Nanoscale sawtooth design and manufacturing method based on machine learning

Abstract

The invention discloses a nanoscale sawtooth design and manufacturing method based on machine learning. The nanoscale sawtooth design and manufacturing method comprises the following steps: S1, collecting and processing initial data related to a nanoscale sawtooth structure; s2, performing analysis and feature extraction on the data by using feature engineering, and determining feature parameters; s3, selecting and constructing a machine learning model, and training the extracted feature parameters; s4, applying the trained machine learning model, and generating a nanoscale sawtooth structure design scheme meeting requirements according to a preset target performance index; s5, performing reverse design on the generated design scheme, and deducing a specific manufacturing process parameter optimization design scheme; s6, in the manufacturing process, real-time monitoring is carried out based on the derived manufacturing process parameters, and dynamic adjustment is carried out in combination with a machine learning model; and S7, after manufacturing is completed, performance testing is carried out on the produced sawtooth structure, and a result is fed back to the machine learning model. According to the method, efficient control and optimization of design and manufacturing of the nanoscale sawtooth structure are achieved.

Description

Technical Field

[0001] The present invention relates to the field of nano-manufacturing technology, and in particular to a nano-scale sawtooth design and manufacturing method based on machine learning. Background Art

[0002] Nanoscale sawtooth structures have important applications in many high-precision technology fields, such as micro-nanoelectronic devices, optical components, microfluidic systems, and biosensors. These application fields place increasingly high demands on the geometric accuracy, manufacturing consistency, and functional complexity of nanostructures. However, existing nanofabrication technologies still have significant limitations in meeting these requirements.

[0003] In the traditional nanoscale structure manufacturing process, photolithography, electron beam etching and nanoimprinting are the most commonly used methods. Although photolithography is widely used in the field of semiconductor manufacturing, it is difficult to achieve ultra-high-resolution nanostructure manufacturing due to the limitations of light source wavelength and photoresist performance. Electron beam etching technology can achieve higher processing resolution, but its manufacturing speed is slow and the equipment cost is high, which is not suitable for large-scale production. Nanoimprinting technology can effectively replicate large-area nanostructures, but it still faces many challenges in mold making and material selection.

[0004] Existing nanofabrication technology usually relies on the experience of process engineers and the gradual adjustment of process parameters to optimize the manufacturing process. This method is not only time-consuming and labor-intensive, but also difficult to ensure the stability of the manufacturing process and the consistency of the product. Especially when manufacturing nano-scale sawtooth structures, any slight geometric deviation will significantly affect the performance of the final product. Therefore, achieving precise control of the manufacturing process to ensure high precision and high consistency of the nanostructure is an urgent problem to be solved in the current technical field.

[0005] In recent years, with the development of computer science and artificial intelligence technology, the application of machine learning in the manufacturing process has gradually received attention. Machine learning technology can automatically establish a mapping relationship between process parameters and manufacturing results by analyzing a large amount of historical data, thereby achieving real-time adjustment and optimization in the manufacturing process. This data-driven optimization method can effectively reduce dependence on engineer experience and improve the degree of automation and manufacturing quality of the manufacturing process.

[0006] However, the current application of machine learning in nanomanufacturing is still in its early stages and faces the following problems:

[0007] 1. Difficulty in extracting feature parameters: There are many process parameters involved in the nanomanufacturing process, and there are often complex nonlinear coupling relationships between the parameters. How to effectively extract the key feature parameters that affect the manufacturing results is the primary difficulty in building a machine learning model. Most existing feature extraction methods rely on traditional data processing technology, which makes it difficult to fully capture the deep relationship between process parameters and manufacturing results.

[0008] 2. Limited data samples: Due to the high experimental costs in the nanofabrication process, the available data samples are often limited. The accuracy and generalization ability of machine learning models depend to a large extent on sufficient data. Limited data samples can easily lead to overfitting of the model, thereby reducing the performance of the model in practical applications.

[0009] 3. Complex model training: In the nanomanufacturing process, there may be high-dimensional nonlinear relationships between the process parameters involved and the target performance indicators. This makes it difficult for traditional machine learning models to effectively handle these complex relationships, resulting in a complex and time-consuming model training process. In addition, when processing real-time data, the response speed of existing machine learning models often cannot meet the real-time adjustment requirements in the manufacturing process, affecting the application effect of the model in actual manufacturing.

[0010] 4. Insufficient real-time dynamic adjustment capabilities: In the actual manufacturing process, process parameters may be affected by a variety of factors, resulting in dynamic changes in process conditions. Existing machine learning models often fail to respond effectively in a timely manner when dealing with such dynamic changes, making it difficult to achieve precise control of the manufacturing process, affecting the consistency and accuracy of manufacturing.

[0011] In view of the above problems, the existing technology still has the defects of insufficient precision, complex process control and poor manufacturing consistency in the manufacturing process of nano-scale sawtooth structures. Especially in the manufacturing of nanostructures with high precision and high consistency requirements, traditional methods and existing machine learning models are difficult to meet the needs of actual production.

[0012] Therefore, how to provide a nanoscale sawtooth design and manufacturing method based on machine learning is an urgent problem that technicians in this field need to solve. Summary of the invention

[0013] One purpose of the present invention is to propose a nanoscale sawtooth design and manufacturing method based on machine learning. The present invention optimizes and controls the design parameters and manufacturing process of the nanoscale sawtooth structure by utilizing machine learning technology, combined with Hodge state decomposition and abnormal cycle dimensional analysis, and describes in detail how to achieve high-precision and high-consistency nanostructure manufacturing through a data-driven method. It has the advantages of high manufacturing accuracy, intelligent process control, and a high degree of automation in the manufacturing process.

[0014] A nano-scale sawtooth design and manufacturing method based on machine learning according to an embodiment of the present invention comprises the following steps:

[0015] S1. Collect and process initial data related to the nano-scale sawtooth structure, including geometric parameters, material properties, process parameters, and target performance indicators;

[0016] S2. Analyze and extract features from the processed data using feature engineering, extract initial feature vectors from the initial data, map the initial feature vectors to a high-dimensional latent space, capture the abnormal cycle patterns and dimensional structures hidden in the data, and calculate the characteristic parameters that affect the performance of the sawtooth structure through multi-scale periodic analysis and singular value decomposition;

[0017] S3. Select and build a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model, train the extracted feature parameters, and predict the complex nonlinear relationship between design parameters and target performance indicators through geometric depth mapping, singular point topological transformation and quantum state optimization process;

[0018] S4. Apply the trained polymorphic neural network model for constructing Hodge state decomposition and abnormal cycle dimension analysis, generate a nanoscale sawtooth structure design scheme according to the predetermined target performance indicators, optimize the design parameters through non-equilibrium fractal Laplace transform and quantum topological interleaving operator, and output a design scheme that meets the predetermined performance requirements;

[0019] S5. Perform reverse design on the generated design scheme to derive the manufacturing process parameters required to implement the design scheme;

[0020] S6. During the manufacturing process, real-time monitoring is performed based on the derived manufacturing process parameters, and dynamic adjustments are made by combining the construction of Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network models to optimize manufacturing accuracy and consistency;

[0021] S7. After manufacturing is completed, the performance of the produced nano-scale sawtooth structure is tested, and the test results are fed back to the construction of the Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to update and optimize the model.

[0022] Optionally, the S2 specifically includes:

[0023] S21, receiving the initial feature vector Z extracted from the original data = [z1, z2,, z p ] and mapped to a high-dimensional latent space To capture the abnormal cycle patterns and dimensional structures implicit in the data, the mapping is defined by the abnormal cycle mapping function C(·):

[0024]

[0025] Among them, α i is the mapping weight, W i is the weight matrix, C latent It is the latent space representation after mapping, aiming to reveal the nonlinear cyclic characteristics in the data;

[0026] S22, the latent space after mapping is represented by C latent Perform multi-scale periodic analysis using multi-scale transformation S j (·) Decompose the cyclic patterns in the latent space and extract abnormal features at different scales:

[0027]

[0028] Among them, β jk is the multi-scale analysis coefficient, ω jk is the frequency parameter, is the phase shift, q j is the number of patterns at the jth scale;

[0029] S23 The abnormal cycle characteristics obtained by multi-scale periodic analysis Combined into a multi-dimensional abnormal cycle feature vector A cycle ;

[0030] S24, for the comprehensive abnormal cycle feature vector A cycle Perform singular value decomposition to decompose the singular value σ that dominates the abnormal cycle characteristics k and the corresponding singular vector u k ;

[0031] S25, the main abnormal cycle feature A after singular value decomposition cycle as output.

[0032] Optionally, the S3 specifically includes:

[0033] S31, receiving the Hodge state component H output by the Hodge state decomposition layer in claim 2 = [H1, H2,, H m ] and the abnormal cycle feature A output by the abnormal cycle dimension analysis layer in claim 3 cycle =[A 1,cycle ,A 2,cycle ,,A n,cycle ], taking the feature parameters as input and applying them to the multidimensional nonlinear mapping space Construct a high-dimensional mapping relationship between the input feature space and the target performance index,

[0034] S32. Construct a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to generate abnormal cycle feature vectors based on high-dimensional mapping relationships;

[0035] S33, perform geometric depth mapping on the input abnormal cycle feature vector, using a high-order surface fitting function G i (·) Perform nonlinear mapping on Hodge state components and abnormal cycle features to generate high-dimensional geometric feature representation F geo :

[0036]

[0037] Among them, ij is the high-dimensional geometric mapping coefficient, ∧ represents the outer product operation, Q ij is a symmetric positive definite matrix. The mapping constructs a feature representation containing complex geometric and topological information by combining the outer product and the exponential function. The abnormal cycle feature A cycle ;

[0038] S34, geometric feature representation F geo Perform singular point topological transformation and introduce the singular point transformation function S based on high-order differential geometry k (·), identifying and amplifying singularities and topological anomalies in high-dimensional latent spaces:

[0039]

[0040] Among them, ξ k is the singular point transformation coefficient, represents the generalized inverse, is the k-th order derivative operator, and the high-dimensional geometric feature representation F geo ;

[0041] S35. The characteristic T after the topological transformation of the singular point through quantum state topological optimization singular Perform quantum state optimization to generate optimized feature representation T quantum ;

[0042] S36. Using high-order generalized convolution to optimize the feature representation of quantum state T quantum Perform convolution operation to generate convolution feature representation F conv :

[0043]

[0044] Among them, ν t is the convolution kernel coefficient, C t (·) is a high-order generalized convolution operation, which extracts nonlinear and multi-scale information in features through multi-order convolution operations and generates deep feature representations for prediction. r: the total number of convolution operations, indicating how many different convolution operations are used to enrich feature extraction.

[0045] S37, through symbolic regression and multi-state prediction model, the convolution feature representation F conv Convert to symbolic prediction expression Generate predictions between design parameters and target performance indicators

[0046]

[0047] in, The final prediction result represents the relationship between the design parameters and the target performance indicators. The P(·) symbolic regression and multi-state prediction model transforms the numerical features into interpretable symbolic expressions. Symbolic regression coefficients adjust the contribution of different features to the prediction results, High-order nonlinear terms represent complex nonlinear relationships in features and enhance the predictive power of the model. The order of the s-symbol regression indicates how many regression terms of different orders are included.

[0048] S38, by adaptively adjusting the geometric depth mapping coefficient ζ ij , singular point transformation coefficient ξ k and quantum state optimization coefficient λ l , the model responds to different input characteristics and target requirements, optimizing the predictions and analyses achieved in a variety of complex design scenarios.

[0049] Optionally, the S32 specifically includes:

[0050] S321, receiving the input data H=[H1, H2, ..., H m ], where H i represents the i-th Hodge state component, including rotation state, gradient state and harmonic state;

[0051] S322, for each Hodge state component H i Perform abnormal cycle dimension analysis using nonlinear mapping functions Extracting potentially abnormal loop patterns:

[0052]

[0053] Among them, β ik is the nonlinear weight, ω ik is the frequency parameter, φ ik is the phase shift, K i is the mode number of the i-th Hodge state component;

[0054] S323, the result after nonlinear mapping Perform abnormal cycle dimension mapping to obtain the abnormal cycle feature A of each Hodge state component i,cycle :

[0055]

[0056] Among them, P il is the projection matrix, p ir is the center point vector, γ il is the weight of the Gaussian kernel function, L i is the number of mapping dimensions of the i-th Hodge state component;

[0057] S3.2.4. For each abnormal cycle feature A i,cycle Perform multi-scale analysis and use wavelet transform to extract the cyclic pattern features at different scales to obtain the multi-scale abnormal cyclic features W i,cycle ;

[0058] S3.2.5. All the multi-scale anomalous cyclic features of the Hodge state components W i,cycle Combined into a comprehensive abnormal cycle feature vector A cycle .

[0059] Optionally, the S4 specifically includes:

[0060] S41, receiving a predetermined target performance indicator, and inputting it into a trained Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model, wherein the Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model is trained and optimized;

[0061] S42, perform fractal dimension transformation on the target performance index, and map the target performance index to a high-dimensional design space through non-equilibrium fractal Laplace transformation:

[0062]

[0063] Among them, p is the target performance index, σ f is the fractal dimension parameter, which controls the scale effect of the transformation, and Γ(·) is the extended gamma function, which includes the fractal Laplace operator. and the nonlinear attenuation term Represents time-varying fractional derivatives, capturing memory effects and nonlinear dynamic behavior of the system;

[0064] S43. Based on the mapped design space, a preliminary design scheme of the nanoscale sawtooth structure is generated by applying the quantum topological interweaving operator. The quantum topological interweaving operator optimizes the design parameters through the topological interweaving effect between quantum states:

[0065]

[0066] Where d is the design parameter ψ i and ψ j Represent different states of quantum state functions, which are used to capture the topological characteristics in the design space. is the quantum state transformation operator, which describes the topological interweaving relationship between design parameters. Tr(·) is the trace operation, which is used to extract the topological features in the design scheme. is a projection operator that captures important intertwining patterns in the quantum state, is the high-order derivative of the design parameters, which is used to describe the mutual influence between the design parameters;

[0067] S45. Output a sawtooth structure design scheme that meets predetermined performance indicators.

[0068] Optionally, the S5 specifically includes:

[0069] S51, receiving the sawtooth structure design scheme generated in S4, taking the sawtooth structure design scheme as input, and preliminarily deriving the manufacturing process parameters required to implement the design scheme through a reverse design method, wherein the process parameters include equipment settings, material processing conditions, processing path planning, and setting of process control points, so that the optimized design scheme can be implemented in the actual manufacturing process;

[0070] S52, optimize the manufacturing process parameters initially derived, using a nonlinear transformation formula G based on multidimensional geometry and time-space coupling t (x) Adjust and optimize process parameters to optimize the quality and consistency of the manufacturing process:

[0071]

[0072] Wherein, x represents the manufacturing process parameter vector, including temperature, pressure, and processing speed parameters; t is the time variable, which reflects the dynamic changes of process parameters in the manufacturing process; α is the order of the fractional derivative, which is used to capture the memory effect and nonlinear dynamic characteristics in the time dimension; λ is the adjustment coefficient, which controls the influence of Ricci curvature on process parameter optimization and helps optimize the geometric stability of the manufacturing process; Ric(g(x)), Ricci curvature, originated from Riemannian geometry, is used to describe the degree of curvature of manufacturing parameters in high-dimensional manifold space and reflects the complexity of topological characteristics in the manufacturing process; κ is the spatial adjustment coefficient, which adjusts the influence of manufacturing process in the spatial dimension; Hess(f(x)), Hessian matrix, represents the second-order partial derivative of manufacturing parameters on the objective function, which helps to identify the mutual influence and sensitivity between process parameters; Ω is the definition domain of the manufacturing process; dμ(x) is the volume element, which represents the integration on the manifold and the optimization of the entire manufacturing space; C is the integral path;

[0073] S53, generating a manufacturing instruction set M according to the optimized process parameters t ,The instruction set includes equipment settings, optimization planning of processing paths, and real-time monitoring parameters;

[0074] S54. During the manufacturing process, simulation technology is used to verify the manufacturing instruction set, simulate actual operations, detect and correct possible process deviations, and optimize the effectiveness and consistency of process parameters in actual manufacturing through simulation optimization;

[0075] S55. Apply the optimized and verified process parameters to actual manufacturing, optimize the manufacturing quality of the nano-scale sawtooth structure to meet the design requirements, and further optimize the process flow through a feedback mechanism.

[0076] Optionally, the S7 specifically includes:

[0077] S61. Collect data related to the operation of manufacturing equipment, process parameters and environmental conditions in real time during the manufacturing process. The data collected through the sensor network and monitoring system covers temperature, pressure, processing speed, vibration frequency process parameters, as well as external environmental factors such as humidity and air quality;

[0078] S62, the data collected in real time is directly input into the trained Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model;

[0079] S63, dynamically adjusting parameters in the manufacturing process according to the analysis results of the polymorphic neural network model based on Hodge state decomposition and abnormal cycle dimension analysis, adjusting the prediction results based on the model, and the adjusted parameters acting on the manufacturing equipment through a feedback mechanism;

[0080] S64. During the entire manufacturing process, all relevant data are stored in the database;

[0081] S65. By establishing a feedback loop Hodge state decomposition and abnormal cycle dimensional analysis polymorphic neural network model, it continuously self-updates and optimizes according to the latest data accumulated during the manufacturing process.

[0082] Beneficial Effects

[0083] 1. Improve design accuracy and manufacturing consistency: By utilizing Hodge state decomposition and abnormal cycle dimension analysis technology, the present invention can accurately extract and analyze the key characteristic parameters that affect the performance of nano-scale sawtooth structures. Combined with these advanced mathematical tools, the machine learning model can accurately predict the relationship between design parameters and target performance indicators, thereby generating highly optimized nanostructure design solutions. During the manufacturing process, by real-time monitoring and dynamic adjustment of process parameters, manufacturing accuracy and structural consistency are further ensured, effectively solving the problem of the manufacturing process relying on experience and low consistency in the prior art.

[0084] 2. Improve the automation and intelligence of the manufacturing process: The present invention realizes real-time data analysis and dynamic adjustment in the manufacturing process by constructing and training a complex polymorphic neural network model. Combined with the steps in the claims, the present invention can automatically collect data, analyze anomalies and adjust process parameters in real time during the manufacturing process, making the manufacturing process more intelligent, reducing dependence on manual intervention, and significantly improving the automation level and efficiency of manufacturing.

[0085] 3. Optimize the derivation and implementation of manufacturing process parameters: Through reverse design and multi-dimensional geometric optimization technology, the present invention can accurately derive the specific process parameters required for manufacturing the sawtooth structure and optimize and implement them in actual manufacturing. This method not only ensures the manufacturability of the design, but also further improves the reliability and accuracy of the manufacturing process through complex time-space coupling optimization, making the manufacturing of high-precision nanostructures efficient.

[0086] 4. Enhance the adaptability and robustness of the model: The Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model in the present invention can not only provide optimization suggestions in the design stage, but also continuously update and optimize itself through the feedback mechanism during the manufacturing process. In this way, the model can continue to improve with the accumulation of data and changes in manufacturing conditions, improving its ability to adapt to complex manufacturing environments and robustness.

[0087] 5. Realize an efficient closed-loop feedback mechanism: By feeding back the performance test results in the manufacturing process to the Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model, the present invention can achieve closed-loop optimization. This feedback mechanism helps to continuously improve the design and manufacturing process, so that each production can provide data support for the next manufacturing, forming an efficient self-improvement system.

[0088] In summary, the present invention not only improves the design accuracy through precise mathematical modeling and optimization algorithms in the design stage, but also ensures the stability and high precision of the manufacturing process through real-time monitoring and dynamic adjustment, ultimately achieving higher manufacturing consistency and automation level, and has significant industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0090] Figure 1 A summary diagram of a nano-scale sawtooth design and manufacturing method based on machine learning proposed by the present invention;

[0091] Figure 2 The figure of claim 3 of the nano-scale sawtooth design and manufacturing method based on machine learning proposed by the present invention. DETAILED DESCRIPTION

[0092] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0093] refer to Figure 1-2 , a nano-scale sawtooth design and manufacturing method based on machine learning, comprising the following steps:

[0094] S1. Collect and process initial data related to the nano-scale sawtooth structure, including geometric parameters, material properties, process parameters, and target performance indicators;

[0095] S2. Analyze and extract features from the processed data using feature engineering, extract initial feature vectors from the initial data, map the initial feature vectors to a high-dimensional latent space, capture the abnormal cycle patterns and dimensional structures hidden in the data, and calculate the characteristic parameters that affect the performance of the sawtooth structure through multi-scale periodic analysis and singular value decomposition;

[0096] S3. Select and build a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model, train the extracted feature parameters, and predict the complex nonlinear relationship between design parameters and target performance indicators through geometric depth mapping, singular point topological transformation and quantum state optimization process;

[0097] S4. Apply the trained polymorphic neural network model for constructing Hodge state decomposition and abnormal cycle dimension analysis, generate a nanoscale sawtooth structure design scheme according to the predetermined target performance indicators, optimize the design parameters through non-equilibrium fractal Laplace transform and quantum topological interleaving operator, and output a design scheme that meets the predetermined performance requirements;

[0098] S5. Perform reverse design on the generated design scheme to derive the manufacturing process parameters required to implement the design scheme;

[0099] S6. During the manufacturing process, real-time monitoring is performed based on the derived manufacturing process parameters, and dynamic adjustments are made by combining the construction of Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network models to optimize manufacturing accuracy and consistency;

[0100] S7. After manufacturing is completed, the performance of the produced nano-scale sawtooth structure is tested, and the test results are fed back to the construction of the Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to update and optimize the model.

[0101] In this implementation, S2 specifically includes:

[0102] S21, receiving the initial feature vector Z extracted from the original data = [z1, z2,, z p ] and mapped to a high-dimensional latent space To capture the abnormal cycle patterns and dimensional structures implicit in the data, the mapping is defined by the abnormal cycle mapping function C(·):

[0103]

[0104] Among them, α i is the mapping weight, W i is the weight matrix, C latent It is the latent space representation after mapping, aiming to reveal the nonlinear cyclic characteristics in the data;

[0105] S22, the latent space after mapping is represented by C latent Perform multi-scale periodic analysis using multi-scale transformation S j (·) Decompose the cyclic patterns in the latent space and extract abnormal features at different scales:

[0106]

[0107] Among them, β jk is the multi-scale analysis coefficient, ω jk is the frequency parameter, is the phase shift, q j is the number of patterns at the jth scale;

[0108] S23 The abnormal cycle characteristics obtained by multi-scale periodic analysis Combined into a multi-dimensional abnormal cycle feature vector A cycle ;

[0109] S24, for the comprehensive abnormal cycle feature vector A cycle Perform singular value decomposition to decompose the singular value σ that dominates the abnormal cycle characteristics k and the corresponding singular vector u k ;

[0110] S25, the main abnormal cycle feature A after singular value decomposition cycle as output.

[0111] In this implementation, S3 specifically includes:

[0112] S31, receiving the Hodge state component H output by the Hodge state decomposition layer in claim 2 = [H1, H2,, H m ] and the abnormal cycle feature A output by the abnormal cycle dimension analysis layer in claim 3 cycle =[A 1,cycle ,A 2,cycle ,,An,cycle ], taking the feature parameters as input and applying them to the multidimensional nonlinear mapping space Construct a high-dimensional mapping relationship between the input feature space and the target performance index,

[0113] S32. Construct a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to generate abnormal cycle feature vectors based on high-dimensional mapping relationships;

[0114] S33, perform geometric depth mapping on the input abnormal cycle feature vector, using a high-order surface fitting function G i (·) Perform nonlinear mapping on Hodge state components and abnormal cycle features to generate high-dimensional geometric feature representation F geo :

[0115]

[0116] Among them, ij is the high-dimensional geometric mapping coefficient, ∧ represents the outer product operation, Q ij is a symmetric positive definite matrix. The mapping constructs a feature representation containing complex geometric and topological information by combining the outer product and the exponential function. The abnormal cycle feature A cycle ;

[0117] S34, geometric feature representation F geo Perform singular point topological transformation and introduce the singular point transformation function S based on high-order differential geometry k (·), identifying and amplifying singularities and topological anomalies in high-dimensional latent spaces:

[0118]

[0119] Among them, ξ k is the singular point transformation coefficient, represents the generalized inverse, is the k-th order derivative operator, and the high-dimensional geometric feature representation F geo ;

[0120] S35. The characteristic T after the topological transformation of the singular point through quantum state topological optimization singular Perform quantum state optimization to generate optimized feature representation T quantum ;

[0121] S36. Using high-order generalized convolution to optimize the feature representation of quantum state T quantum Perform convolution operation to generate convolution feature representation F conv :

[0122]

[0123] Among them, ν tis the convolution kernel coefficient, C t (·) is a high-order generalized convolution operation, which extracts nonlinear and multi-scale information in features through multi-order convolution operations and generates deep feature representations for prediction. r: the total number of convolution operations, indicating how many different convolution operations are used to enrich feature extraction.

[0124] S37, through symbolic regression and multi-state prediction model, the convolution feature representation F conv Convert to symbolic prediction expression Generate predictions between design parameters and target performance indicators

[0125]

[0126] in, The final prediction result represents the relationship between the design parameters and the target performance indicators. The P(·) symbolic regression and multi-state prediction model transforms the numerical features into interpretable symbolic expressions. Symbolic regression coefficients adjust the contribution of different features to the prediction results, High-order nonlinear terms represent complex nonlinear relationships in features and enhance the predictive power of the model. The order of the s-symbol regression indicates how many regression terms of different orders are included.

[0127] S38, by adaptively adjusting the geometric depth mapping coefficient ζ ij , singular point transformation coefficient ξ k and quantum state optimization coefficient λ l , the model responds to different input characteristics and target requirements, optimizing the predictions and analyses achieved in a variety of complex design scenarios.

[0128] In this implementation manner, the S32 specifically includes:

[0129] S321, receiving the input data H=[H1, H2, ..., H m ], where H i represents the i-th Hodge state component, including rotation state, gradient state and harmonic state;

[0130] S322, for each Hodge state component H i Perform abnormal cycle dimension analysis using nonlinear mapping functions Extracting potentially abnormal loop patterns:

[0131]

[0132] Among them, β ik is the nonlinear weight, ω ik is the frequency parameter, φ ikis the phase shift, K i is the mode number of the i-th Hodge state component;

[0133] S323, the result after nonlinear mapping Perform abnormal cycle dimension mapping to obtain the abnormal cycle feature A of each Hodge state component i,cycle :

[0134]

[0135] Among them, P il is the projection matrix, p ir is the center point vector, γ il is the weight of the Gaussian kernel function, L i is the number of mapping dimensions of the i-th Hodge state component;

[0136] S3.2.4. For each abnormal cycle feature A i,cycle Perform multi-scale analysis and use wavelet transform to extract the cyclic pattern features at different scales to obtain the multi-scale abnormal cyclic features W i,cycle ;

[0137] S3.2.5. All the multi-scale anomalous cyclic features of the Hodge state components W i,cycle Combined into a comprehensive abnormal cycle feature vector A cycle .

[0138] In this implementation, S4 specifically includes:

[0139] S41, receiving a predetermined target performance indicator, and inputting it into a trained Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model, wherein the Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model is trained and optimized;

[0140] S42, perform fractal dimension transformation on the target performance index, and map the target performance index to a high-dimensional design space through non-equilibrium fractal Laplace transformation:

[0141]

[0142] Among them, p is the target performance index, σ f is the fractal dimension parameter, which controls the scale effect of the transformation, and Γ(·) is the extended gamma function, which includes the fractal Laplace operator. and the nonlinear attenuation term Represents time-varying fractional derivatives, capturing memory effects and nonlinear dynamic behavior of the system;

[0143] S43. Based on the mapped design space, a preliminary design scheme of the nanoscale sawtooth structure is generated by applying the quantum topological interweaving operator. The quantum topological interweaving operator optimizes the design parameters through the topological interweaving effect between quantum states:

[0144]

[0145] Where d is the design parameter ψ i and ψ j Represent different states of quantum state functions, which are used to capture the topological characteristics in the design space. is the quantum state transformation operator, which describes the topological interweaving relationship between design parameters. Tr(·) is the trace operation, which is used to extract the topological features in the design scheme. is a projection operator that captures important intertwining patterns in the quantum state, is the high-order derivative of the design parameters, which is used to describe the mutual influence between the design parameters;

[0146] S45. Output a sawtooth structure design scheme that meets predetermined performance indicators.

[0147] In this implementation manner, S5 specifically includes:

[0148] S51, receiving the sawtooth structure design scheme generated in S4, taking the sawtooth structure design scheme as input, and preliminarily deriving the manufacturing process parameters required to implement the design scheme through a reverse design method, wherein the process parameters include equipment settings, material processing conditions, processing path planning, and setting of process control points, so that the optimized design scheme can be implemented in the actual manufacturing process;

[0149] S52, optimize the manufacturing process parameters initially derived, using a nonlinear transformation formula G based on multidimensional geometry and time-space coupling t (x) Adjust and optimize process parameters to optimize the quality and consistency of the manufacturing process:

[0150]

[0151] Wherein, x represents the manufacturing process parameter vector, including temperature, pressure, and processing speed parameters; t is the time variable, which reflects the dynamic changes of process parameters in the manufacturing process; α is the order of the fractional derivative, which is used to capture the memory effect and nonlinear dynamic characteristics in the time dimension; λ is the adjustment coefficient, which controls the influence of Ricci curvature on process parameter optimization and helps optimize the geometric stability of the manufacturing process; Ric(g(x)), Ricci curvature, originated from Riemannian geometry, is used to describe the degree of curvature of manufacturing parameters in high-dimensional manifold space and reflects the complexity of topological characteristics in the manufacturing process; κ is the spatial adjustment coefficient, which adjusts the influence of manufacturing process in the spatial dimension; Hess(f(x)), Hessian matrix, represents the second-order partial derivative of manufacturing parameters on the objective function, which helps to identify the mutual influence and sensitivity between process parameters; Ω is the definition domain of the manufacturing process; dμ(x) is the volume element, which represents the integration on the manifold and the optimization of the entire manufacturing space; C is the integral path;

[0152] S53, generating a manufacturing instruction set M according to the optimized process parameters t ,The instruction set includes equipment settings, optimization planning of processing paths, and real-time monitoring parameters;

[0153] S54. During the manufacturing process, simulation technology is used to verify the manufacturing instruction set, simulate actual operations, detect and correct possible process deviations, and optimize the effectiveness and consistency of process parameters in actual manufacturing through simulation optimization;

[0154] S55. Apply the optimized and verified process parameters to actual manufacturing, optimize the manufacturing quality of the nano-scale sawtooth structure to meet the design requirements, and further optimize the process flow through a feedback mechanism.

[0155] In this implementation manner, the S7 specifically includes:

[0156] S61. Collect data related to the operation of manufacturing equipment, process parameters and environmental conditions in real time during the manufacturing process. The data collected through the sensor network and monitoring system covers temperature, pressure, processing speed, vibration frequency process parameters, as well as external environmental factors such as humidity and air quality;

[0157] S62, the data collected in real time is directly input into the trained Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model;

[0158] S63, dynamically adjusting parameters in the manufacturing process according to the analysis results of the polymorphic neural network model based on Hodge state decomposition and abnormal cycle dimension analysis, adjusting the prediction results based on the model, and the adjusted parameters acting on the manufacturing equipment through a feedback mechanism;

[0159] S64. During the entire manufacturing process, all relevant data are stored in the database;

[0160] S65. By establishing a feedback loop Hodge state decomposition and abnormal cycle dimensional analysis polymorphic neural network model, it continuously self-updates and optimizes according to the latest data accumulated during the manufacturing process.

[0161] Example

[0162] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a micro-nano manufacturing plant in a certain city. In the field of high-precision micro-nano electronic device manufacturing, especially in the manufacture of tiny structures in integrated circuits, the precise manufacture of nanoscale sawtooth structures is crucial to the performance of the device. These structures are generally used to improve the flow characteristics of electric currents, enhance the optical properties of photonic crystals, or optimize the flow characteristics of microfluidic channels. However, traditional manufacturing technologies have major deficiencies in accuracy and consistency, resulting in fluctuations in product performance and may even lead to manufacturing failures. In order to solve these problems, we conducted optimization experiments on the process of producing nanoscale sawtooth structures in a micro-nano manufacturing plant located in Shenzhen. The experimental time span is three months, aiming to verify the effectiveness of the design and manufacturing method based on machine learning proposed in the present invention.

[0163] In the experiment, we selected the nanoscale sawtooth structure commonly used in integrated circuits as the test object. This structure has complex geometry and strict dimensional tolerance requirements. We first collected historical data accumulated in the factory's previous production process, including geometric parameters, material properties, process parameters, and target performance indicators. Then, we analyzed and extracted features from this data through feature engineering to identify the key parameters that have the greatest impact on the performance of the sawtooth structure.

[0164] Based on these key parameters, we built a polymorphic neural network model and trained it using Hodge state decomposition and abnormal cycle dimension analysis to enable it to accurately predict the relationship between design parameters and target performance. Once the model is trained, we apply it to the production line to monitor and adjust process parameters in real time during the manufacturing process.

[0165] In the specific application, we first used the trained Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to generate an optimized nanoscale sawtooth structure design scheme based on the target performance indicators. This design scheme is more accurate than the scheme generated by traditional methods and can better meet the performance requirements of the product. Then, we reverse engineered the generated design scheme and derived the specific manufacturing process parameters required to implement the design. In order to ensure the optimality of the process parameters, we further refined and adjusted them using a multidimensional geometric optimization algorithm.

[0166] During the production process, we collect real-time operating data of manufacturing equipment through sensor networks, including temperature, pressure, processing speed and vibration frequency, as well as external environmental data, humidity and air quality. All of this data is input into the Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model, which analyzes this data in real time, identifies any possible anomalies or deviations, and automatically adjusts process parameters to ensure that the manufacturing process is always in the best state.

[0167] Through this series of intelligent adjustments, the factory successfully produced a batch of high-precision nano-scale serrated structure products within one month, and the geometric accuracy and consistency of the products reached the expected goals.

[0168] Table 1: Manufacturing accuracy comparison

[0169] Manufacturing method Average size deviation (nm) Improvement in geometric accuracy (%) Traditional crafts ±15 - Method of the present invention ±5 66.7

[0170] Table 2: Manufacturing consistency comparison

[0171]

[0172] Table 3: Production efficiency comparison

[0173] Manufacturing method Average debugging time (hours) Production efficiency improvement (%) Traditional crafts 3 - Method of the present invention 1 66

[0174] Table 4: Process stability comparison

[0175] Manufacturing method Parameter adjustment frequency Sensitivity to environmental impact Stability improvement (%) Traditional crafts high high - Method of the present invention Low Low 80

[0176] Table 5: Feedback optimization effect

[0177] Optimization rounds Dimensional deviation reduction (nanometers) Product qualification rate improvement (%) First optimization 1.0 0.5 Second optimization 0.5 0.5

[0178] It can be seen that, by comparing the production results of the traditional process with that of the method of the present invention, it is found that the present invention significantly improves the manufacturing accuracy of the nano-scale sawtooth structure, reduces the average size deviation from ±15 nanometers to ±5 nanometers, and improves the accuracy by about 66.7%; the product qualification rate is increased from 85% to 98%, and the consistency is significantly improved; the production efficiency is increased by about 66%, and the average debugging time is shortened from 3 hours to 1 hour; the process stability is enhanced, the number of parameter adjustments is reduced by about 80%, and the sensitivity to environmental changes is reduced; in addition, through feedback optimization, the size deviation is further reduced by 1.5 nanometers, and the qualification rate is increased by 1%.

[0179] Through the application of the method of the present invention, the micro-nano manufacturing factory has greatly improved the manufacturing accuracy and consistency of the products in the production of nano-scale sawtooth structures, and also significantly improved the production efficiency and process stability. The results fully prove the significant advantages of the present invention in practical applications, and provide an efficient and reliable solution for the field of high-precision nano-manufacturing.

[0180] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A nano-scale sawtooth design and manufacturing method based on machine learning, characterized in that: The steps include: S1. Collect and process initial data related to the nano-scale sawtooth structure, including geometric parameters, material properties, process parameters, and target performance indicators; S2. Analyze and extract features from the processed data using feature engineering, extract initial feature vectors from the initial data, map the initial feature vectors to a high-dimensional latent space, capture the abnormal cycle patterns and dimensional structures hidden in the data, and calculate the characteristic parameters that affect the performance of the sawtooth structure through multi-scale periodic analysis and singular value decomposition; S3. Select and build a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model, train the extracted feature parameters, and predict the complex nonlinear relationship between design parameters and target performance indicators through geometric depth mapping, singular point topological transformation and quantum state optimization process; S4. Apply the trained polymorphic neural network model for constructing Hodge state decomposition and abnormal cycle dimension analysis, generate a nanoscale sawtooth structure design scheme according to the predetermined target performance indicators, optimize the design parameters through non-equilibrium fractal Laplace transform and quantum topological interleaving operator, and output a design scheme that meets the predetermined performance requirements; S5. Perform reverse design on the generated design scheme to derive the manufacturing process parameters required to implement the design scheme; S6. During the manufacturing process, real-time monitoring is performed based on the derived manufacturing process parameters, and dynamic adjustments are made by combining the construction of Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network models to optimize manufacturing accuracy and consistency; S7. After manufacturing is completed, the performance of the produced nano-scale sawtooth structure is tested, and the test results are fed back to the construction of the Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to update and optimize the model.

2. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S2 specifically includes: S21, receiving the initial feature vector Z extracted from the original data = [z1, z2,, z p ] and mapped to a high-dimensional latent space To capture the abnormal cycle patterns and dimensional structures implicit in the data, the mapping is defined by the abnormal cycle mapping function C(·): Among them, α i is the mapping weight, W i is the weight matrix, C latent It is the latent space representation after mapping, aiming to reveal the nonlinear cyclic characteristics in the data; S22, the latent space after mapping is represented by C latent Perform multi-scale periodic analysis using multi-scale transformation S j (·) Decompose the cyclic patterns in the latent space and extract abnormal features at different scales: Among them, β jk is the multi-scale analysis coefficient, ω jk is the frequency parameter, is the phase shift, q j is the number of patterns at the jth scale; S23 The abnormal cycle characteristics obtained by multi-scale periodic analysis Combined into a multi-dimensional abnormal cycle feature vector A cycle ; S24, for the comprehensive abnormal cycle feature vector A cycle Perform singular value decomposition to decompose the singular value σ that dominates the abnormal cycle characteristics k and the corresponding singular vector u k ; S25, the main abnormal cycle feature A after singular value decomposition cycle as output.

3. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S3 specifically includes: S31, receiving the Hodge state component H output by the Hodge state decomposition layer in claim 2 = [H1, H2,, H m ] and the abnormal cycle feature A output by the abnormal cycle dimension analysis layer in claim 3 cycle =[A 1,cycle ,A 2,cycle ,,A n,cycle ], taking the feature parameters as input and applying them to the multidimensional nonlinear mapping space Construct a high-dimensional mapping relationship between the input feature space and the target performance index, S32. Construct a Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model to generate abnormal cycle feature vectors based on high-dimensional mapping relationships; S33, perform geometric depth mapping on the input abnormal cycle feature vector, using a high-order surface fitting function G i (·) Perform nonlinear mapping on Hodge state components and abnormal cycle features to generate high-dimensional geometric feature representation F geo : Among them, ij is the high-dimensional geometric mapping coefficient, ∧ represents the outer product operation, Q ij is a symmetric positive definite matrix. The mapping constructs a feature representation containing complex geometric and topological information by combining the outer product and the exponential function. The abnormal cycle feature A cycle ; S34, geometric feature representation F geo Perform singular point topological transformation and introduce the singular point transformation function S based on high-order differential geometry k (·), identifying and amplifying singularities and topological anomalies in high-dimensional latent spaces: Among them, ξ k is the singular point transformation coefficient, represents the generalized inverse, is the k-th order derivative operator, and the high-dimensional geometric feature representation F geo ; S35. The characteristic T after the topological transformation of the singular point through quantum state topological optimization singular Perform quantum state optimization to generate optimized feature representation T quantum ; S36. Using high-order generalized convolution to optimize the feature representation of quantum state T quantum Perform convolution operation to generate convolution feature representation F conv : Among them, v t is the convolution kernel coefficient, C t (·) is a high-order generalized convolution operation, which extracts nonlinear and multi-scale information in features through multi-order convolution operations and generates deep feature representations for prediction. r: the total number of convolution operations, indicating how many different convolution operations are used to enrich feature extraction. S37, through symbolic regression and multi-state prediction model, the convolution feature representation F conv Convert to symbolic prediction expression Generate predictions between design parameters and target performance indicators in, The final prediction result represents the relationship between the design parameters and the target performance indicators. The P(·) symbolic regression and multi-state prediction model transforms the numerical features into interpretable symbolic expressions. Symbolic regression coefficients adjust the contribution of different features to the prediction results, High-order nonlinear terms represent complex nonlinear relationships in features and enhance the predictive power of the model. The order of the s-symbol regression indicates how many regression terms of different orders are included. S38, by adaptively adjusting the geometric depth mapping coefficient ζ ij , singular point transformation coefficient ξ k and quantum state optimization coefficient λ l , the model responds to different input characteristics and target requirements, optimizing the predictions and analyses achieved in a variety of complex design scenarios.

4. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S32 specifically includes: S321, receiving the input data H after being processed by the Hodge state decomposition layer in claim 2 = [H1, H2, ..., H m ], where H i represents the i-th Hodge state component, including rotation state, gradient state and harmonic state; S322, for each Hodge state component H i Perform abnormal cycle dimension analysis using nonlinear mapping functions Extracting potentially abnormal loop patterns: Among them, β ik is the nonlinear weight, ω ik is the frequency parameter, φ ik is the phase shift, K i is the mode number of the i-th Hodge state component S323, the result after nonlinear mapping Perform abnormal cycle dimension mapping to obtain the abnormal cycle feature A of each Hodge state component i，cycle : Among them, P il is the projection matrix, p ir is the center point vector, γ il is the weight of the Gaussian kernel function, L i is the number of mapping dimensions of the i-th Hodge state component; S3.2.

4. For each abnormal cycle feature A i，cycle Perform multi-scale analysis and use wavelet transform to extract the cyclic pattern features at different scales to obtain the multi-scale abnormal cyclic features W i，cycle ; S3.2.

5. All the multi-scale anomalous cyclic features of the Hodge state components W i，cycle Combined into a comprehensive abnormal cycle feature vector A cycle .

5. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S4 specifically includes: S41, receiving a predetermined target performance indicator, and inputting it into a trained Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model, wherein the Hodge state decomposition and abnormal cycle dimensionality analysis polymorphic neural network model is trained and optimized; S42, perform fractal dimension transformation on the target performance index, and map the target performance index to a high-dimensional design space through non-equilibrium fractal Laplace transformation: Among them, p is the target performance index, σ f is the fractal dimension parameter, which controls the scale effect of the transformation, and Γ(·) is the extended gamma function, which includes the fractal Laplace operator. and the nonlinear attenuation term Represents time-varying fractional derivatives, capturing memory effects and nonlinear dynamic behavior of the system; S43. Based on the mapped design space, a preliminary design scheme of the nanoscale sawtooth structure is generated by applying the quantum topological interweaving operator. The quantum topological interweaving operator optimizes the design parameters through the topological interweaving effect between quantum states: Where d is the design parameter ψ i and ψ j Represent different states of quantum state functions, which are used to capture the topological characteristics in the design space. is the quantum state transformation operator, which describes the topological interweaving relationship between design parameters. Tr(·) is the trace operation, which is used to extract the topological features in the design scheme. is a projection operator that captures important intertwining patterns in the quantum state, is the high-order derivative of the design parameters, which is used to describe the mutual influence between the design parameters; S45. Output a sawtooth structure design scheme that meets predetermined performance indicators.

6. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S5 specifically includes: S51, receiving the sawtooth structure design scheme generated in S4, taking the sawtooth structure design scheme as input, and preliminarily deriving the manufacturing process parameters required to implement the design scheme through a reverse design method, wherein the process parameters include equipment settings, material processing conditions, processing path planning, and setting of process control points, so that the optimized design scheme can be implemented in the actual manufacturing process; S52, optimize the manufacturing process parameters initially derived, using a nonlinear transformation formula G based on multidimensional geometry and time-space coupling t (x) Adjust and optimize process parameters to optimize the quality and consistency of the manufacturing process: Wherein, x represents the manufacturing process parameter vector, including temperature, pressure, and processing speed parameters; t is the time variable, which reflects the dynamic changes of process parameters in the manufacturing process; α is the order of the fractional derivative, which is used to capture the memory effect and nonlinear dynamic characteristics in the time dimension; λ is the adjustment coefficient, which controls the influence of Ricci curvature on process parameter optimization and helps optimize the geometric stability of the manufacturing process; Ric(g(x)), Ricci curvature, originated from Riemannian geometry, is used to describe the degree of curvature of manufacturing parameters in high-dimensional manifold space and reflects the complexity of topological characteristics in the manufacturing process; κ is the spatial adjustment coefficient, which adjusts the influence of manufacturing process in the spatial dimension; Hess(f(x)), Hessian matrix, represents the second-order partial derivative of manufacturing parameters on the objective function, which helps to identify the mutual influence and sensitivity between process parameters; Ω is the definition domain of the manufacturing process; dμ(x) is the volume element, which represents the integration on the manifold and the optimization of the entire manufacturing space; C is the integral path; S53, generating a manufacturing instruction set M according to the optimized process parameters t ,The instruction set includes equipment settings, optimization planning of processing paths, and real-time monitoring parameters; S54. During the manufacturing process, simulation technology is used to verify the manufacturing instruction set, simulate actual operations, detect and correct possible process deviations, and optimize the effectiveness and consistency of process parameters in actual manufacturing through simulation optimization; S55. Apply the optimized and verified process parameters to actual manufacturing, optimize the manufacturing quality of the nano-scale sawtooth structure to meet the design requirements, and further optimize the process flow through a feedback mechanism.

7. The method for designing and manufacturing nano-scale sawtooth based on machine learning according to claim 1, characterized in that: The S7 specifically includes: S61. Collect data related to the operation of manufacturing equipment, process parameters and environmental conditions in real time during the manufacturing process. The data collected through the sensor network and monitoring system covers temperature, pressure, processing speed, vibration frequency process parameters, as well as external environmental factors such as humidity and air quality; S62, the data collected in real time is directly input into the trained Hodge state decomposition and abnormal cycle dimension analysis polymorphic neural network model; S63, dynamically adjusting parameters in the manufacturing process according to the analysis results of the polymorphic neural network model based on Hodge state decomposition and abnormal cycle dimension analysis, adjusting the prediction results based on the model, and the adjusted parameters acting on the manufacturing equipment through a feedback mechanism; S64. During the entire manufacturing process, all relevant data are stored in the database; S65. By establishing a feedback loop Hodge state decomposition and abnormal cycle dimensional analysis polymorphic neural network model, it continuously self-updates and optimizes according to the latest data accumulated during the manufacturing process.

Images