Dimension reduction of manufacturing process non-dimensional parameters and method and system for obtaining control function thereof

Abstract

The application provides a manufacturing process dimensionless parameter dimension reduction and control function acquisition method and system, including: step 1: obtaining an experimental data set, including process parameters in a manufacturing process and corresponding target physical characteristic values under each process parameter, and constructing a parameter space of the manufacturing process; step 2: based on the parameter space of the manufacturing process, performing dimension analysis on the manufacturing process parameters to obtain a series of dimensionless parameters, and establishing a function relationship between the dimensionless parameters and the target physical characteristic values; step 3: using the experimental data set, training and optimizing the function relationship between the dimensionless parameters and the target physical characteristic values by a polynomial fitting or a machine learning method, and finally obtaining key dimensionless parameters describing the target physical characteristic values and a related function relationship. The application can simplify a high-dimensional process parameter space into a form involving only a few dimensionless parameters, effectively reducing the difficulty of process design and parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of materials processing and manufacturing technology, specifically to a method and system for dimensionless parameter reduction of manufacturing processes and acquisition of their control functions. Background Technology

[0002] In some complex manufacturing processes, the final performance of a part is affected by a variety of factors, including materials, processes, and post-processing. These factors can number in the dozens, and even a small change in each parameter can significantly impact material properties. The large number of process parameters makes parameter optimization during manufacturing extremely difficult, rendering traditional trial-and-error methods almost impossible. This problem is particularly pronounced in additive manufacturing, a complex process with multiple parameters.

[0003] Taking cold spray additive manufacturing as an example, as a solid-state deposition technology, it accelerates micron-sized powder particles to supersonic speeds using compressed airflow, utilizing kinetic energy to bond materials together. Throughout the process, the material temperature remains well below its melting point, effectively avoiding the adverse effects of high-temperature melting. It boasts advantages such as less material oxidation, lower residual stress, and minimal thermal impact on the substrate. Compared to traditional manufacturing technologies, cold spray additive manufacturing can solve the problem of machining complex-shaped parts, achieving near-net-shape forming of complex structural parts. Furthermore, it allows for the design of the composition and microstructure of different regions within the material according to the needs of the part, improving the part's adaptability to application conditions. It has become an important technology driving innovation and reform in the manufacturing system.

[0004] As the types of materials used in cold spray additive manufacturing gradually increase and the shapes of manufactured parts become more complex, determining the optimal cold spray process parameters to achieve the performance requirements of the processed parts is crucial. However, due to the complexity of the cold spray additive manufacturing process, numerous process parameters affect the final forming performance of the parts. More than 20 parameters have a crucial impact on the final quality of the parts. If the method of cross-experiment optimization is used, the corresponding combinations of process parameters can reach 1020. Under such a massive number of process parameter combinations, it is almost impossible to optimize the performance of the parts through experimental trial and error.

[0005] Physical model-based simulation reduces the economic cost of optimizing process parameters and promotes a better understanding of the relationship between multiple process parameters, microstructure, and macroscopic properties in cold spray additive manufacturing. However, due to the large number of parameters in cold spray additive manufacturing, any change in parameters such as carrier gas temperature, carrier gas pressure, powder type, or substrate heating temperature necessitates remodeling, updating relevant parameter settings, and recalculating. Each calculation typically takes several hours or even tens of hours, resulting in a significant burden on computational time costs. Due to economic and time constraints, the above methods can only optimize within a narrow processing window, leading to a degree of randomness in the optimization results and hindering the further development of the industrial application of cold spray additive manufacturing.

[0006] Patent document CN110045002A (application number: CN201910398662.1) discloses a method for extracting dimensionless feature parameters of magnetoacoustic emission signals. It decomposes the signal into single-component functions using the Unwinding AFD algorithm and selects single-component functions with small energy differences to form a reconstructed signal. Traditional and dimensionless feature parameters of the magnetoacoustic emission signals under different excitation conditions are extracted for low-cycle fatigue and high-cycle fatigue states. After normalization, the traditional and dimensionless feature parameters are compared to obtain the analysis results of the dimensionless feature parameters on the microstructure and stress state of early-stage faults. However, this patent cannot solve the existing technical problems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for dimensionless parameter reduction in manufacturing processes and the acquisition of their control functions.

[0008] The method for dimensionless parameter reduction and control function acquisition of manufacturing processes according to the present invention includes:

[0009] Step 1: Obtain the experimental dataset, including the process parameters in the manufacturing process and the corresponding target physical feature values ​​under each process parameter, and construct the parameter space of the manufacturing process;

[0010] Step 2: Based on the parameter space of the manufacturing process, perform dimensional analysis on the manufacturing process parameters to obtain a series of dimensionless parameters, and establish the functional relationship between the dimensionless parameters and the target physical characteristic values;

[0011] Step 3: Using the experimental dataset, train and optimize the functional relationship between the dimensionless parameters and the target physical feature values ​​through multinomial fitting or machine learning methods, and finally obtain the key dimensionless parameters and related functional relationships describing the target physical feature values.

[0012] Preferably, the process parameters are adjustable parameters during the cold spray additive manufacturing process, and the performance of the sprayed material is controlled by changing the process parameters; the process parameters include the carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P and matrix hardness H S ;

[0013] The target physical characteristic value is a dimensionless number that reflects the material properties, including the particle length-to-length ratio and the material porosity; the larger the length-to-length ratio of the deformed particles, the better the material properties; the smaller the internal porosity of the material, the better the material properties.

[0014] Preferably, step 2 includes:

[0015] The relationship between the target physical characteristic value Y and the process parameters is expressed as follows:

[0016] Y = f(p1, p2, ... p) n )=f(p)

[0017] Where p is the product of powers of the input variables, expressed as: The power of a variable is represented by a vector w, as: w = [w1, w2, ..., w n ] T n is the number of process parameters in the complex process; p1 is a specific parameter in the complex process, and w1 is the power of that parameter. arrive The product of these components constitutes the dimensionless parameter p;

[0018] The dimensional matrix D of the manufacturing process parameters is established based on length L, time T, mass M, and temperature Θ, and its expression is:

[0019] D = [v(p1),v(p2),…v(p...] n )]

[0020] For a carrier gas pressure P, the corresponding dimension vector v(P) = [-1, -2, 1, 0]. T ;

[0021] Perform row operations on the dimensional matrix D and calculate the rank r of the matrix;

[0022] After dimensionless processing of the manufacturing process parameter space, the solution w satisfying the equation Dw=0 is expressed as:

[0023] w = γ1w1 + γ2w2 + ... + γ K w K

[0024] K = nr

[0025] Where γ1, γ2, ... γ K is a dimensional parameter.

[0026] Preferably, step 3 includes: using polynomial fitting or a neural network to train the dimensionless numbers obtained from the above-mentioned sample training set, and establishing a functional relationship between the target physical feature value and the dimensionless parameter;

[0027] When the constructed polynomial is a fifth-order polynomial, its expression is written as: Y = f(p) = β0 + β1p + β2p 2 +β3p 3 +β4p 4 +β5p 5 The correlation coefficient vector β = [β0, β1, β2, β3, β4, β5], where β0, β1, β2, β3, β4, β5 are polynomial coefficients.

[0028] Preferably, a two-stage optimization method is used to determine the two sets of unknown parameters in the regression problem: the dimensional parameter γ and the polynomial coefficient β;

[0029] The optimization process includes multiple iterations. While keeping the second-order polynomial coefficients β constant, the first-order dimensional parameter γ is adjusted. Then, keeping γ constant, the polynomial coefficients β are optimized, and this process is repeated. The correlation coefficient R2 of the test dataset is used as a measure of data learning performance. Through the above iterative process, the dimensionless number corresponding to a high R2 value is found, which is used as the most important dimensionless parameter describing the physical characteristic, and the correlation function relationship is obtained.

[0030] The system for dimensionless parameter reduction and control function acquisition of manufacturing processes according to the present invention includes:

[0031] Module M1: Acquire experimental datasets, including process parameters in the manufacturing process and the corresponding target physical feature values ​​under each process parameter, and construct the parameter space of the manufacturing process;

[0032] Module M2: Based on the parameter space of the manufacturing process, dimensional analysis is performed on the manufacturing process parameters to obtain a series of dimensionless parameters, and a functional relationship between the dimensionless parameters and the target physical characteristic values ​​is established.

[0033] Module M3: Using experimental datasets, the functional relationship between dimensionless parameters and target physical feature values ​​is trained and optimized through multinomial fitting or machine learning methods, ultimately obtaining key dimensionless parameters and related functional relationships that describe the target physical feature values.

[0034] Preferably, the process parameters are adjustable parameters during the cold spray additive manufacturing process, and the performance of the sprayed material is controlled by changing the process parameters; the process parameters include the carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P and matrix hardness H S ;

[0035] The target physical characteristic value is a dimensionless number that reflects the material properties, including the particle length-to-length ratio and the material porosity; the larger the length-to-length ratio of the deformed particles, the better the material properties; the smaller the internal porosity of the material, the better the material properties.

[0036] Preferably, the module M2 includes:

[0037] The relationship between the target physical characteristic value Y and the process parameters is expressed as follows:

[0038] Y = f(p1, p2, ... p) n )=f(p)

[0039] Where p is the product of powers of the input variables, expressed as: The power of a variable is represented by a vector w, as: w = [w1, w2, ..., w n ] T n is the number of process parameters in the complex process; p1 is a specific parameter in the complex process, and w1 is the power of that parameter. arrive The product of these components constitutes the dimensionless parameter p;

[0040] The dimensional matrix D of the manufacturing process parameters is established based on length L, time T, mass M, and temperature Θ, and its expression is:

[0041] D = [v(p1),v(p2),…v(p...] n )]

[0042] For a carrier gas pressure P, the corresponding dimension vector v(P) = [-1, -2, 1, 0]. T ;

[0043] Perform row operations on the dimensional matrix D and calculate the rank r of the matrix;

[0044] After dimensionless processing of the manufacturing process parameter space, the solution w satisfying the equation Dw=0 is expressed as:

[0045] w = γ1w1 + γ2w2 + ... + γ K w K

[0046] K = nr

[0047] Where γ1, γ2, ... γ K is a dimensional parameter.

[0048] Preferably, module M3 includes: using polynomial fitting or a neural network to train the dimensionless numbers obtained from the above-mentioned sample training set, and establishing a functional relationship between the target physical feature value and the dimensionless parameter;

[0049] When the constructed polynomial is a fifth-order polynomial, its expression is written as: Y = f(p) = β0 + β1p + β2p 2 +β3p 3 +β4p 4 +β5p 5 The correlation coefficient vector β = [β0, β1, β2, β3, β4, β5], where β0, β1, β2, β3, β4, β5 are polynomial coefficients.

[0050] Preferably, a two-stage optimization method is used to determine the two sets of unknown parameters in the regression problem: the dimensional parameter γ and the polynomial coefficient β;

[0051] The optimization process includes multiple iterations. While keeping the second-order polynomial coefficients β constant, the first-order dimensional parameter γ is adjusted. Then, keeping γ constant, the polynomial coefficients β are optimized, and this process is repeated. The correlation coefficient R2 of the test dataset is used as a measure of data learning performance. Through the above iterative process, the dimensionless number corresponding to a high R2 value is found, which is used as the most important dimensionless parameter describing the physical characteristic, and the correlation function relationship is obtained.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] (1) The method proposed in this invention is a physical-based process parameter dimensionality reduction method, which can simplify complex, multi-parameter controlled processing technology into a form described by several dimensionless parameters with clear physical meaning, and more simply and intuitively describe the relationship between process parameters and material performance characteristics.

[0054] (2) This invention transforms numerous process features into low-dimensional models that are not affected by units and scale, enabling the method to train an accurate and universal model under limited experimental data conditions, and to achieve accurate prediction of some physical features in the cold spray additive manufacturing process.

[0055] (3) The two-level optimization method proposed in this invention can more flexibly select machine learning methods. Under the condition that the learning method of a certain level remains unchanged, the parameter optimization method of another level can be flexibly selected according to the process parameters. For example, in the second level, either multinomial regression or tree-based regression analysis method can be selected. Attached Figure Description

[0056] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0057] Figure 1 This is a schematic diagram of the basic process of the present invention;

[0058] Figure 2 A schematic diagram of the process for cold spray additive manufacturing;

[0059] Figure 3 This is a schematic diagram illustrating the powder deformation characteristics during cold spraying.

[0060] Figure 4 Metallographic images of 2024 aluminum alloy material for cold spray additive manufacturing;

[0061] Figure 5 This is a schematic diagram of a machine learning method that represents dimensionless parameters and their functional relationship with a certain physical feature. Detailed Implementation

[0062] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0063] Example 1:

[0064] This invention proposes a data-driven method for dimensionless parameter reduction and control function acquisition in manufacturing processes. It identifies dimensionless parameters and corresponding functional relationships from limited experimental data that describe the physical characteristics of the process. This method is a physics-based process parameter reduction approach that simplifies complex, multi-parameter controlled processes into descriptions using a few dimensionless parameters with clear physical meanings. This provides a simpler and more intuitive description of the relationship between process parameters and material properties. A flowchart is shown below. Figure 1 As shown, applying this to the optimization of process parameters in cold spray additive manufacturing can effectively avoid the high economic and time costs associated with extensive experiments or simulations, significantly improving the applicability and effectiveness of process parameter optimization.

[0065] Step 1: Determine the process parameter space of cold spray additive manufacturing and obtain the experimental dataset. The dataset includes the process parameters in the manufacturing process and the corresponding target physical feature values ​​under each process parameter.

[0066] There are many methods for obtaining experimental data. For example, this embodiment provides the following method: obtaining experimental data on cold spray additive manufacturing process parameters—target physical characteristic parameters—from publicly available papers or through self-conducted experiments. The established database includes a process parameter space, such as carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P Matrix hardness H S A schematic diagram of the cold spray additive manufacturing process is shown below. Figure 2 As shown. The target physical characteristic parameters are quantitative parameters, reflecting the material's performance through an accurate numerical value, such as the aspect ratio of powder particles (a higher aspect ratio indicates better performance), such as... Figure 3 As shown; and the material porosity (closest to 0 indicates better performance), such as Figure 4 As shown, the above data is compiled into a material library, which includes manufacturing process parameters and their corresponding target physical characteristics. In this implementation case, porosity is used as the target characteristic parameter.

[0067] Step 2: Based on the above cold spray additive manufacturing process parameter space, perform dimensional analysis on the additive manufacturing process parameters to obtain a series of dimensionless parameters, and establish a functional relationship between the dimensionless parameters and the target physical characteristic parameter porosity Φ.

[0068] The relationship between porosity and process parameters is expressed as follows:

[0069] Φ=f(T g ,P,L s ,ρ,L P H P H S …)=f(p); p is the product of powers of the input variables, expressed as:

[0070] The power vector w of variables is represented as: w = [w1, w2, ..., w7] T .

[0071] Determine the units for the process parameters and express them using basic unit length L, time T, mass M, and temperature Θ.

[0072] parameter symbol unit LTMΘ Dimension Carrier gas temperature <![CDATA[T g ]]> K <![CDATA[L 0 T 0 M 0 Θ 1 ]]> Carrier gas pressure P Pa <![CDATA[L -1 T -2 M 1 Θ 0 ]]> Spraying distance <![CDATA[L S ]]> m <![CDATA[L 1 T 0 M 0 Θ 0 ]]> powder density ρ <![CDATA[Kg / m 3 ]]> <![CDATA[L -3 T 0 M 1 Θ 0 ]]> Powder particle size <![CDATA[L P ]]> m <![CDATA[L 1 T 0 M 0 Θ 0 ]]> Powder Hardness <![CDATA[H P ]]> Hv <![CDATA[L -1 T -2 M 1 Θ 0 ]]> matrix hardness <![CDATA[H S ]]> Hv <![CDATA[L -1 T -2 M 1 Θ 0 ]]>

[0073] Based on the fundamental dimensions of LTMΘ, a dimensional matrix is ​​established, and the dimensional matrix D is shown below:

[0074] <![CDATA[T g ]]> P <![CDATA[L S ]]> ρ <![CDATA[L P ]]> <![CDATA[H P ]]> <![CDATA[H S ]]> L 0 -1 1 -3 1 -1 -1 T 0 -2 0 0 0 -2 -2 M 0 1 0 1 0 1 1 Θ 1 0 0 0 0 0 0

[0075] By performing row operations on the matrix, the final matrix has a rank of 4.

[0076] Row transformations of a matrix are a fundamental technique in matrix operations. The main methods used are: (1) adding K times a row to another row; (2) swapping two rows; and (3) multiplying a row by K (K≠0) to transform the matrix into a row echelon matrix. The number of non-zero rows in this row echelon matrix is ​​the rank r of the matrix.

[0077] The manufacturing process parameter space is dimensionless, and the vector w is solved to make Dw = 0. Since this equation is underdetermined, there are infinitely many solutions satisfying it, indicating that infinitely many dimensionless parameters can be calculated. All solutions satisfying the equation can be expressed as a linear combination of three basis vectors: w = γ1w b1 +γ2w b2 +γ3w b3 The number of basis vectors is equal to the number of process parameters (7 in this case) minus the rank of the dimensional equation (4 in this case).

[0078] The expressions for the three basis vectors are obtained using Buckingham's PI theory:

[0079] w b1 =[0,0,1,0,-1,0,0] T

[0080] w b1 =[0,1,0,0,0,-1,0] T

[0081] w b1 =[0,1,0,0,0,0,-1] T

[0082] Therefore, the coefficients of the basis vectors (referred to as "basis coefficients") are unknowns to be determined. Once the coefficients γ1, γ2, and γ3 of the basis vectors are determined, a definite dimensionless parameter П can be obtained.

[0083] Step 3: Using the dataset established in Step 1, the functional relationship between the target physical feature parameters and dimensionless parameters is trained and optimized by using a multinomial fitting method. Finally, the most important dimensionless parameters and related functional relationships that can describe the physical feature are obtained.

[0084] To determine the values ​​of the aforementioned basis coefficients using experimental datasets, a model needs to be constructed that represents the relationship between the target physical quantity and the dimensionless parameter. For example, when the constructed relationship model is a fifth-order polynomial, its expression is written as: Y = f(p) = β0 + β1П + β2П 2 +β3П 3 +β4pП 4 +β5П 5 The correlation coefficient vector β = [β0, β1, β2, β3, β4, β5] is introduced. At this time, another set of position parameters β is introduced, and the coefficient β is a polynomial coefficient.

[0085] To obtain the two sets of unknown parameters in the regression problem, namely the basic coefficients γ and the polynomial coefficients β, an iterative two-stage optimization scheme was designed, such as... Figure 5 As shown. The first-order basis coefficients γ have a clear physical meaning; they are related to the powers that produce dimensionless numbers. Therefore, these values ​​must be rational numbers to maintain dimensionality invariance. Furthermore, their typical range is finite; in most dimensionless coefficients and polynomial relationships, the absolute value of the coefficients is less than 4. The optimization scheme involves multiple iterative steps. In each step, the first-order basis coefficients γ are adjusted while keeping the second-order polynomial coefficients β constant, and then the second-order polynomial coefficients β are optimized while keeping the first-order basis coefficients γ constant. This process is repeated. The correlation coefficient R² of the test dataset is used as a measure of data learning performance. By continuously adjusting the basis coefficients γ and polynomial coefficients β, the correlation coefficient R² of the test dataset is made closest to 1, ultimately obtaining the most important dimensionless parameters (γ values) and correlation function relationships (β values) that can describe this physical characteristic.

[0086] This invention achieves dimensionality reduction of complex process parameters through a two-level optimization method, obtaining dimensionless parameters and corresponding functional relationships that describe the physical characteristics of the process. Applying this method to the optimization of process parameters in cold spray additive manufacturing effectively avoids the high economic and time costs associated with extensive experiments or simulations, significantly improving the applicability and effectiveness of process parameter optimization. Besides cold spray additive manufacturing, this method is also applicable to parameter dimensionality reduction in other complex processes.

[0087] Example 2:

[0088] The present invention also provides a system for dimensionless parameter reduction of manufacturing process and acquisition of its control function. The system for dimensionless parameter reduction of manufacturing process and acquisition of its control function can be implemented by executing the process steps of the method for acquisition of dimensionless parameter reduction of manufacturing process and acquisition of its control function. That is, those skilled in the art can understand the method for acquisition of dimensionless parameter reduction of manufacturing process and acquisition of its control function as a preferred embodiment of the system for acquisition of dimensionless parameter reduction of manufacturing process and acquisition of its control function.

[0089] The system for dimensionless parameter reduction and control function acquisition of manufacturing processes according to the present invention includes: Module M1: acquiring experimental datasets, including process parameters in the manufacturing process and corresponding target physical feature values ​​under each process parameter, and constructing a parameter space for the manufacturing process; Module M2: performing dimensional analysis on the manufacturing process parameters based on the parameter space of the manufacturing process to obtain a series of dimensionless parameters and establishing a functional relationship between the dimensionless parameters and the target physical feature values; Module M3: using the experimental dataset, training and optimizing the functional relationship between the dimensionless parameters and the target physical feature values ​​through polynomial fitting or machine learning methods, and finally obtaining the key dimensionless parameters describing the target physical feature values ​​and related functional relationships.

[0090] Process parameters are adjustable parameters during cold spray additive manufacturing, which control the properties of the sprayed material. These parameters include the carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P and matrix hardness H S ;

[0091] The target physical characteristic value is a dimensionless number that reflects the material properties, including the particle length-to-length ratio and the material porosity; the larger the length-to-length ratio of the deformed particles, the better the material properties; the smaller the internal porosity of the material, the better the material properties.

[0092] The module M2 includes:

[0093] The relationship between the target physical characteristic value Y and the process parameters is expressed as follows:

[0094] Y = f(p1, p2, ... p) n )=f(p)

[0095] Where p is the product of powers of the input variables, expressed as: The power of a variable is represented by a vector w, as: w = [w1, w2, ..., w n ] T n is the number of process parameters in the complex process; p1 is a specific parameter in the complex process, and w1 is the power of that parameter. arrive The product of these components constitutes the dimensionless parameter p;

[0096] The dimensional matrix D of the manufacturing process parameters is established based on length L, time T, mass M, and temperature Θ, and its expression is:

[0097] D = [v(p1),v(p2),…v(p...] n )]

[0098] For example: the dimension vector v(P) corresponding to the carrier gas pressure P is [-1, -2, 1, 0]. T ;

[0099] Perform row operations on the dimensional matrix D and calculate the rank r of the matrix;

[0100] After dimensionless processing of the manufacturing process parameter space, the solution w satisfying the equation Dw=0 is expressed as:

[0101] w = γ1w1 + γ2w2 + ... + γ K w K

[0102] K = nr

[0103] Where γ1, γ2, ... γ K is a dimensional parameter.

[0104] The module M3 includes: using polynomial fitting or a neural network to train the dimensionless numbers obtained from the above sample training set, and establishing a functional relationship between the target physical feature value and the dimensionless parameter;

[0105] When the constructed polynomial is a fifth-order polynomial, its expression is written as: Y = f(p) = β0 + β1p + β2p 2 +β3p 3 +β4p 4 +β5p 5 The correlation coefficient vector β = [β0, β1, β2, β3, β4, β5], where β0, β1, β2, β3, β4, β5 are polynomial coefficients.

[0106] A two-stage optimization method is used to determine two sets of unknown parameters in the regression problem: the dimensional parameter γ and the polynomial coefficients β.

[0107] The optimization process includes multiple iterations. While keeping the second-order polynomial coefficients β constant, the first-order dimensional parameter γ is adjusted. Then, keeping γ constant, the polynomial coefficients β are optimized, and this process is repeated. The correlation coefficient R2 of the test dataset is used as a measure of data learning performance. Through the above iterative process, the dimensionless number corresponding to a high R2 value is found, which is used as the most important dimensionless parameter describing the physical characteristic, and the correlation function relationship is obtained.

[0108] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0109] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for dimensionless parameter reduction and control function acquisition in a manufacturing process, characterized in that, include: Step 1: Obtain the experimental dataset, including the process parameters in the manufacturing process and the corresponding target physical feature values ​​under each process parameter, and construct the parameter space of the manufacturing process; Step 2: Based on the parameter space of the manufacturing process, perform dimensional analysis on the manufacturing process parameters to obtain a series of dimensionless parameters, and establish the functional relationship between the dimensionless parameters and the target physical characteristic values; Step 3: Using the experimental dataset, train and optimize the functional relationship between the dimensionless parameters and the target physical feature values ​​through multinomial fitting or machine learning methods, and finally obtain the key dimensionless parameters and related functional relationships that describe the target physical feature values. Step 2 includes: The relationship between the target physical characteristic value Y and the process parameters is expressed as follows: in, The expression for the product of powers of the input variables is: The exponentiation of a variable is expressed using a vector. This is represented as: n is the number of process parameters in a complex process; It is a parameter in a complex process. It is a power of the process parameter. arrive The product of these components constitutes the dimensionless parameter p; Establish a dimensional matrix of manufacturing process parameters based on length L, time T, mass M, and temperature Θ. The expression is: For carrier gas pressure The corresponding dimension vector =[-1,-2,1,0] T ; Perform row operations on the dimensional matrix D and calculate the rank r of the matrix; The manufacturing process parameter space is dimensionless to satisfy the equation. Solution Represented as: in, Dimensional parameter; Step 3 includes: using polynomial fitting or a neural network to train the dimensionless numbers obtained from the above sample training set, and establishing a functional relationship between the target physical feature value and the dimensionless parameter; When the constructed polynomial is a fifth-order polynomial, its expression is written as: Correlation coefficient vector ;in, These are the polynomial coefficients; Process parameters are adjustable parameters during cold spray additive manufacturing, which control the properties of the sprayed material. These parameters include the carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P and matrix hardness H S ; The target physical characteristic value is a dimensionless number that reflects the material properties, including the particle length-to-length ratio and the material porosity; the larger the length-to-length ratio of the deformed particles, the better the material properties; the smaller the internal porosity of the material, the better the material properties. A two-stage optimization method is used to determine two sets of unknown parameters in the regression problem: the dimensional parameter γ and the polynomial coefficients β. The optimization process includes multiple iterations. While keeping the second-order polynomial coefficients β constant, the first-order dimensional parameter γ is adjusted. Then, keeping γ constant, the polynomial coefficients β are optimized, and this process is repeated. The correlation coefficient R2 of the test dataset is used as a measure of data learning performance. Through the above iterative process, the dimensionless number corresponding to the R2 value is found, which is used as the dimensionless parameter describing the physical characteristic, and the correlation function relationship is obtained.

2. A system for dimensionless parameter reduction and control function acquisition in a manufacturing process, characterized in that, include: Module M1: Acquire experimental datasets, including process parameters in the manufacturing process and the corresponding target physical feature values ​​under each process parameter, and construct the parameter space of the manufacturing process; Module M2: Based on the parameter space of the manufacturing process, dimensional analysis is performed on the manufacturing process parameters to obtain a series of dimensionless parameters, and a functional relationship between the dimensionless parameters and the target physical characteristic values ​​is established. Module M3: Using experimental datasets, the functional relationship between dimensionless parameters and target physical feature values ​​is trained and optimized through multinomial fitting or machine learning methods, ultimately obtaining the key dimensionless parameters describing the target physical feature values ​​and related functional relationships; The module M2 includes: The relationship between the target physical characteristic value Y and the process parameters is expressed as follows: in, The expression for the product of powers of the input variables is: The exponentiation of a variable is expressed using a vector. This is represented as: n is the number of process parameters in a complex process; It is a parameter in a complex process. It is a power of the process parameter. arrive The product of these components constitutes the dimensionless parameter p; Establish a dimensional matrix of manufacturing process parameters based on length L, time T, mass M, and temperature Θ. The expression is: For carrier gas pressure The corresponding dimension vector =[-1,-2,1,0] T ; Perform row operations on the dimensional matrix D and calculate the rank r of the matrix; The manufacturing process parameter space is dimensionless to satisfy the equation. Solution Represented as: in, Dimensional parameter; The module M3 includes: using polynomial fitting or a neural network to train the dimensionless numbers obtained from the above sample training set, and establishing a functional relationship between the target physical feature value and the dimensionless parameter; When the constructed polynomial is a fifth-order polynomial, its expression is written as: Correlation coefficient vector ;in, These are the polynomial coefficients; Process parameters are adjustable parameters during cold spray additive manufacturing, which control the properties of the sprayed material. These parameters include the carrier gas temperature T. g Carrier gas pressure P, spraying distance L S Powder density ρ, powder particle size L P Powder hardness H P and matrix hardness H S ; The target physical characteristic value is a dimensionless number that reflects the material properties, including the particle length-to-length ratio and the material porosity; the larger the length-to-length ratio of the deformed particles, the better the material properties; the smaller the internal porosity of the material, the better the material properties. A two-stage optimization method is used to determine two sets of unknown parameters in the regression problem: the dimensional parameter γ and the polynomial coefficients β. The optimization process includes multiple iterations. While keeping the second-order polynomial coefficients β constant, the first-order dimensional parameter γ is adjusted. Then, keeping γ constant, the polynomial coefficients β are optimized, and this process is repeated. The correlation coefficient R2 of the test dataset is used as a measure of data learning performance. Through the above iterative process, the dimensionless number corresponding to the R2 value is found, which is used as the dimensionless parameter describing the physical characteristic, and the correlation function relationship is obtained.

Images