Adaptive layering co-optimization method for metal additive 3D printing

Abstract

This invention discloses an adaptive layered collaborative optimization method for metal additive 3D printing, relating to the field of 3D printing. The method includes: simultaneously acquiring real-time morphology, temperature field distribution, and dynamic fluctuation parameters of the molten pool during the metal additive 3D printing process using a high-speed vision acquisition module and an infrared temperature measurement module; performing noise reduction preprocessing on the acquired multi-dimensional raw data of the molten pool, simultaneously extracting core features, and performing standardized quantization on the core features to generate a molten pool state vector. This invention can accurately perceive the molten pool state, quantify core features, adaptively generate layer thickness and matching process parameters, select the optimal solution through adaptation judgment, and dynamically adjust parameters throughout the process, thereby improving deposition stability and forming consistency, reducing printing defects, adapting to various metal materials, improving part precision and quality, reducing scrap rate, and optimizing printing efficiency.

Description

Technical Field

[0001] This invention relates to the field of 3D printing technology, specifically to an adaptive layered collaborative optimization method for metal additive 3D printing. Background Technology

[0002] Metal additive 3D printing technology, with its ability to integrally form complex components, has become a core process in high-end manufacturing fields such as aerospace and medical devices. Layer slicing, as a crucial link connecting the 3D model and the actual printing, directly determines the forming accuracy, surface quality, and production efficiency of the part. While traditional uniform contour slicing methods are computationally simple, they cannot balance accuracy and efficiency, and are prone to producing noticeable stair-step effects on complex curved surfaces.

[0003] Chinese invention patent application No. 201710972487.3 discloses an adaptive layering processing method, system and additive manufacturing equipment for additive manufacturing. This application aims to solve the problems of "large error in uniform layering and low printing efficiency".

[0004] However, existing technologies only optimize the layer thickness from a geometric perspective, without achieving coordinated adjustment of the layering scheme and printing process parameters, and thus cannot match the actual deposition state of the molten pool changing in real time during the printing process.

[0005] To address this, we propose an adaptive hierarchical collaborative optimization method for metal additive 3D printing. Summary of the Invention

[0006] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an adaptive layered collaborative optimization method for metal additive 3D printing, which can effectively solve the problems of the existing technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses an adaptive hierarchical collaborative optimization method for metal additive 3D printing, comprising: The real-time morphology, temperature field distribution, and dynamic fluctuation parameters of the molten pool during the metal additive 3D printing process are simultaneously acquired by a high-speed vision acquisition module and an infrared temperature measurement module. The acquired multi-dimensional raw data of the molten pool undergoes noise reduction preprocessing, core features are extracted synchronously, and standardized quantization is performed on the core features to generate a molten pool state vector. Based on the molten pool state vector, several candidate layer thickness schemes with gradient distribution are generated within a pre-defined layer thickness range. For each candidate layer thickness scheme, corresponding combinations of process parameters including laser power, scanning speed, powder feeding rate, and scanning path planning are synchronously generated. A collaborative adaptability judgment matrix including deposition stability evaluation indicators and forming consistency evaluation indicators is established. The process parameter combinations of each candidate scheme are substituted into the matrix to calculate the adaptability score. The adaptability score of each combination is compared with a preset qualified threshold to complete the item-by-item adaptability verification. The combination that passes the verification and has the highest adaptability score is selected and sent to the motion controller of the printing equipment. The molten pool state is continuously acquired throughout the printing process to trigger dynamic parameter iteration. The core features include the width of the molten pool, the depth of the molten pool, the temperature gradient, and the fluctuation frequency.

[0008] Furthermore, when performing noise reduction preprocessing on the collected multi-dimensional raw data of the molten pool, the real-time morphology of the molten pool, the temperature field distribution of the molten pool, and the dynamic fluctuation parameters of the molten pool are time-stamped and aligned. Subpixel-level spatiotemporal registration is performed on the timestamp-aligned melt pool morphology sequence and temperature field sequence to generate a three-dimensional spatiotemporally correlated melt pool data cube. Adaptive median filtering based on neighborhood difference is then performed on the melt pool data cube to obtain denoised morphology and temperature field data. Adaptive threshold denoising based on wavelet transform is performed on the timestamp-aligned dynamic fluctuation parameters of the molten pool to obtain the denoised dynamic fluctuation parameters.

[0009] Furthermore, in the stage of performing standardized quantization on the core features to generate the melt pool state vector, the melt pool width, temperature gradient and fluctuation frequency are simultaneously subjected to linear interval mapping standardization processing to map them to a preset numerical range. During the processing, outliers that exceed the preset original data range are truncated. The melting depth is subjected to nonlinear quantization based on the radial gradient distribution of the temperature field to obtain the standardized melting depth characteristic value; The standardized molten pool width, molten depth, temperature gradient, and fluctuation frequency are spliced ​​together in a preset order to generate a molten pool state vector with fixed dimensions. The nonlinear quantization of melting depth follows the following rules: ; In the formula: These are the standardized melt depth characteristic values; This is a preset reference melting depth value; Preset quantization coefficients; As the center of the molten pool The maximum temperature gradient within a circular region with radius . The average temperature gradient of the entire molten pool; This is the preset feature radius of the molten pool.

[0010] Furthermore, within a pre-defined range of layer thickness values, several groups of candidate layer thickness schemes with a gradient distribution are generated: The melt depth characteristic value in the melt pool state vector is used as the reference value. Generate a base value within a preset layer thickness range. Several groups of candidate layer thicknesses centered on a logarithmic gradient distribution; Candidate values ​​that exceed the preset range of layer thickness are removed, and the remaining candidate values ​​are sorted in ascending order to obtain the final set of candidate layer thickness schemes.

[0011] Furthermore, when simultaneously generating the corresponding combination of process parameters for laser power, scanning speed, powder feeding rate, and scanning path planning for each set of candidate layer thickness schemes, the following applies: The basic adjustment coefficient of laser power is determined based on the ratio of candidate layer thickness to melt depth characteristic value; The correction factor for the scanning speed is determined based on the difference between the molten pool fluctuation frequency and the preset stable fluctuation frequency. The matching value of the powder feeding rate is determined based on the product of the laser power and the scanning speed. The scan line spacing and overlap rate of the scan path are determined based on the molten pool width and the candidate layer thickness. Among them, the basic adjustment coefficient of laser power ; In the formula: h is the candidate layer thickness, The standardized melt depth characteristic value. To preset the power index, This is the preset power attenuation coefficient.

[0012] Furthermore, the generation of process parameter combinations for scan path planning follows the following: By employing a partitioned interleaved scanning strategy, the current printing layer is divided into several regular polygonal partitions of equal area, with the circumcircle radius of each partition being... , represents the preset basic circumcircle radius, k represents the preset partition size adjustment coefficient, and f represents the molten pool fluctuation frequency; The scanning directions of adjacent layers are staggered at a preset angle, and the scanning starting points of adjacent partitions within the same layer are staggered by a preset distance. Each zone uses a continuous reciprocating scanning method, while the zones use a skip scanning method. During the skipping process, the laser power is reduced to a preset standby power.

[0013] Furthermore, the calculation method for sedimentary stability evaluation indicators is as follows: The coefficients of variation of the melt pool width, melt depth and temperature gradient within a consecutive preset number of frames are calculated respectively. The three coefficients of variation are substituted into the preset deposition stability evaluation function to obtain the deposition stability evaluation index value. Among them, the sedimentation stability evaluation function is the reciprocal of the geometric mean of the three coefficients of variation. The larger the sedimentation stability evaluation index value, the more stable the sedimentation process.

[0014] Furthermore, the calculation method for the forming consistency evaluation index is as follows: Calculate the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector; Calculate the normalized Euclidean distance between the current layer candidate process parameter combination and the previous layer actual process parameter combination, with normalization based on the preset value range of each process parameter; Substitute the cosine similarity and normalized Euclidean distance into the preset forming consistency evaluation function. C represents the forming consistency evaluation index value. This represents the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector. The normalized Euclidean distance between the current candidate process parameter combination and the previous actual process parameter combination is used to obtain the forming consistency evaluation index value. Among them, the larger the value of the forming consistency evaluation index, the better the forming consistency between layers.

[0015] Furthermore, a synergistic adaptability judgment matrix is ​​established that includes deposition stability evaluation index and formation consistency evaluation index. The process parameter combinations of each candidate scheme are substituted into the matrix to calculate the adaptability score. A two-dimensional synergistic adaptability judgment matrix is ​​constructed, with the rows of the matrix corresponding to the preset level range of the deposition stability evaluation index and the columns corresponding to the preset level range of the formation consistency evaluation index. The deposition stability evaluation index value and the formation consistency evaluation index value of each candidate scheme are mapped to the corresponding level interval position of the judgment matrix respectively. Based on the preset matrix weight distribution, calculate the suitability score of each candidate solution; The matrix weight distribution is dynamically adjusted as the number of printing layers increases, and the closer to the printing completion stage, the higher the weight of the forming consistency evaluation index.

[0016] Furthermore, during the continuous acquisition of the molten pool state throughout the printing process to trigger dynamic parameter iteration, the molten pool state is acquired once every preset time interval to generate the current molten pool state vector. The Mahalanobis distance between the current molten pool state vector and the molten pool state vector at the time of the last parameter update is calculated as the state difference degree. When the state difference degree exceeds the preset difference threshold, the printing process is paused after the current scan segment is completed, and the candidate layer thickness scheme generation, process parameter matching, and adaptability verification process are re-executed to obtain the updated optimal process parameter combination and send it to the motion controller, and then the printing process is resumed.

[0017] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention simultaneously acquires multi-dimensional data of the molten pool through high-speed vision and infrared temperature measurement modules, and performs precise noise reduction preprocessing and standardized quantization. It can obtain a stable and reliable molten pool state vector in real time, providing accurate data support for the optimization of layer thickness and process parameters. Based on the melt depth, it generates candidate layer thickness schemes with an equal logarithmic gradient distribution, which can reasonably select layer parameters according to the actual molten pool state during the printing process, avoiding the problem of mismatch between thickness settings and molten pool characteristics. It synchronously matches laser power, scanning speed, powder feeding rate and scanning path planning parameters to achieve synergistic adaptation between layer thickness and core process parameters. Furthermore, it uses a synergistic adaptability judgment matrix constructed with dual indicators of deposition stability and forming consistency to screen the optimal parameter combination, which can effectively improve the deposition stability and interlayer forming consistency during the printing process, and reduce the occurrence rate of defects such as over-melting and incomplete melting. In addition, it continuously acquires the molten pool state and triggers dynamic parameter iteration throughout the printing process, which can respond to state changes in a timely manner during the printing process, ensuring the stability of the entire metal additive 3D printing process and the forming quality. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 This is a flowchart illustrating an adaptive hierarchical collaborative optimization method for metal additive 3D printing. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0021] The present invention will be further described below with reference to embodiments.

[0022] Example: This embodiment presents an adaptive layered collaborative optimization method for metal additive 3D printing, such as... Figure 1 As shown, it includes: The real-time morphology, temperature field distribution, and dynamic fluctuation parameters of the molten pool during the metal additive 3D printing process are collected simultaneously by a high-speed vision acquisition module and an infrared temperature measurement module. The high-speed vision acquisition module uses an industrial high-speed camera with a frame rate ≥2000fps, resolution ≥1280×1024, and lens focal length of 8-25mm. It is installed on the outside of the forming chamber of the printing equipment, at a horizontal distance of 150-300mm from the molten pool and a downward angle of 30°-60°. The infrared temperature measurement module uses an infrared thermal imager with a temperature measurement range of 500-3000℃, a response wavelength of 1.0-1.7μm, and a sampling frequency ≥1kHz. It is coaxially installed with the high-speed vision acquisition module. Both types of modules use hardware-triggered synchronous acquisition, with the laser output signal of the printing equipment as the synchronous trigger source, to achieve time-synchronous acquisition of molten pool morphology, temperature field, and dynamic fluctuation parameters. The collected multi-dimensional raw data of the molten pool is preprocessed for denoising, and core features are extracted simultaneously. The core features are then standardized and quantized to generate the molten pool state vector. When performing noise reduction preprocessing on the collected multi-dimensional raw data of the molten pool, the real-time morphology of the molten pool, the temperature field distribution of the molten pool, and the dynamic fluctuation parameters of the molten pool are time-stamped and aligned. Among them, the real-time morphology data of the molten pool includes the molten pool outline, molten pool area, and molten pool aspect ratio; the temperature field distribution data of the molten pool includes the spatial distribution of the temperature field, the peak temperature of the molten pool, and the cooling rate of the molten pool; and the dynamic fluctuation parameters of the molten pool include the fluctuation frequency, fluctuation amplitude, and fluctuation period of the molten pool. Subpixel-level spatiotemporal registration is performed on the timestamp-aligned melt pool morphology sequence and temperature field sequence to generate a three-dimensional spatiotemporally correlated melt pool data cube. Adaptive median filtering based on neighborhood difference is then performed on the melt pool data cube to obtain denoised morphology and temperature field data. An adaptive threshold denoising process based on wavelet transform is performed on the timestamp-aligned dynamic fluctuation parameters of the molten pool to obtain the denoised dynamic fluctuation parameters. The adaptive threshold denoising based on wavelet transform uses the db4 wavelet basis function to perform three-level wavelet decomposition on the dynamic fluctuation parameter sequence of the molten pool. Threshold denoising is performed only on the high-frequency detail components, while the low-frequency approximate components remain unchanged. After denoising, the clean dynamic fluctuation parameters are reconstructed. The kernel size of the adaptive median filter is dynamically adjusted according to the following formula: ; This formula uses a preset basic filter kernel side length as a benchmark, and is based on the ratio of the standard deviation of the current pixel's neighborhood melt pool data to the global standard deviation. This represents the sample standard deviation of the grayscale values ​​of the melt pool morphology or the temperature values ​​of the temperature field within a 3×3 neighborhood of the current pixel. It is the overall standard deviation of the shape grayscale value or temperature field temperature value of all pixels within the effective area of ​​the melt pool in the current frame (excluding the background noise area). Both are calculated using standard statistical formulas, and the effective area of ​​the melt pool is extracted by the threshold segmentation method. By using the hyperbolic tangent function to achieve dynamic adjustment, it can adapt to the noise distribution characteristics of the molten pool image. While preserving the minimum identifiable features of the molten pool edge, it can efficiently filter out mixed noise and improve the processing accuracy of molten pool morphology and temperature field data. The threshold for adaptive threshold denoising is calculated using the following formula: ; This formula is calculated by combining the basic threshold coefficient, the length of the dynamic fluctuation parameter sequence, the preset noise standard deviation, and the wavelet coefficient standard deviation. It can adaptively set the denoising intensity according to the signal characteristics of the molten pool fluctuation parameters. While filtering out most of the Gaussian white noise in the system, it completely preserves the inherent fluctuation characteristics of the molten pool and ensures the authenticity of the dynamic fluctuation parameters. In the formula: The edge length of the filter kernel at the current pixel position; The preset basic filter kernel side length; is the preset kernel size adjustment factor; is the standard deviation of the molten pool data within the 3×3 neighborhood of the current pixel; The global standard deviation of the current frame's molten pool data; The wavelet threshold for the current decomposition level; This is the preset basic threshold coefficient; The length of the dynamic fluctuation parameter sequence; The preset noise standard deviation; The standard deviation of the wavelet coefficients at the current decomposition level; in, The value is determined based on the pixel resolution of the high-speed vision acquisition module and the average projection size of the target metal molten pool on the imaging plane. The value is the smallest odd integer that can completely cover the smallest identifiable feature of the molten pool edge. The value is determined based on the typical fluctuation range of salt-and-pepper noise and Gaussian noise in the molten pool image during the additive manufacturing process of the target metal material. The value is a positive real number that ensures that the maximum dynamic adjustment of the filter kernel side length does not exceed half of the preset basic filter kernel side length. The value is determined based on the original signal-to-noise ratio of the dynamic fluctuation parameters of the molten pool and the signal distortion allowed by the process. The value is a dimensionless coefficient that can filter out more than 90% of the Gaussian white noise of the system while fully preserving the inherent fluctuation characteristics of the molten pool. In the stage of standardizing and quantizing the core features to generate the molten pool state vector, the molten pool width, temperature gradient and fluctuation frequency are simultaneously subjected to linear interval mapping standardization processing and mapped to a preset numerical range. The preset numerical range is uniformly set to [0,1], that is, Tmin=0 and Tmax=1. This range is applicable to the standardization processing of molten pool features of all metal additive 3D printing, which can eliminate the differences in the dimensions of different features and ensure the numerical uniformity of the molten pool state vector. During the processing, outliers that exceed the preset original data range are truncated. The standardization process for linear interval mapping is implemented using the following formula: ; In the formula: These are the original eigenvalues; These are the standardized eigenvalues; , These are the preset minimum and maximum values ​​of the original data for this feature; , These are the lower and upper limits of the preset numerical range, respectively; The above formula achieves feature standardization through linear transformation between the original feature extrema and the target interval extrema, truncates outliers, unifies the numerical scale of features such as melt pool width and temperature gradient, eliminates the influence of dimensional differences, and provides a standardized quantitative basis for the generation of melt pool state vectors. When the original eigenvalue At that time, take When the original eigenvalue At that time, take ; The melting depth is subjected to nonlinear quantization based on the radial gradient distribution of the temperature field to obtain the standardized melting depth characteristic value; The standardized molten pool width, molten depth, temperature gradient, and fluctuation frequency are spliced ​​together in a preset order to generate a molten pool state vector with fixed dimensions. The nonlinear quantization of melting depth follows the following rules: ; In the formula: These are the standardized melt depth characteristic values; This is a preset reference melting depth value; Preset quantization coefficients; As the center of the molten pool The maximum temperature gradient within a circular region with radius . The average temperature gradient of the entire molten pool; The preset molten pool feature radius; The above formula takes the reference melting depth as the benchmark, combines the ratio of the maximum temperature gradient to the average temperature gradient in the characteristic region of the molten pool, and completes the melting depth quantification through an exponential function. It fits the thermophysical relationship between the melting depth and temperature gradient of metal materials and more accurately reflects the actual forming state of the melting depth. in, The determination is based on the melting point and thermal conductivity of the printing material, as well as the equipment's reference laser power and reference scanning speed. Specifically, it is the measured steady-state melting depth of the printing material under the reference process parameters. The width of the printing material is determined based on the steady-state molten pool width under the baseline process parameters, specifically as a preset ratio of the steady-state molten pool width. The value is determined based on the ratio of the thermal diffusivity to the specific heat capacity of the printing material, and is taken within a preset coefficient range. Based on the molten pool state vector, several groups of candidate layer thickness schemes with gradient distribution are generated within a pre-defined range of layer thickness values. The stage of generating several groups of candidate layer thickness schemes with gradient distribution within a pre-defined range of layer thickness values: The melt depth characteristic value in the melt pool state vector is used as the reference value. Generate a base value within a preset layer thickness range. Several groups of candidate layer thicknesses centered on a logarithmic gradient distribution; Candidate values ​​that exceed the preset range of layer thickness are removed, and the remaining candidate values ​​are sorted in ascending order to obtain the final set of candidate layer thickness schemes. Among them, the candidate layer thicknesses, centered on the baseline value and exhibiting an equal logarithmic gradient distribution, follow the following rules during generation: Determine the total number N of candidate layer thicknesses based on the baseline value. The center-symmetric distribution is directed towards directions greater than the reference value and less than the reference value. The number of candidate values ​​generated for the direction; If the preset total quantity N is odd, then the baseline value As a candidate value itself, (N-1) / 2 candidate values ​​are generated on both the top and bottom sides; if the preset total number N is even, then N / 2 candidate values ​​are generated on both the top and bottom sides. Based on the upper limit of the preset layer thickness range and lower limit Calculate the logarithmic step size ; This formula is calculated by the ratio of the natural logarithmic difference between the upper and lower limits of the layer thickness range to the number of candidates, so that the candidate layer thickness has an equal logarithmic gradient distribution, taking into account the sampling uniformity of different thickness ranges, and adapting to the process value constraints of metal additive printing. Based on the benchmark value Starting from the reference value, candidate layer thicknesses are generated sequentially in the directions greater than and less than the reference value, with logarithmic steps. The thickness of the i-th candidate layer greater than the reference value is calculated using the following formula: ; This formula starts with the melt depth characteristic benchmark value and generates thick layer candidate values ​​through exponential operation with logarithmic step size. It conforms to the physical law of thick layer deposition in the melt pool, avoids process adaptation deviation caused by linear gradient, and ensures the stability of thick layer deposition. The thickness of the i-th candidate layer that is less than the baseline value is calculated using the following formula: ; This formula starts with the melt depth characteristic benchmark value and generates thin layer candidate values ​​through exponential operation with a negative logarithmic step size, thereby improving the sampling accuracy of thin layer intervals and matching the process requirements of the minimum depositable thickness of printing equipment. Continue until the number of candidate layer thicknesses generated reaches the preset total number N; Optionally, the total number of candidate layer thicknesses is preset to N=7, which is an odd number, and candidate values ​​with an equal logarithmic gradient are generated with the melt depth characteristic value h0 as the center; the preset safety factor is 0.9, that is, the upper limit of the layer thickness range is the maximum melt depth of the current printing material × 0.9, to avoid incomplete melting defects in thick layer deposition; The lower limit of the preset layer thickness range is the preset minimum depositable thickness jointly calibrated by the motion control accuracy of the printing equipment, the minimum powder feeding rate, and the minimum laser output power. The upper limit is the maximum melting depth of the current printing material at the preset maximum laser power and the preset minimum scanning speed multiplied by the preset safety factor. The preset safety factor is a preset constant less than 1. For each group of candidate layer thickness schemes, the corresponding combination of process parameters such as laser power, scanning speed, powder feeding rate and scanning path planning is generated simultaneously. For each set of candidate layer thickness schemes, when simultaneously generating the corresponding combination of process parameters for laser power, scanning speed, powder feeding rate, and scanning path planning, the following rules apply: The basic adjustment coefficient of laser power is determined based on the ratio of candidate layer thickness to melt depth characteristic value. In the formula: h is the candidate layer thickness, The standardized melt depth characteristic value. To preset the power index, The preset power attenuation coefficient; This formula combines the power operation of the ratio of layer thickness to melting depth, and corrects the ratio deviation through exponential decay. It can dynamically adjust the laser power to avoid defects such as over-melting and incomplete melting of the molten pool and maintain the stability of the deposition process. The correction factor for the scanning speed is determined based on the difference between the molten pool fluctuation frequency and the preset stable fluctuation frequency. In the formula: This represents the current frequency of molten pool fluctuations. To preset a stable fluctuation frequency, This is the preset speed correction coefficient; This formula is calculated using an exponential function based on the relative deviation between the molten pool fluctuation frequency and the preset stable frequency. It can quickly respond to the molten pool fluctuation state and adjust the scanning speed to ensure the stability of dynamic molten pool forming. The matching value for the powder feeding rate is determined based on the product of laser power and scanning speed. In the formula: This is a preset correction factor for the combined effects of material density and deposition efficiency. The width of the molten pool This is the preset spacing adjustment coefficient. Preset base scan speed; This formula integrates multiple parameters such as materials, processes, and molten pool characteristics, and combines the ratio of layer thickness to melting depth for compensation calculation, so as to accurately match the powder feeding rate with laser energy and scanning speed, ensuring that the deposition supply and fusion efficiency are consistent. Based on the molten pool width and candidate layer thickness, determine the scan line spacing and overlap ratio of the scan path. In the formula: This refers to the scan line spacing. This refers to the overlap rate of the scan lines. The above formula is based on the width of the molten pool and dynamically adjusts the spacing by combining the layer thickness and the ratio of molten depth to adapt to the current layered molten pool forming characteristics, ensuring the uniformity of deposition within the layer and the density of interlayer bonding. The overlap rate is calculated by the ratio of the difference between the scan line spacing and the molten pool width, which precisely controls the tightness of the molten pool overlap and effectively improves the surface forming quality of the printed parts. in, By conducting a standard single-pass melting depth calibration experiment on the target metal material, the steady-state melting depth value under different laser powers was measured. The power law relationship curve between melting depth and laser power was fitted using the least squares method, and the resulting power exponent is the preset power exponent of the material. In a gradient experiment where the ratio of layer thickness to melting depth deviates from 1, the maximum laser power adjustment range that ensures no overmelting or incomplete melting defects in the molten pool is recorded. The exponential decay curve of the power adjustment coefficient with the deviation of the ratio is fitted, and the resulting decay constant is the preset power decay coefficient. By gradually changing the scanning speed and simultaneously collecting the molten pool fluctuation frequency under the condition of fixed laser power and powder feeding rate, the exponential relationship curve between the scanning speed correction amount and the relative deviation of the molten pool fluctuation frequency is fitted, and the obtained proportional coefficient is the preset speed correction coefficient. By measuring the actual deposition mass per unit length of the scanning path through a standard single-pass deposition experiment, the ratio of this mass to the theoretical deposition mass (material density × cross-sectional area × length) is calculated. Multiplying this ratio by the material density yields the comprehensive correction factor. By conducting multi-pass lap forming experiments at different layer thicknesses, the optimal lap ratio is determined by using surface roughness and interlayer bonding strength as evaluation indicators. The linear relationship between the scan line spacing and the melt pool width and the layer thickness-melt depth ratio is then calculated, and the resulting slope is the preset spacing adjustment coefficient. The generation of process parameter combinations for scan path planning follows the following rules: By employing a partitioned interleaved scanning strategy, the current printing layer is divided into several regular polygonal partitions of equal area, with the circumcircle radius of each partition being... , represents the preset basic circumcircle radius, k represents the preset partition size adjustment coefficient, and f represents the molten pool fluctuation frequency; This formula uses the radius of the basic circumscribed circle as a reference, and combines the partition size adjustment coefficient and the dynamic correction of the molten pool fluctuation frequency to adapt to the thermal diffusion and laser power density characteristics of different materials, and optimize the forming effect of partitioned staggered scanning. The scanning directions of adjacent layers are staggered at a preset angle, and the scanning starting points of adjacent partitions within the same layer are staggered by a preset distance. The scanning directions of adjacent layers are preset to intersect at an angle of 90°; the scanning starting points of adjacent partitions within the same layer are staggered by 1 / 2 of the scanning line spacing; the number of regular polygon partitions in each layer is determined according to the area of ​​the printed layer, with 4 to 9 partitions per unit area per cm², ensuring that the partition areas are equal and adapted to the molten pool forming characteristics. Each zone uses a continuous reciprocating scanning method, while the zones use a skip scanning method. During the skipping process, the laser power is reduced to a preset standby power. The preset partition size adjustment coefficient k ranges from 0.05 to 0.5, and its value is determined based on the thermal conductivity of the printing material and the laser power density. When the thermal conductivity of the printing material is greater than 200 W / (m·K) or the laser power density is greater than When k is 0.2~0.5; when the thermal conductivity of the printing material is 50~200W / (m·K) and the laser power density is... When k is less than 0.1~0.2; when the thermal conductivity of the printing material is less than 50 W / (m·K) or the laser power density is less than When k is in this case, it ranges from 0.05 to 0.1. Divide the current printing layer into several regular polygon partitions of equal area. The partition types must include at least: Square: Divides the current print layer into several square partitions of equal area, with the radius of the outer circle of each partition being... Calculate the corresponding square side length using the formula for the circumcircle radius. The scanning directions of adjacent layers are staggered at 90°, and the scanning starting points of adjacent square partitions within the same layer are offset by half the distance between the scanning lines. Each square partition uses a continuous reciprocating scanning method, while the partitions use a skip scanning method. During the skipping process, the laser power is reduced to a preset standby power of 50W. Regular hexagon: Divides the current print layer into several regular hexagonal partitions of equal area, with the radius of the circumcircle of each partition being... Based on the formula for the radius of the circumcircle, the scanning directions of adjacent layers are staggered at 90°. Within the same layer, the scanning starting points of adjacent regular hexagonal partitions are offset by half the distance between the scanning lines. Each regular hexagonal partition uses a continuous reciprocating scanning method, with the scanning direction parallel to a pair of opposite sides of the regular hexagon. A skip scanning method is used between partitions, during which the laser power is reduced to a preset standby power of 50W. Regular octagon: Divides the current print layer into several regular octagonal partitions of equal area, with the radius of the circumcircle of each partition being... Based on the formula for the radius of the circumcircle, the scanning directions of adjacent layers are staggered at 90°. Within the same layer, the scanning starting points of adjacent regular octagonal partitions are offset by half the distance between the scanning lines. Each regular octagonal partition uses a continuous reciprocating scanning method, with the scanning direction parallel to a pair of opposite sides of the regular octagon. A skip scanning method is used between partitions, during which the laser power is reduced to a preset standby power of 50W. Preset partition size adjustment coefficient Examples of possible values: When the printing material is aluminum alloy (thermal conductivity approximately 237 W / (m·K)), the laser power density is... hour, Take 0.3; When the printing material is stainless steel (thermal conductivity approximately 16 W / (m·K)), the laser power density is... hour, Take 0.08; When the printing material is titanium alloy (thermal conductivity approximately 17 W / (m·K)), the laser power density is: hour, Take 0.12; Establish a collaborative adaptability judgment matrix that includes deposition stability evaluation index and formation consistency evaluation index. Substitute the process parameter combination of each candidate scheme into the matrix to calculate the adaptability score. Compare the adaptability score of each combination with the preset qualified threshold one by one to complete the item-by-item verification of adaptability. The calculation method for sediment stability evaluation index is as follows: The coefficients of variation of the melt pool width, melt depth and temperature gradient within a consecutive preset number of frames are calculated respectively. The three coefficients of variation are substituted into the preset deposition stability evaluation function to obtain the deposition stability evaluation index value. Among them, the sedimentation stability evaluation function is the reciprocal of the geometric mean of the three coefficients of variation. The larger the sedimentation stability evaluation index value, the more stable the sedimentation process. Wherein, the coefficient of variation is the ratio of the standard deviation of the corresponding parameter within a consecutive preset number of frames to the arithmetic mean of the parameter within the same number of frames; The calculation method for the forming consistency evaluation index is as follows: Calculate the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector; Calculate the normalized Euclidean distance between the current layer candidate process parameter combination and the previous layer actual process parameter combination, with normalization based on the preset value range of each process parameter; Substitute the cosine similarity and normalized Euclidean distance into the preset forming consistency evaluation function. C represents the forming consistency evaluation index value. This represents the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector. The normalized Euclidean distance between the current candidate process parameter combination and the previous actual process parameter combination is used to obtain the forming consistency evaluation index value. The above formula obtains the evaluation index by multiplying the interlayer cosine similarity of the melt pool state vector with the normalized Euclidean distance of the process parameter combination. By combining the similarity of the melt pool state with the continuity of the process parameters, the matching degree of interlayer forming can be accurately quantified. The larger the index value, the better the consistency of interlayer forming, which can provide an intuitive and reliable evaluation basis for the stable control of interlayer quality in the printing process. Among them, the larger the value of the forming consistency evaluation index, the better the forming consistency between layers; Establish a synergistic adaptability judgment matrix that includes deposition stability evaluation index and formation consistency evaluation index. Substitute the process parameter combination of each candidate scheme into the matrix to calculate the adaptability score stage, and construct a two-dimensional synergistic adaptability judgment matrix. The rows of the matrix correspond to the preset level range of the deposition stability evaluation index, and the columns correspond to the preset level range of the formation consistency evaluation index. The deposition stability evaluation index value and the formation consistency evaluation index value of each candidate scheme are mapped to the corresponding level interval position of the judgment matrix respectively. Based on the preset matrix weight distribution, calculate the suitability score of each candidate solution; Among them, the matrix weight distribution is dynamically adjusted as the number of printing layers increases, and the closer to the printing completion stage, the higher the weight of the forming consistency evaluation index. The preset grade ranges for the deposition stability evaluation index are: 0-5 (low), 5-10 (medium), and ≥10 (high); the preset grade ranges for the shape consistency evaluation index are: 0-0.6 (low), 0.6-0.8 (medium), and ≥0.8 (high); the dynamic adjustment rules for the matrix weights are as follows: when the number of printed layers is ≤50% of the total number of layers, the deposition stability weight is 0.7 and the shape consistency weight is 0.3; when the number of printed layers is >50% and ≤80% of the total number of layers, the deposition stability weight is 0.5 and the shape consistency weight is 0.5; when the number of printed layers is >80% of the total number of layers, the deposition stability weight is 0.3 and the shape consistency weight is 0.7. The combination that passes the verification and has the highest compatibility score is selected and sent to the motion controller of the printing equipment. The status of the molten pool is continuously collected throughout the printing process to trigger dynamic parameter iteration. During the continuous acquisition of the molten pool status to trigger dynamic parameter iteration throughout the printing process, the molten pool status is acquired once every preset time interval to generate the current molten pool status vector. The Mahalanobis distance between the current molten pool status vector and the molten pool status vector at the time of the last parameter update is calculated as the state difference degree. When the state difference degree exceeds the preset difference threshold, the printing process is paused after the current scan segment is completed. The candidate layer thickness scheme generation, process parameter matching and adaptability verification process are re-executed to obtain the updated optimal process parameter combination and send it to the motion controller. Then the printing process is resumed. Among them, the preset time interval for collecting the state of the molten pool throughout the printing process is 2 seconds; the preset difference threshold for the state difference is 0.5, that is, when the Mahalanobis distance is >0.5, the parameter dynamic iteration is triggered; the preset pass threshold for the adaptability score is 80 points, and only candidate solutions with a score ≥80 points can pass the verification. The core features include the width of the molten pool, the depth of the molten pool, the temperature gradient, and the fluctuation frequency.

[0023] When the method described in the above embodiments is used for metal additive 3D printing, it can accurately capture the real-time state of the molten pool and complete data processing and feature quantization. It can adaptively generate matching layer thickness and process parameters such as laser power and scanning speed. After collaborative adaptation and selection of the optimal solution, it can also dynamically iterate and adjust parameters throughout the process. This can effectively improve the stability of printing deposition and the consistency of interlayer forming, reduce defects such as over-melting and incomplete melting, adapt to the printing needs of different metal materials, improve the forming accuracy and quality of parts, optimize printing efficiency, reduce scrap rate, and make the metal additive printing process more stable and the forming effect more reliable.

[0024] See the above embodiments for an example of the application of this method: Using TC4 titanium alloy as the printing material, this embodiment manufactures small load-bearing structural components for aerospace applications using laser selective melting metal additive 3D printing equipment. The entire process is implemented in this example. After printing is started, the high-speed vision acquisition module and infrared temperature measurement module on the device are activated simultaneously to collect multi-dimensional operating data of the printing melt pool in real time. The melt pool morphology data includes the melt pool outline, melt pool area, and melt pool aspect ratio; the temperature field distribution data includes the temperature field spatial distribution, melt pool peak temperature, and melt pool cooling rate; and the melt pool dynamic fluctuation parameters include the melt pool fluctuation frequency, fluctuation amplitude, and fluctuation period. The data collected by the two types of modules are kept synchronized in time.

[0025] Subsequently, the collected raw molten pool data underwent denoising preprocessing. First, all data were timestamped and aligned. Then, the aligned molten pool morphology sequence and temperature field sequence were spatiotemporally registered at the subpixel level to generate a three-dimensional spatiotemporally correlated molten pool data cube. Adaptive median filtering was applied to this cube for denoising, and the filter kernel side length for the current pixel position was finally determined to be 5 pixels, thus completing the purification of the morphology and temperature field data. At the same time, wavelet transform adaptive threshold denoising was applied to the dynamic fluctuation parameters of the molten pool. The wavelet threshold for the current decomposition level was finally determined to be 0.82, which can filter out more than 92% of the system Gaussian white noise, resulting in interference-free dynamic fluctuation parameters.

[0026] Next, four core features—melt pool width, melt depth, temperature gradient, and fluctuation frequency—are extracted and standardized for quantization. Melt pool width, temperature gradient, and fluctuation frequency are standardized using linear interval mapping, with outliers exceeding the original data range truncated. The standardized values ​​for these three features are 0.35, 0.41, and 0.29, respectively. Melt depth is standardized using nonlinear quantization based on the radial gradient distribution of the temperature field, resulting in a standardized melt depth feature value of 0.68. The four standardized features are then concatenated in a fixed order to generate a fixed-dimensional melt pool state vector.

[0027] Based on the melt depth characteristic value of 0.68 in the melt pool state vector, candidate layer thicknesses with equal logarithmic gradient distribution are generated within a preset layer thickness range of 0.05 mm to 0.3 mm. After interval screening and sorting, 7 groups of effective candidate layer thicknesses are obtained, namely 0.06 mm, 0.09 mm, 0.12 mm, 0.18 mm, 0.23 mm, 0.27 mm, and 0.29 mm.

[0028] For each candidate layer thickness, corresponding process parameters were matched. Taking a layer thickness of 0.18mm as an example, the basic adjustment coefficient for laser power was 1.02, matching a laser power of 285W; the correction coefficient for scanning speed was 0.97, matching a scanning speed of 1150mm / s; the powder feeding rate was matched at 12.3g / min; the scanning line spacing was determined to be 0.12mm, and the scanning line overlap rate was 32%. The scanning path adopted a square partitioned staggered scanning method, with a circumscribed circle radius of 1.2mm for each partition, corresponding to a square side length of 1.69mm. The scanning directions of adjacent layers were staggered at 90°, and the scanning starting points of adjacent partitions in the same layer were offset by 0.06mm. Continuous reciprocating scanning was performed within each partition, with inter-partition skipping scanning and the laser power reduced to 50W for standby.

[0029] A collaborative adaptability judgment matrix including deposition stability and forming consistency was constructed, and the adaptability score of each scheme was calculated. The deposition stability index of this scheme is 8.6, the forming consistency index is 0.8464, and combined with the pre-printing weight allocation rule, the adaptability score is 92.5, which is higher than the qualified threshold of 80, making it the best among all schemes.

[0030] The optimal parameters are sent to the device motion controller, and the device executes printing according to the parameters. During the printing process, the molten pool status is collected every 2 seconds, and the status difference is calculated. In this example, the difference does not exceed the threshold, so there is no need to pause the iteration. If the difference exceeds the standard in the future, the process is paused after the current scan segment is completed, candidate schemes are regenerated, parameters are matched and verified, the optimal parameters are updated, and printing is resumed.

[0031] In summary, the method in this embodiment simultaneously acquires multi-dimensional data of the molten pool through high-speed vision and infrared temperature measurement modules, and performs precise noise reduction preprocessing and standardized quantization. This enables the real-time acquisition of a stable and reliable molten pool state vector, providing accurate data support for the optimization of layer thickness and process parameters. Candidate layer thickness schemes with an equal logarithmic gradient distribution are generated based on the melt depth, allowing for the reasonable selection of layer parameters that closely match the actual molten pool state during the printing process. This avoids the problem of mismatch between thickness settings and molten pool characteristics. Simultaneous matching of laser power, scanning speed, powder feeding rate, and scanning path planning parameters achieves coordinated adaptation between layer thickness and core process parameters. Furthermore, the optimal parameter combination is selected using a coordinated adaptability judgment matrix constructed with both deposition stability and forming consistency indicators. This effectively improves deposition stability and interlayer forming consistency during the printing process, reducing the incidence of defects such as over-melting and incomplete melting. In addition, continuous acquisition of the molten pool state and triggering dynamic parameter iteration throughout the printing process ensures timely response to state changes and guarantees the stability and forming quality of the entire metal additive 3D printing process.

[0032] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A self-adaptive layering co-optimization method for metal additive 3D printing, characterized in that, include: The real-time morphology, temperature field distribution, and dynamic fluctuation parameters of the molten pool during the metal additive 3D printing process are collected simultaneously by a high-speed vision acquisition module and an infrared temperature measurement module. The collected multi-dimensional raw data of the molten pool is preprocessed for denoising, and core features are extracted simultaneously. The core features are then standardized and quantized to generate the molten pool state vector. Based on the molten pool state vector, several groups of candidate layer thickness schemes with gradient distribution are generated within a pre-defined range of layer thickness values. For each group of candidate layer thickness schemes, the corresponding combination of process parameters such as laser power, scanning speed, powder feeding rate and scanning path planning is generated simultaneously. Establish a collaborative adaptability judgment matrix that includes deposition stability evaluation index and formation consistency evaluation index. Substitute the process parameter combination of each candidate scheme into the matrix to calculate the adaptability score. Compare the adaptability score of each combination with the preset qualified threshold one by one to complete the item-by-item verification of adaptability. The combination that passes the verification and has the highest compatibility score is selected and sent to the motion controller of the printing equipment. The status of the molten pool is continuously collected throughout the printing process to trigger dynamic parameter iteration. The core features include the width of the molten pool, the depth of the molten pool, the temperature gradient, and the fluctuation frequency.

2. The adaptive layered collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, When performing noise reduction preprocessing on the collected multi-dimensional raw data of the molten pool, the real-time morphology of the molten pool, the temperature field distribution of the molten pool, and the dynamic fluctuation parameters of the molten pool are time-stamped. The timestamp-aligned melt pool morphology sequence and temperature field sequence are spatiotemporally registered at the subpixel level to generate a three-dimensional spatiotemporally correlated melt pool data cube. Adaptive median filtering based on neighborhood difference is then performed on the melt pool data cube to obtain denoised morphology and temperature field data. Adaptive threshold denoising based on wavelet transform is performed on the timestamp-aligned dynamic fluctuation parameters of the molten pool to obtain the denoised dynamic fluctuation parameters.

3. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, In the stage of standardizing and quantizing the core features to generate the melt pool state vector, the melt pool width, temperature gradient and fluctuation frequency are simultaneously subjected to linear interval mapping standardization processing to map them to the preset numerical range. During the processing, outliers that exceed the preset original data range are truncated. The melting depth is subjected to nonlinear quantization based on the radial gradient distribution of the temperature field to obtain the standardized melting depth characteristic value; The standardized molten pool width, molten depth, temperature gradient, and fluctuation frequency are spliced ​​together in a preset order to generate a molten pool state vector with fixed dimensions. The nonlinear quantization of melting depth follows the following rules: ； In the formula: These are the standardized melt depth characteristic values; This is a preset reference melting depth value; Preset quantization coefficients; As the center of the molten pool The maximum temperature gradient within a circular region with radius . The average temperature gradient of the entire molten pool; This is the preset feature radius of the molten pool.

4. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, The stage of generating several groups of candidate layer thickness schemes with gradient distribution within a pre-defined range of layer thickness values: The melt depth characteristic value in the melt pool state vector is used as the reference value. Generate a base value within a preset layer thickness range. Several groups of candidate layer thicknesses centered on a logarithmic gradient distribution; Candidate values ​​that exceed the preset range of layer thickness are removed, and the remaining candidate values ​​are sorted in ascending order to obtain the final set of candidate layer thickness schemes.

5. The adaptive layered collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, When simultaneously generating the corresponding combination of process parameters for laser power, scanning speed, powder feeding rate, and scanning path planning for each group of candidate layer thickness schemes, the following applies: The basic adjustment coefficient of laser power is determined based on the ratio of candidate layer thickness to melt depth characteristic value; The correction factor for the scanning speed is determined based on the difference between the molten pool fluctuation frequency and the preset stable fluctuation frequency. The matching value of the powder feeding rate is determined based on the product of the laser power and the scanning speed. The scan line spacing and overlap rate of the scan path are determined based on the molten pool width and the candidate layer thickness. Among them, the basic adjustment coefficient of laser power ; In the formula: h is the candidate layer thickness, The standardized melt depth characteristic value. To preset the power index, This is the preset power attenuation coefficient.

6. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, The generation of the process parameter combination for the scanning path planning follows the following: By employing a partitioned interleaved scanning strategy, the current printing layer is divided into several regular polygonal partitions of equal area, with the circumcircle radius of each partition being... , represents the preset basic circumcircle radius, k represents the preset partition size adjustment coefficient, and f represents the molten pool fluctuation frequency; The scanning directions of adjacent layers are staggered at a preset angle, and the scanning starting points of adjacent partitions within the same layer are staggered by a preset distance. Each zone uses a continuous reciprocating scanning method, while the zones use a skip scanning method. During the skipping process, the laser power is reduced to a preset standby power.

7. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, The calculation method for the sedimentation stability evaluation index is as follows: The coefficients of variation of the melt pool width, melt depth and temperature gradient within a consecutive preset number of frames are calculated respectively. The three coefficients of variation are substituted into the preset deposition stability evaluation function to obtain the deposition stability evaluation index value. The sedimentation stability evaluation function is the reciprocal of the geometric mean of the three coefficients of variation. The larger the sedimentation stability evaluation index value, the more stable the sedimentation process.

8. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, The calculation method for the forming consistency evaluation index is as follows: Calculate the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector; Calculate the normalized Euclidean distance between the current layer candidate process parameter combination and the previous layer actual process parameter combination, wherein the normalization is based on the preset value range of each process parameter; Substitute the cosine similarity and normalized Euclidean distance into the preset forming consistency evaluation function. C represents the forming consistency evaluation index value. This represents the cosine similarity between the current layer's molten pool state vector and the previous layer's molten pool state vector. The normalized Euclidean distance between the current candidate process parameter combination and the previous actual process parameter combination is used to obtain the forming consistency evaluation index value. Among them, the larger the value of the forming consistency evaluation index, the better the forming consistency between layers.

9. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, Establish a synergistic adaptability judgment matrix that includes deposition stability evaluation index and formation consistency evaluation index. Substitute the process parameter combination of each candidate scheme into the matrix to calculate the adaptability score stage, and construct a two-dimensional synergistic adaptability judgment matrix. The rows of the matrix correspond to the preset level range of the deposition stability evaluation index, and the columns correspond to the preset level range of the formation consistency evaluation index. The deposition stability evaluation index value and the formation consistency evaluation index value of each candidate scheme are mapped to the corresponding level interval position of the judgment matrix respectively. Based on the preset matrix weight distribution, calculate the suitability score of each candidate solution; The matrix weight distribution is dynamically adjusted as the number of printing layers increases, and the closer to the printing completion stage, the higher the weight of the forming consistency evaluation index.

10. The adaptive hierarchical collaborative optimization method for metal additive 3D printing according to claim 1, characterized in that, When continuously collecting the molten pool state throughout the printing process to trigger dynamic parameter iteration, the molten pool state is collected once every preset time interval to generate the current molten pool state vector. The Mahalanobis distance between the current molten pool state vector and the molten pool state vector at the time of the last parameter update is calculated as the state difference degree. When the state difference degree exceeds the preset difference threshold, the printing process is paused after the current scan segment is printed, and the candidate layer thickness scheme generation, process parameter matching and adaptability verification process are re-executed to obtain the updated optimal process parameter combination and send it to the motion controller, and then the printing process is resumed.

Images