning

CN122021252ACN 122021252 ACN122021252 ACN 122021252ACN-122021252-A

Abstract

The invention belongs to the technical field of additive manufacturing process optimization and material performance regulation and control, and relates to a LPBF ventilation steel process parameter optimization method based on deep learning. Generating physical experiment parameters and simulation parameters by using Latin hypercube design. Correcting simulation parameters through a small amount of experimental data to ensure that the deviation between simulation and experimental data is less than or equal to 8%, setting the porosity, the pore size and the air permeability coefficient of the air permeable steel according to requirements, setting the constraint range of technological parameters according to LPBF equipment, obtaining a Pareto solution through an NSGA-III algorithm, and selecting the solution with the highest score as the optimal solution through weighted scoring. And carrying out closed loop verification by adopting the optimal parameters to prepare the breathable steel sample test target performance, wherein the relative error of each index is less than or equal to +/-5%.

Inventors

  • LI JIAMING
  • Chai Pengtao
  • CHENG YUHANG
  • ZHANG GUOLIANG

Assignees

  • 上海镭镆科技有限公司

Dates

Publication Date
20260512
Application Date
20251226

Claims (10)

  1. 1. The LPBF ventilation steel technological parameter optimization method based on deep learning is characterized by comprising the following steps of: Step S1, LPBF, collecting and preprocessing process data and simulation physical field data; s2, constructing a multi-target prediction model based on deep learning; The input layer is characterized by laser power, scanning speed, scanning interval, layer thickness, bath peak temperature, cooling rate, thermal stress peak value and strain distribution, the feature extraction layer is provided with 3X 3, 5X 5 and 3X 3 convolution cores, the output channel number is 64, 128 and 256 in sequence, and the multi-head attention layer is connected through the residual error to avoid gradient disappearance, the porosity, aperture size and permeability coefficient feature vector are output, and the normalized predicted value is output by adopting a Sigmoid activation function; Determining initial super parameters by adopting a Bayesian optimization algorithm, adjusting through cross validation, adopting AdamW optimizers, adopting a loss function as weighted cross entropy loss, inhibiting overfitting through L2 regularization, stopping training when the relative errors of porosity, aperture size and air permeability coefficient prediction on a validation set are less than or equal to +/-5%, and storing an optimal model; Step S3, solving the technological parameters based on a multi-objective optimization algorithm; Setting a process parameter boundary based on LPBF equipment performance and material characteristics; defining a target range, constructing a judgment matrix by adopting an analytic hierarchy process, wherein a target layer is 'multi-performance optimal', a criterion layer is porosity, aperture size and air permeability coefficient, and a weight vector is determined by consistency test; The method comprises the steps of performing 1000 times of sampling on a process parameter space based on Latin hypercube sampling, removing obvious unreasonable parameter combinations, inputting effective parameter combinations into a prediction model which is completed by training, outputting a prediction performance index, and optimizing by adopting an NSGA-III algorithm to obtain a Pareto front solution; Calculating the weighted score of each Pareto solution according to the weight vector determined by the AHP, selecting the solution with the highest score as the unique optimal technological parameter combination, and outputting a specific parameter value; Step S4, closed loop feedback iteration; Inputting optimal technological parameters into LPBF equipment for printing experiments, testing actual performance indexes, if the relative error between the actual performance and a predicted value is less than or equal to +/-5%, combining corresponding parameters into a final scheme, if the error is > +/-5%, supplementing corresponding parameter combination data to an original data set, retraining a model, and repeating the step S3 until the error meets the requirement to form closed loop optimization.
  2. 2. The method for optimizing parameters of a deep learning-based LPBF gas permeable steel process according to claim 1, comprising, in step S1: Selecting laser power, scanning speed, scanning interval and layer thickness in LPBF process core parameters as basic input parameters, adopting Latin hypercube design experimental scheme with uniform parameter space coverage and strong sample representativeness, and setting a parameter range; Adopting thermal-stress coupling simulation software and generating multi-physical field data in batches based on technological parameters, wherein the multi-physical field data comprise the peak temperature of a molten pool, cooling rate, thermal stress peak value and strain distribution, and the initial parameters of a simulation model refer to the thermal conductivity, specific heat capacity and thermal expansion coefficient of a material; developing a calibration experiment to verify the validity of the simulation data; preprocessing data, including removing abnormal data, normalizing input characteristics and normalizing output performance indexes; adopting an SMOTE algorithm to expand small sample data, performing physical consistency check on a synthesized sample, removing redundant samples with the similarity of more than or equal to 95% with an original sample, adding random noise to standardized input and output data, and improving the generalization capability of a model; And finally forming the labeling data set.
  3. 3. The method for optimizing parameters of LPBF gas permeable steel process based on deep learning as set forth in claim 2, wherein the ranges of laser power, scan speed, scan pitch, layer thickness parameters are set according to LPBF equipment.
  4. 4. The method for optimizing LPBF ventilation steel process parameters based on deep learning of claim 2, wherein verifying the validity of the simulation data comprises selecting 20 groups of typical process parameter combinations for LPBF printing experiments, measuring porosity by using an Archimedes drainage method, obtaining pore size by image analysis by a scanning optical microscope, and testing ventilation coefficient by using a gas permeation test device.
  5. 5. The method for optimizing the parameters of LPBF ventilated steel based on deep learning as claimed in claim 2, wherein abnormal data caused by simulation convergence failure and experimental equipment errors are removed by adopting a 3 sigma criterion, the input characteristics are normalized by adopting a Z-score, the output performance index is mapped to a [0,1] interval by adopting a Min-Max normalization, and the influence of dimension differences on model training is eliminated.
  6. 6. The method for optimizing parameters of LPBF gas permeable steel process based on deep learning as set forth in claim 2, wherein the labeling dataset comprises physical experiment data, calibrated simulation data, SMOTE synthesis data and noise enhancement data, and the data format is one-to-one correspondence of input feature vectors and output performance vectors.
  7. 7. The method for optimizing parameters of LPBF permeable steel based on deep learning as claimed in claim 1, wherein the attention head number of the multi-head attention layer is 8, and the weight attenuation coefficient of the adamw optimizer is 0.01.
  8. 8. The method for optimizing parameters of LPBF gas permeable steel process based on deep learning as set forth in claim 1, wherein the range of porosity, pore size, and gas permeability coefficient can be customized by a user.
  9. 9. The method for optimizing parameters of LPBF permeable steel process based on deep learning according to claim 1, wherein the NSGA-III algorithm has a population size of 100 and a number of iterations of 200.
  10. 10. The deep learning-based LPBF gas permeable steel process parameter optimization method of claim 1, wherein the weighted score is proportional to the product of the performance prediction value and its corresponding weight.

Description

LPBF ventilation steel process parameter optimization method based on deep learning Technical Field The invention relates to the technical field of additive manufacturing process optimization and material performance regulation, in particular to a LPBF (Laser Powder Bed Fusion, laser powder bed melting) ventilation steel process parameter optimization method based on deep learning, which is used for integrating process parameters and multi-physical field data, and is particularly suitable for accurate parameter regulation of customized ventilation steel prepared by LPBF (or SLM, selective Laser Melting selective laser melting) process. Background The breathable steel is used as a special metal material with a large number of communicated or semi-communicated pores inside, has the characteristics of good breathability, low density, high specific strength and the like, and is widely applied to the fields of biomedical, aerospace, injection mold exhaust systems and the like, and the main performance indexes of the breathable steel, such as porosity, pore size, breathability coefficient and the like, directly determine the application effect. At present, the preparation process of the breathable steel mainly comprises foaming agent forming, powder metallurgy, spark plasma sintering and the like, but the traditional processes have obvious limitations that parts with complex shapes are difficult to manufacture, pore structures are difficult to accurately control, and popularization and application of the breathable steel are severely limited. The LPBF process can realize the preparation of near-net-shaped metal parts with complex structures and excellent mechanical properties by virtue of the characteristics of rapid melting and layered printing, can realize the connectivity and controllability of pore structures by regulating and controlling process parameters, and provides a new path for the preparation of breathable steel. However, LPBF processes are used for preparing the breathable steel, wherein the core technical pain points are 1. The process parameters (laser power, scanning speed and the like) have a strong coupling nonlinear relation with multiple physical fields and multiple performance indexes, hundreds of physical experiments are required to be carried out by a traditional trial-and-error method, the cost of a single experiment is over ten thousands yuan, the optimization period is 3-6 months, the full parameter space cannot be covered, 2. The influence rule of the multiple physical fields on the pore structure is difficult to quantify, the parameter optimization blindness is high, the problem of poor multi-performance cooperativity such as 'the porosity reaches the standard but the air permeability coefficient is insufficient' is frequently caused, and 3. The closed loop scheme of 'data acquisition-model prediction-parameter optimization-feedback iteration' is not formed in the prior art, and the customization requirements of different scenes cannot be met. The deep learning technology has strong nonlinear mapping and multi-objective fitting capability, and has potential in additive manufacturing process optimization. However, the prior art has not combined deep learning with multi-objective optimization depth of LPBF process for preparing the breathable steel, and especially lacks a closed-loop optimization scheme for integrating multi-physical-field data and supporting real-time correction, and cannot meet the requirements of customization, high precision and high efficiency of the breathable steel. Therefore, it is needed to provide a LPBF process parameter multi-objective optimization method based on deep learning, which solves the above-mentioned technical pain. It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the invention and thus may include information that does not form the prior art that is already known to those of ordinary skill in the art. Disclosure of Invention Aiming at the problems of low parameter optimization efficiency, high trial and error cost, poor multi-performance cooperativity and lack of customized closed-loop scheme in the conventional LPBF process for preparing the breathable steel, the invention provides a LPBF breathable steel process parameter optimization method based on deep learning, which is characterized in that a high-precision multi-objective prediction model is constructed by integrating process parameters and multi-physical-field data, a closed-loop system of data acquisition, model training, parameter optimization and feedback iteration is formed, the trial and error cost is reduced, and the qualified rate of finished products and the generalization capability of parameter optimization are improved. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: A LPBF ventilated steel process parameter optimization method based on