Film coating process optimization method and film coating equipment for improving uniformity of film on inner wall of hemispherical harmonic oscillator

Abstract

A film coating process optimization method for improving uniformity of a film on the inner wall of a hemispherical resonator comprises the steps that a finite element simulation model in the film coating process of the hemispherical resonator is established, and multi-angle deposition parameters are obtained; discretizing the process into a time step sequence, and establishing an accumulated film thickness state space; a reinforcement learning framework based on near-end strategy optimization (PPO) is constructed, a state space is designed as the current accumulated film thickness and distribution, and an action space is an optimal angle selected from predefined deposition angles; generating an action strategy through a neural network, and performing iterative optimization to a target thickness; defining a reward function to minimize film thickness uniformity as a target; dynamically adjusting a deposition angle selection strategy; and an optimal strategy with the minimum film thickness uniformity value is screened out, and a corresponding deposition angle and a time distribution scheme thereof are output. The deposition angle and time distribution are dynamically optimized through reinforcement learning, the non-uniformity of the film thickness is remarkably reduced, and the method is suitable for film coating process optimization of harmonic oscillators of different sizes.

Description

Technical Field

[0001] This invention relates to the field of coating technology, and specifically to a coating process optimization method and coating equipment for improving the uniformity of the thin film on the inner wall of a hemispherical harmonic oscillator. Background Art

[0002] As a core component of high-precision inertial devices, the uniformity of the thin film on the inner wall of a hemispherical resonator directly affects the stability of its resonant frequency and its mechanical performance. Traditional coating processes often rely on trial and error or static parameter control, making it difficult to dynamically adapt to the complex nonlinear changes during the deposition process. This can easily lead to uneven film thickness distribution, which in turn causes performance degradation of the resonator. Especially in high-precision applications, even nanometer-level thickness deviations can significantly reduce the device's quality factor. Therefore, developing adaptive optimization algorithms to achieve uniform coating has significant engineering value.

[0003] Existing optimization methods often face problems such as policy adjustment lag and difficulty in resolving multi-parameter coupling, while reinforcement learning (RL) technology provides a new approach to solving such dynamic decision-making problems. The Proximal Policy Optimization (PPO) algorithm, as an efficient method in the RL field, ensures training stability by constraining the policy update magnitude. Its objective function can be expressed as:

[0004]

[0005] Where Et is the expectation, r t (θ) Strategy ratio, ∈ represents the estimated advantage function, and ∈ is the truncation hyperparameter. This mechanism maximizes the cumulative reward while ensuring convergence. The PPO algorithm dynamically generates the action probability distribution through a neural network and adaptively adjusts the deposition strategy in multiple iterations, overcoming the limitations of traditional methods that rely on fixed rules.

[0006] Currently, improving the performance of hemispherical resonator gyroscopes faces significant technical challenges, with the core constraint being that the fabrication process of curved metallized thin films has not yet reached an ideal level. This technical bottleneck directly leads to problems such as insufficient measurement accuracy and poor operational stability in gyroscopes. In particular, the precision thin film deposition technology for complex curved surfaces has become a key factor hindering the development of my country's inertial navigation field.

[0007] To address the strong demand for high-precision gyroscopes in high-end applications such as space exploration, it is urgent to overcome the key technical challenges of precision coating on curved surfaces. Systematic research into the thin film growth mechanism of curved substrates, the development of advanced coating processes, and the establishment of a comprehensive curved surface coating technology system will provide crucial technical support for my country's national defense modernization. Summary of the Invention

[0008] The purpose of this invention is to obtain data on the inner wall film thickness distribution under different angles between the resonator and the evaporation source during coating through finite element simulation. The obtained data is then imported into a reinforcement learning algorithm based on proximal policy optimization (PPO). The algorithm has policy optimization function, and can output a coating process with high uniformity of the inner wall of the resonator through the imported simulation data.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] An optimization algorithm for coating process to improve the uniformity of the thin film on the inner wall of a hemispherical harmonic oscillator is implemented by the following steps:

[0011] Step 1) Obtain film thickness distribution data at different deposition angles by establishing a finite element model;

[0012] Step 2) Discretize the coating process into discrete time steps, with each step size corresponding to the thickness increment detected by the crystal oscillator. Define the initial state as zero film thickness.

[0013] Step 3) Construct a reinforcement learning framework based on proximal policy optimization (PPO), designing the state space as the current accumulated film thickness and distribution, and the action space as the selection of the optimal angle from predefined deposition angles;

[0014] Step 4) Generate the action probability distribution through the neural network policy model, accumulate it for a single time step, terminate the calculation when the thickness reaches the target film thickness, and output the film thickness uniformity obtained in a single cycle.

[0015] Step 5) Define a reward function with the objective of minimizing film thickness uniformity;

[0016] Step 6) Iterate through multiple optimization loops to dynamically adjust the deposition angle selection strategy;

[0017] Step 7) Record the film thickness uniformity value for each iteration, select the optimal strategy with the smallest film thickness uniformity value, and output the corresponding deposition angle and its time allocation scheme.

[0018] In step 1), when the harmonic oscillator in the finite element model is rotating, the film thickness data of the inner wall is obtained when the angle between the harmonic oscillator and the evaporation source is 0°, 10°, 20°, 30°, 40°, 50°, 60°, 70°, and 80°.

[0019] In step 5), the reward function is:

[0020]

[0021] Among them, V max To achieve the target film thickness, the maximum value of the inner wall film thickness of the resonator is V. minThe minimum thickness of the inner wall of the harmonic oscillator when the target film thickness is achieved.

[0022] Compared with existing technologies, the present invention has the following advantages:

[0023] This invention reduces the number of experiments by combining finite element simulation data with a pre-trained model. Furthermore, it employs reinforcement learning (PPO algorithm) to achieve closed-loop feedback optimization through dynamic adjustment of deposition angle and duration, significantly improving coating uniformity. In addition, the neural network-based strategy model can be applied to resonators of different sizes, requiring only adjustments to the simulation input data. The algorithm framework also supports extension to other coating processes. Compared to traditional coating processes that rely heavily on fixed parameters or manual experience adjustments, requiring repeated experimentation and trial and error, resulting in high costs and long cycles, this invention solves the three major pain points of traditional coating processes: poor uniformity, low efficiency, and reliance on experience, providing a standardized intelligent solution for high-precision resonator manufacturing. Attached Figure Description

[0024] Figure 1 This is a flowchart of the algorithm of the present invention;

[0025] Figure 2 This is a diagram of a single-cycle recurrent neural network for the algorithm of this invention;

[0026] Figure 3 To input film thickness distribution diagrams for different coating angles;

[0027] Figure 4 This is a diagram showing the film thickness distribution after the coating process has been optimized using a reinforcement learning algorithm.

[0028] Figure 5 The convergence graph of the reinforcement learning algorithm; DETAILED DESCRIPTION

[0029] The invention will now be further described with reference to the accompanying drawings.

[0030] like Figures 1 to 5 As shown, an optimization algorithm for a coating process to improve the uniformity of the thin film on the inner wall of a hemispherical resonator is implemented through the following steps:

[0031] Step 1) Obtain film thickness distribution data at different deposition angles by establishing a finite element model;

[0032] Step 2) Discretize the coating process into discrete time steps, with each step size corresponding to the thickness increment detected by the crystal oscillator. Define the initial state as zero film thickness.

[0033] Step 3) Construct a reinforcement learning framework based on proximal policy optimization (PPO), designing the state space as the current accumulated film thickness and distribution, and the action space as the selection of the optimal angle from predefined deposition angles;

[0034] Step 4) Generate the action probability distribution through the neural network policy model, accumulate it for a single time step, terminate the calculation when the thickness reaches the target film thickness, and output the film thickness uniformity obtained in a single cycle.

[0035] Step 5) Define a reward function with the objective of minimizing film thickness uniformity;

[0036] Step 6) Iterate through multiple optimization loops to dynamically adjust the deposition angle selection strategy;

[0037] Step 7) Record the film thickness uniformity value for each iteration, select the optimal strategy with the smallest film thickness uniformity value, and output the corresponding deposition angle and its time allocation scheme.

[0038] In step 1), when the harmonic oscillator in the finite element model is rotating, the film thickness data of the inner wall is obtained when the angle between the harmonic oscillator and the evaporation source is 0°, 10°, 20°, 30°, 40°, 50°, 60°, 70°, and 80°.

[0039] In step 5), the reward function is:

[0040]

[0041] Among them, V max To achieve the target film thickness, the maximum value of the inner wall film thickness of the resonator is V. min The minimum thickness of the inner wall of the harmonic oscillator when the target film thickness is achieved.

[0042] The specific operation of the method of the present invention will be further explained below:

[0043] In step 1), the finite element simulation software, COMSOL's free molecular flow module, can be used to simulate the coating process. During the simulation, the resonator needs to rotate. When the simulated reading of the crystal oscillator is 1 nm, the film thickness distribution along the meridian direction on the inner wall is recorded, and simulations are performed at the angles between each resonator and the evaporation source. When the target film thickness is 100 nm, the coating process is discretized into 100 steps. In each step, a PPO-based reinforcement learning algorithm is used to select an optimal angle from the input angles for coating. A reward function is set, aiming to make the film thickness as uniform as possible; the smaller the difference between the thickest and thinnest points, the higher the reward. The reward function determines the next step of the algorithm, and the above steps are repeated until the film reaches the target thickness. The algorithm undergoes 1000 virtual coating experiments through repeated trial and error. After training, the algorithm outputs an optimal strategy, where the strategy is represented by the coating time corresponding to the angles between the resonator and the evaporation source: 0°, 10°, 20°, 30°, 40°, 50°, 60°, 70°, and 80°.

[0044] This invention pre-obtains film thickness distribution data at different deposition angles by establishing a finite element simulation model, and then combines this with reinforcement learning (PPO algorithm) for dynamic optimization, significantly reducing the number of experiments and improving optimization efficiency. It solves the problem that traditional methods cannot dynamically adapt to nonlinear changes in the coating process, achieving data-driven intelligent optimization.

[0045] A near-end strategy optimization (PPO) algorithm is employed to design a state space (current film thickness distribution) and an action space (deposition angle selection). A neural network strategy model is used to adjust deposition parameters in real time. Key breakthrough: Compared to traditional fixed-angle deposition, this method can dynamically adjust the deposition strategy to adapt to the optimization needs of different film thickness growth stages and improve uniformity.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing the coating process to improve the uniformity of the thin film on the inner wall of a hemispherical harmonic oscillator, characterized in that, The steps include: Step 1) Establish a finite element model of the coating process of the hemispherical harmonic oscillator and obtain the film thickness distribution data under different deposition angles; Step 2) Discretize the coating process of the hemispherical harmonic oscillator into multiple time steps, each step size corresponding to the thickness increment detected by the crystal oscillator, and define the initial state as the film thickness being zero; Step 3) Construct a reinforcement learning framework based on proximal policy optimization (PPO), with the state space being the current accumulated film thickness and distribution, and the action space being the selection of the optimal angle from predefined deposition angles; Step 4) Generate the action probability distribution through the neural network strategy model, accumulate the film thickness for a single time step, terminate the calculation when the target film thickness is reached, and output the film thickness uniformity obtained in a single cycle. Step 5) Define a reward function with the objective of minimizing film thickness uniformity; Step 6) Iterate through multiple optimization loops to dynamically adjust the deposition angle selection strategy; Step 7) Record the film thickness uniformity value for each iteration, select the optimal strategy with the smallest film thickness uniformity value, and output the corresponding deposition angle and its time allocation scheme.

2. The coating process optimization algorithm for improving the uniformity of the thin film on the inner wall of a hemispherical harmonic oscillator as described in claim 1, characterized in that: In step 1), the finite element model is established to obtain the film thickness data of the inner wall when the angle between the harmonic oscillator and the evaporation source is 0°, 10°, 20°, 30°, 40°, 50°, 60°, 70°, and 80° under the condition of the harmonic oscillator rotation.

3. The method for optimizing the coating process to improve the uniformity of the thin film on the inner wall of a hemispherical harmonic oscillator as described in claim 1, characterized in that: In step 5), the reward function is defined as: Among them, V max To achieve the target film thickness, the maximum value of the inner wall film thickness of the resonator is V. min The minimum thickness of the inner wall of the harmonic oscillator when the target film thickness is achieved.

4. A coating equipment, characterized in that, The coating process optimization method described in any one of claims 1-3 is used to dynamically control the angle between the evaporation source and the resonator and the deposition time to achieve highly uniform thin film deposition.

5. A hemispherical harmonic oscillator, characterized in that, The inner wall film is prepared by the coating process optimization method described in any one of claims 1-3.

Images