Autonomous optimization system and method for parameters of metal ceramic bionic multilayer film

Abstract

The invention relates to a cermet bionic multilayer film parameter autonomous optimization system and method, and belongs to the technical field of cermet bionic multilayer film research and development, the cermet bionic multilayer film parameter autonomous optimization system comprises a data management module, a machine learning agent model module, an optimization algorithm module, an automatic experiment platform and a central control module, the central control module is in communication connection with the data management module, the machine learning agent model module, the optimization algorithm module and the automatic experiment platform, and is used for controlling data streams and instruction streams among the modules and driving the whole autonomous optimization loop iteration. According to the method, unmanned operation is realized by constructing a closed loop of'machine learning-optimization algorithm-automatic experiment ', the traditional research and development period needing several years is shortened to several months and even several weeks, meanwhile, a large amount of labor, material and energy cost is saved, and the method is more efficient and more reliable by adopting the combination of Gaussian process regression and Bayesian optimization algorithm. And the prediction mean value and the uncertainty can be utilized at the same time.

Description

Technical Field

[0001] This invention relates to the field of metal-ceramic biomimetic multilayer film research and development technology, and in particular to a system and method for autonomous optimization of parameters of metal-ceramic biomimetic multilayer films. Background Technology

[0002] Metal-ceramic biomimetic multilayer films are artificial synthetic materials inspired by natural biomaterials (such as nacre in seashells). They are created by stacking tough metals and hard ceramics alternately at the nanometer or micrometer scale using advanced thin film fabrication techniques (such as magnetron sputtering) to form a periodic multilayer film structure with a large number of interfaces. This structure cleverly combines the advantages of both metals and ceramics, aiming to achieve high strength, high hardness, high toughness, good wear resistance, and fatigue resistance that are difficult to achieve with a single material. It is widely used in precision tool coatings, protection of critical aerospace components, and other fields. Its performance depends to a great extent on the structural parameters of the multilayer film (such as single-layer thickness, modulation ratio, material system, total number of layers, etc.) and process parameters (such as deposition power, temperature, gas pressure, etc.).

[0003] Currently, the design and development of this type of material mainly relies on researchers' experience and the traditional "trial and error" method, which has the following inherent drawbacks:

[0004] 1. Low efficiency: The parameter space is vast, the experimental cycle is long, and the cost is high;

[0005] 2. Prone to local optima: Relying on human experience makes it difficult to find the global optimum in a complex multidimensional parameter space;

[0006] 3. High dependence on human resources: The entire process relies heavily on the intuition and judgment of senior experts, resulting in poor replicability and universality;

[0007] To address this, we propose a system and method for autonomous optimization of parameters in biomimetic multilayer metal ceramic films. By constructing a closed loop of "machine learning-optimization algorithm-automated experiment", we can achieve fully unmanned operation from intelligent design and automated preparation to performance characterization, thereby efficiently and accurately discovering the optimal parameter formulation. Summary of the Invention

[0008] This invention provides a system and method for autonomous optimization of parameters of biomimetic multilayer metal ceramic films, which solves the problems of low research and development efficiency, difficulty in obtaining the global optimal solution, and difficulty in multi-parameter collaborative optimization in the traditional research and development field of biomimetic multilayer metal ceramic films.

[0009] The solution of the present invention to solve the above-mentioned technical problems is as follows: a metal-ceramic biomimetic multilayer film parameter autonomous optimization system and method, including a data management module, a machine learning agent model module, an optimization algorithm module, an automated experimental platform, and a central control module. The central control module is communicatively connected to the data management module, the machine learning agent model module, the optimization algorithm module, and the automated experimental platform, respectively, and is used to control the data flow and instruction flow between the modules, and drive the entire autonomous optimization loop to iterate.

[0010] The optimization method includes the following steps:

[0011] S1, Initialize the dataset: Enter the initial dataset in the data management module;

[0012] S2, Model Training and Update: The machine learning agent model module trains or updates the agent model based on the current dataset, establishing a mapping relationship between input parameters and output performance.

[0013] S3, Parameter optimization and recommendation: The optimization algorithm module calculates and recommends one or a set of optimal candidate parameter combinations in the parameter space based on the current agent model and the preset optimization objective.

[0014] S4, fully automated experiment execution: The central control module sends the recommended candidate parameters to the automated experimental platform. The automated preparation unit prepares the sample according to the parameters, and then the automated characterization unit tests the sample to obtain performance data, which is then transmitted to the central control module.

[0015] S5, Data entry and loop termination determination: Store the parameter-performance data pairs of the new experiment into the data management module to form an expanded dataset. The data is sent back to the central control module to determine whether the termination condition is met. If not, return to step S2 and proceed to the next iteration; if the condition is met, terminate the loop and output the optimal parameter combination and the corresponding performance.

[0016] Based on the above technical solution, the present invention can be further improved as follows.

[0017] Furthermore, the data management module uses an SQL database to store and manage historical and real-time generated datasets. These datasets include input parameters and corresponding performance metrics. The use of a mature SQL database (such as MySQL or PostgreSQL) ensures the reliability, security, and consistency of data management, preventing data loss or corruption, and conforming to industry standards. The SQL database provides structured, relational data management capabilities, enabling efficient and stable storage and querying of complex experimental data.

[0018] Furthermore, the input parameters include at least one of the following: material system, metal layer thickness, ceramic layer thickness, modulation cycle, number of layers, deposition power, deposition temperature, and working gas pressure. The performance indicators include at least one of the following: hardness, toughness, wear resistance, and film-substrate adhesion. All the listed parameters are key performance indicators that affect the practical application of metal-ceramic multilayer films.

[0019] Furthermore, the machine learning surrogate model module uses a Gaussian Process Regression (GPR) model as a surrogate model. This module is connected to the data management module and trains the machine learning model based on existing datasets to learn the complex nonlinear mapping relationship between input parameters and performance indicators. It is also capable of predicting performance for new parameter combinations. GPR is particularly suitable for handling small-sample, high-cost experimental data optimization problems. It not only provides predicted values ​​but also estimates of prediction uncertainty, which provides crucial information for subsequent Bayesian optimization and forms the basis for achieving a "balanced exploration and utilization." Compared to models using ordinary neural networks, which can only provide point estimates, it offers better predictive certainty.

[0020] Furthermore, the optimization algorithm module employs a Bayesian optimization algorithm to maximize (hardness × H) 3 / E 2 As the optimization objective, the optimization algorithm module is connected to the machine learning surrogate model module to set the optimization objective (such as maximizing the product of hardness and toughness). Based on the prediction results of the surrogate model, the optimization algorithm (such as Bayesian optimization and genetic algorithm) searches and recommends the next batch of parameter combinations that are expected to improve the target performance in the parameter space. This process balances the exploration of unknown parameter regions and the utilization of existing high-performance regions, and uses the uncertainty of the GPR model to intelligently select the next most valuable experimental point.

[0021] Furthermore, the automated experimental platform includes an automated preparation unit and an automated characterization unit. The automated preparation unit is a multi-target magnetron sputtering system that can switch between Cu (metal) targets and TiN (ceramic) targets through program control, and precisely control the deposition time of each layer to adjust the thickness, power, and gas pressure. The automated characterization unit is a nanoindenter with an integrated robotic arm. The robotic arm automatically transfers the sample from the preparation chamber to the testing stage for testing and feeds the results back to the central control module.

[0022] Furthermore, the central control module is an industrial computer that runs control software and coordinates the work of each module, making it suitable for long-term continuous operation in laboratory or even future industrial environments.

[0023] Furthermore, in step S1, the initial dataset is derived from historical databases, literature data, or pre-designed initial experimental data.

[0024] Furthermore, in step S3, the optimization objective is to maximize / minimize a single performance index or to maximize a weighted comprehensive objective of multiple performance indexes. When recommending candidate parameter combinations, it is necessary to balance exploring the unknown parameter region and utilizing the existing high-performance region.

[0025] Furthermore, in step S5, the termination condition is at least one of the following: the performance index reaches a predetermined threshold, the number of iterations reaches an upper limit, or the performance improvement is below the tolerance for several consecutive generations, making it suitable for long-term continuous operation in laboratory or even future industrial environments.

[0026] The beneficial effects of this invention are as follows: This invention provides a system and method for autonomous optimization of parameters of biomimetic multilayer films in metal ceramics, which has the following advantages:

[0027] 1. Solve the problem of low R&D efficiency in the traditional field of biomimetic multilayer metal-ceramic films. The traditional "trial and error method" relies heavily on the experience and intuition of R&D personnel, and requires a large number of time-consuming, material-consuming, and energy-consuming experiments in a vast parameter space, resulting in a long R&D cycle and high cost for new materials. By constructing a closed loop of "machine learning-optimization algorithm-automated experiment", unmanned operation is achieved, shortening the traditional R&D cycle of several years to several months or even several weeks, while saving a lot of labor, material and energy costs.

[0028] 2. To address the problem of obtaining the global optimal solution in the traditional field of metal-ceramic biomimetic multilayer film research and development, manual experimental design is prone to getting stuck in the local performance optimal region, making it difficult to effectively explore the broad and non-intuitive parameter combinations and discover the global optimal formula with better performance, thus restricting the breakthrough of the upper limit of material performance. The system adopts a combination of Gaussian process regression (GPR) and Bayesian optimization algorithm. Its core advantage is that it can simultaneously utilize the predicted mean and uncertainty. The system not only "utilizes" the currently known high-performance region, but also actively "explores" the unknown but potentially huge parameter space, thus having a greater probability of finding the global optimal solution with better comprehensive performance.

[0029] 3. This system addresses the challenge of multi-parameter synergistic optimization in the traditional field of biomimetic multilayer metal-ceramic films. The performance of multilayer metal-ceramic films is determined by numerous structural and process parameters, which exhibit complex nonlinear interactions. Simultaneous optimization of multiple parameters to achieve the optimal balance of overall performance is difficult. This system can easily handle the synergistic optimization of a dozen or even dozens of variables by defining parameters such as "maximizing (hardness × H)". 3 / E 2 The system uses composite objective functions such as strength, toughness, and wear resistance to automatically find the parameter combination that achieves the best balance between multiple key performance indicators, perfectly embodying the design concept of biomimetic materials.

[0030] 4. This system addresses the challenges of knowledge transfer and standardization in the traditional field of biomimetic multilayer metal-ceramic films. The research and development process heavily relies on the experience of individual experts, exhibiting subjectivity and non-reproducibility. Valuable experimental data fails to be systematically managed and utilized to form effective digital assets. Using an SQL database for management ensures the integrity, security, and traceability of all experimental data. The entire optimization process is driven by algorithms and code, eliminating accidental errors and subjective biases from human operation. This results in a highly standardized experimental process and excellent reproducibility. All data generated by the system is saved and used for iterative model training. As the number of runs increases, the internal proxy model becomes increasingly accurate and intelligent, transforming into a continuously self-improving "expert system" that converts valuable research and development experience into sustainably reusable digital assets.

[0031] 5. The system core uses an industrial computer as the control center, and the key hardware adopts a mature multi-target magnetron sputtering system and an automated nanoindenter. It also integrates a robotic arm to realize automatic sample transfer. This design ensures that the system can operate stably and reliably for a long time in scientific research laboratories and even in future industrial production environments.

[0032] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0033] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0034] Figure 1 This is a flowchart of a method for autonomous optimization of parameters of a metal-ceramic biomimetic multilayer film, provided in an embodiment of the present invention.

[0035] Figure 2 This is a system architecture diagram of a metal-ceramic biomimetic multilayer film parameter autonomous optimization system and method provided in an embodiment of the present invention. Detailed Implementation

[0036] The following is in conjunction with the appendix Figure 1-2The principles and features of the present invention are described below. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the invention will become clearer from the following description. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the invention.

[0037] It should be noted that when a component is described as "fixed to" another component, it can be directly on the other component or may have a component in between. When a component is considered "connected to" another component, it can be directly connected to the other component or may have a component in between. When a component is considered "set on" another component, it can be directly set on the other component or may have a component in between. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0039] like Figure 1-2 As shown, the present invention provides a system and method for autonomous optimization of parameters of biomimetic multilayer films in metal ceramics, including a data management module, a machine learning agent model module, an optimization algorithm module, an automated experimental platform, and a central control module. The central control module is communicatively connected to the data management module, the machine learning agent model module, the optimization algorithm module, and the automated experimental platform, and is used to control the data flow and instruction flow between the modules to drive the entire autonomous optimization loop iteratively.

[0040] Preferably, the data management module uses an SQL database. The data management module stores and manages historical and real-time generated datasets. The datasets include input parameters and corresponding performance indicators. Using a mature SQL database (such as MySQL or PostgreSQL) ensures the reliability, security, and consistency of data management, avoids data loss or chaos, and conforms to industry standards. The SQL database provides structured and relational data management capabilities, and can efficiently and stably store and query complex experimental data.

[0041] Preferably, the input parameters include at least one of the following: material system, metal layer thickness, ceramic layer thickness, modulation cycle, number of layers, deposition power, deposition temperature, and working gas pressure. The performance indicators include at least one of the following: hardness, toughness, wear resistance, and film-substrate adhesion. All the listed parameters are key performance indicators that affect the practical application of metal-ceramic multilayer films.

[0042] Preferably, the machine learning surrogate model module uses a Gaussian Process Regression (GPR) model as the surrogate model. This module is connected to the data management module and trains the machine learning model based on existing datasets to learn the complex nonlinear mapping relationship between input parameters and performance indicators. It is also capable of predicting performance for new parameter combinations. GPR is particularly suitable for handling small-sample, high-cost experimental data optimization problems. It not only provides predicted values ​​but also estimates of prediction uncertainty, which provides crucial information for subsequent Bayesian optimization and forms the basis for achieving a "balanced exploration and exploitation." Compared to models using ordinary neural networks, which can only provide point estimates, it offers better predictive certainty.

[0043] Preferably, the optimization algorithm module employs a Bayesian optimization algorithm to maximize (hardness × H) 3 / E 2 As the optimization objective, the optimization algorithm module is connected to the machine learning surrogate model module to set the optimization objective (such as maximizing the product of hardness and toughness). Based on the prediction results of the surrogate model, the optimization algorithm (such as Bayesian optimization and genetic algorithm) searches and recommends the next batch of parameter combinations that are expected to improve the target performance in the parameter space. This process balances the exploration of unknown parameter regions and the utilization of existing high-performance regions, and uses the uncertainty of the GPR model to intelligently select the next most valuable experimental point.

[0044] Preferably, the automated experimental platform includes an automated preparation unit and an automated characterization unit. The automated preparation unit is a multi-target magnetron sputtering system that can switch between Cu (metal) targets and TiN (ceramic) targets through program control and precisely control the deposition time of each layer to adjust the thickness, power, and gas pressure. The automated characterization unit is a nanoindenter with an integrated robotic arm. The robotic arm automatically transfers the sample from the preparation chamber to the testing stage for testing and feeds the results back to the central control module.

[0045] Preferably, the central control module is an industrial computer that runs control software and coordinates the work of each module, making it suitable for long-term continuous operation in laboratory or even future industrial environments.

[0046] The specific working principle and usage method of this invention are as follows:

[0047] S1. Initialize the dataset as the starting point of the optimization loop. Build a structured database in the data management module. The database fields should include at least the input parameters and their corresponding performance metrics.

[0048] Input parameters include the thickness of the metal layer (Cu) (nm), the thickness of the ceramic layer (TiN) (nm), the sputtering power of the Cu target (W), the sputtering power of the TiN target (W), the working pressure (Pa), and the deposition temperature (°C);

[0049] Output performance includes nanoindentation hardness (GPa), elastic modulus (GPa), and H is calculated. 3 / E 2 The H value (GPa) is used as a comprehensive index to evaluate toughness and resistance to plastic deformation. In this embodiment, the optimization objective is set to maximize H. 3 / E 2 value;

[0050] Through the human-computer interaction interface of the central control module, the initial "seed dataset" is imported. This dataset can be derived from 10-20 sets of historical experimental data or orthogonal experimental data designed based on physical vapor deposition experience, providing initial learning samples for the machine learning proxy model.

[0051] S2, Model Training and Update: The central control module instructs the machine learning proxy model module to retrieve the current complete dataset from the data management module. The machine learning proxy model module then calls its embedded Gaussian process regression model for training. The Gaussian process is fully defined by the mean function m(x) and the covariance function k(x,x'), and its formal expression is as follows:

[0052] f(x)~GP(m(x),k(x,x'));

[0053] Wherein, the input x is a vector representing a set of process parameters (e.g., [Cu thickness, TiN thickness, Cu power, ...]), and the output f(x) is the predicted performance index (H). 3 / E 2 );

[0054] In this embodiment, the covariance function k is selected from the Malton kernel, and its formula is as follows:

[0055] k(x_i,x_j)=(1+√3*d / σ)*exp(-√3*d / σ), where d=||x_i-x_j||_2;

[0056] This kernel function is suitable for handling nonlinear, smoothly changing physical processes. The model optimizes the hyperparameter σ of the kernel function by maximizing the marginal likelihood function. After training, the surrogate model has the ability to predict the mean μ(x_) and uncertainty variance σ of any input parameter x_.2 The ability of (x_*);

[0057] S3, Parameter Optimization and Recommendation: The central control module instructs the optimization algorithm module to start. The optimization algorithm module uses the Bayesian optimization algorithm and uses the expected boost as the acquisition function to guide the search. The optimization algorithm module obtains the predicted mean and variance of all points in the current parameter space from the surrogate model. The formalized algorithm for calculating the expected boost function is as follows:

[0058] The function calculates the expected improvement (current best performance f_best, candidate point x):

[0059] Obtain the mean μ and standard deviation σ of the performance predictions for candidate point x from the surrogate model;

[0060] The standardized improvement amount z = (μ - f_best - ξ) / σ / / ξ is a small constant (e.g., 0.01) for balancing exploration and utilization;

[0061] The increase PI is calculated as follows: PI = (μ - f_best - ξ) * Φ(z) + σ * φ(z) / / Φ is the standard normal cumulative distribution function, and φ is the standard normal probability density function;

[0062] Return PI;

[0063] End function;

[0064] Main loop: For the set of candidate points in the parameter space, call the `calculate expected improvement` function and select the candidate point x_next that maximizes the expected improvement value PI as the parameter combination for the next experiment;

[0065] End the main loop;

[0066] The algorithm module sends the recommended optimal candidate parameter combination x_next (e.g., Cu thickness = 8.5nm, TiN thickness = 16.2nm, Cu power = 150W, ...) to the central control module;

[0067] S4, fully automated experiment execution: The central control module receives the recommended parameter combination x_next and converts it into control commands, which are then sent to the automated experimental platform. The automated preparation unit of the automated experimental platform is an ultra-high vacuum multi-target magnetron sputtering device (e.g., the MULTI450 model ultra-high vacuum multi-target magnetron sputtering device) that integrates a programmable logic controller (PLC) and a robot transfer system.

[0068] Specific execution process:

[0069] 1. The central control module sends the process recipe (thickness of each layer, power, time, etc.) to the PLC of the sputtering equipment via Ethernet;

[0070] 2. The PLC automatically controls the vacuum system to evacuate to the base vacuum, fill with working gas (Ar, N2), and precisely adjust to the set pressure;

[0071] 3. The robot transfers the substrate to the Cu target. The PLC calculates the required deposition time based on the thickness and controls the RF power supply to turn on at the set power to deposit the Cu layer.

[0072] 4. After completion, the robot transfers the sample to the TiN target site, and deposits the TiN layer in the same manner;

[0073] 5. Repeat the cycle until the preset number of layers is reached, and the preparation is complete;

[0074] Subsequently, the automated characterization unit was activated, which includes a fully automated nanoindenter (e.g., NHT). 3 (Model number) Its sample stage is controlled by a robotic arm, which automatically transports the prepared sample from the sputtering chamber to the test stage of the nanoindenter. The nanoindenter automatically performs indentation tests at multiple points on the sample surface according to a preset program, collects load-displacement curves, and automatically calculates the hardness H and elastic modulus E using the Oliver-Pharr method, thereby calculating H. 3 / E 2 After the test is completed, the characterization unit will automatically upload the result data back to the central control module.

[0075] S5, Data Entry and Loop Termination Determination: The central control module writes the complete data pair of this experiment {input parameter: x_next, output parameter: y_next} into the database of the data management module, expanding the dataset. Subsequently, the central control module executes the termination condition judgment logic.

[0076] The judgment conditions include:

[0077] 1. Performance meets the standard: Does the latest obtained y_next exceed the preset threshold T (e.g., H)? 3 / E 2 >30GPa);

[0078] 2. Number of iterations: Whether the number of completed loops has reached the maximum set value N (e.g., N = 50);

[0079] 3. Convergence judgment: Whether the improvement in target performance in M ​​consecutive iterations (e.g., M=10) is less than the tolerance ε (e.g., ε=0.1%).

[0080] If any termination condition is met, the central control module terminates the entire optimization loop and outputs the historical best parameter combination and its performance data. At the same time, it can instruct the machine learning agent model module to export the final trained model for analysis.

[0081] If no termination condition is met, the process automatically jumps to step S2 to begin the next iteration. The new database will be used to train a more accurate surrogate model, thus starting a new and smarter optimization loop.

[0082] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Content not described in detail in this specification is prior art known to those skilled in the art.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A system for autonomous optimization of parameters of biomimetic multilayer metal-ceramic films, comprising a data management module, a machine learning proxy model module, an optimization algorithm module, an automated experimental platform, and a central control module, characterized in that, The central control module is connected to the data management module, the machine learning agent model module, the optimization algorithm module and the automated experimental platform respectively, and is used to control the data flow and instruction flow between the modules, driving the entire autonomous optimization loop to iterate. The optimization method includes the following steps: S1, Initialize the dataset: Enter the initial dataset in the data management module; S2, Model Training and Update: The machine learning agent model module trains or updates the agent model based on the current dataset, establishing a mapping relationship between input parameters and output performance. S3, Parameter optimization and recommendation: The optimization algorithm module calculates and recommends one or a set of optimal candidate parameter combinations in the parameter space based on the current agent model and the preset optimization objective. S4, fully automated experiment execution: The central control module sends the recommended candidate parameters to the automated experimental platform. The automated preparation unit prepares the sample according to the parameters, and then the automated characterization unit tests the sample to obtain performance data, which is then transmitted to the central control module. S5, Data entry and loop termination determination: Store the parameter-performance data pairs of the new experiment into the data management module to form an expanded dataset. The data is sent back to the central control module to determine whether the termination condition is met. If not, return to step S2 to proceed to the next iteration. If the condition is met, terminate the loop and output the optimal parameter combination and the corresponding performance.

2. The metal-ceramic biomimetic multilayer film parameter autonomous optimization system according to claim 1, characterized in that, The data management module uses an SQL database to store and manage historical and real-time generated datasets, which include input parameters and corresponding performance metrics.

3. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 2, characterized in that, The input parameters include at least one of the following: material system, metal layer thickness, ceramic layer thickness, modulation cycle, number of layers, deposition power, deposition temperature, and working gas pressure. The performance indicators include at least one of the following: hardness, toughness, wear resistance, and film-substrate adhesion.

4. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, The machine learning proxy model module uses a Gaussian process regression model as a proxy model. The machine learning proxy model module is connected to the data management module and trains the machine learning model based on the existing dataset to learn the complex nonlinear mapping relationship between input parameters and performance indicators and to predict the performance of new parameter combinations.

5. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, The optimization algorithm module employs a Bayesian optimization algorithm with maximization as the optimization objective. This module is connected to a machine learning surrogate model module to set the optimization objective and, based on the prediction results of the surrogate model, uses the optimization algorithm to search and recommend the next batch of parameter combinations that are expected to improve the target performance in the parameter space. This process balances the exploration of unknown parameter regions with the utilization of existing high-performance regions.

6. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, The automated experimental platform includes an automated preparation unit and an automated characterization unit. The automated preparation unit is a multi-target magnetron sputtering system that can switch between metal and ceramic targets through program control and precisely control the deposition time of each layer to adjust the thickness, power, and gas pressure. The automated characterization unit is a nanoindenter with an integrated robotic arm. The robotic arm automatically transfers the sample from the preparation chamber to the testing stage for testing and feeds the results back to the central control module.

7. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, The central control module is an industrial computer that runs control software and coordinates the work of each module.

8. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, In step S1, the initial dataset is derived from historical databases, literature data, or pre-designed initial experimental data.

9. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, In step S3, the optimization objective is to maximize / minimize a single performance index or to maximize a weighted comprehensive objective of multiple performance indexes. When recommending candidate parameter combinations, it is necessary to balance exploring the unknown parameter region and utilizing the existing high-performance region.

10. The self-optimization system for parameters of a metal-ceramic biomimetic multilayer film according to claim 1, characterized in that, In step S5, the termination condition is at least one of the following: the performance index reaches a predetermined threshold, the number of iterations reaches the upper limit, or the performance improvement is below the tolerance for several consecutive generations.

Images