Evaporation process monitoring method and system based on data analysis and medium

Abstract

The invention relates to the technical field of evaporation process monitoring, and discloses an evaporation process monitoring method and system based on data analysis and a medium. The method comprises the following steps: acquiring data of a sensor in an evaporation cavity, and performing filtering standardization treatment to obtain process parameter data; performing time-frequency analysis and correlation calculation on the process parameter data, and extracting key parameter features; setting a dynamic threshold based on the key parameter characteristics, and calculating a deviation degree to obtain an abnormal index; and adjusting the radio frequency power and the substrate temperature in real time according to the abnormal index, and generating a control scheme. According to the method, the complex incidence relation among the multi-source sensing data can be analyzed in real time, the abnormal mode can be recognized in time, potential problems can be predicted before the abnormal condition is obviously expressed, the technological parameters are automatically adjusted through the intelligent control strategy, and therefore the technical defects that in the prior art, response of a monitoring system is lagged, and the technological parameters cannot be adjusted in real time are overcome; and the stability of the evaporation process and the product quality consistency are improved.

Description

Technical Field

[0001] The present application relates to the technical field of evaporation process monitoring, and in particular to a method, system and medium for monitoring an evaporation process based on data analysis. Background Art

[0002] Evaporation technology is an important method of modern thin film deposition and is widely used in the manufacturing processes of semiconductor devices, optical coatings, solar cells, etc. During the evaporation process, the raw materials are deposited onto the surface of the substrate by heating evaporation or sputtering in a vacuum environment, forming a thin film structure with specific functions. Traditional evaporation process monitoring mainly relies on empirical parameter settings and manual intervention, and is controlled by pre-set process parameters such as chamber temperature, pressure, and RF power. With the development of technology, automated monitoring systems have begun to be applied to evaporation production, using a fixed threshold method to monitor key parameters and triggering an alarm when the parameters exceed the preset range. Some advanced systems introduce simple data analysis methods, such as moving average and linear regression, to perform preliminary processing of process parameters to reduce false alarms and improve monitoring accuracy.

[0003] However, existing technologies have the following significant shortcomings: First, parameter collection is fragmented, making it difficult to establish a unified data standard. Data from each monitoring point is stored in isolation and lacks integrated analysis. Second, although the evaporation process generates a large amount of data, the analysis methods are overly simplistic, mostly remaining at the stage of comparing single parameter thresholds, unable to explore the deep correlations and complex patterns between parameters. Third, the monitoring system has a lag in response, making it impossible to adjust process parameters in real time. There is a significant delay in identifying and handling abnormal situations, resulting in unnecessary waste of materials and energy. Fourth, existing systems lack the ability to learn autonomously, with low recognition rates for new abnormal patterns and difficulty adapting to changing process conditions. Fifth, inter-system coordination is poor, and the monitoring, analysis, and control links are separated from each other, making it difficult to form a complete optimization closed loop. These problems are particularly prominent in scenarios with high-precision requirements, such as the difficulty in meeting the requirements for micro-nano-scale film thickness uniformity control and optical performance stability.

[0004] This application solves the core problem that traditional monitoring systems cannot achieve accurate anomaly detection and automatic control; in this application, a technical system is constructed from raw data processing to deep feature extraction, and then to anomaly detection and classification based on deep learning, which solves the technical problems of high data dimension, complex patterns and diverse anomaly types in the evaporation process. Summary of the Invention

[0005] The present application provides a method, system and medium for monitoring an evaporation process based on data analysis, which can analyze the complex correlation between multi-source sensor data in real time, identify abnormal patterns in a timely manner, and predict potential problems before abnormal conditions become obvious. It automatically adjusts process parameters through intelligent control strategies, thereby overcoming the technical defects of the existing technology such as the delayed response of the monitoring system and the inability to adjust the process parameters in real time, and improving the stability of the evaporation process and the consistency of product quality.

[0006] In the first aspect, the present application provides a method for monitoring an evaporation process based on data analysis, and the method comprises: collecting raw data from a temperature sensor, a pressure sensor, a power monitor, and a thickness monitoring system installed in an evaporation chamber, and obtaining evaporation process parameter data through filtering and standardization; performing time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extracting the variation characteristics of the evaporation source temperature, chamber pressure, and deposition rate, and obtaining key parameter characteristics of the evaporation process; based on the key parameter characteristics of the evaporation process, performing deviation analysis by setting a dynamic threshold, calculating the degree of deviation between the real-time value of the parameter and the standard value, and obtaining an abnormality index of the evaporation process; and adjusting the RF power input and substrate temperature control in real time according to the abnormality index of the evaporation process, generating a correction control instruction, and obtaining a target control scheme for thin film growth in the evaporation chamber.

[0007] In a first embodiment of the first aspect, collecting raw data from a temperature sensor, a pressure sensor, a power monitor, and a thickness monitoring system installed in the evaporation chamber, and obtaining evaporation process parameter data through filtering and normalization processing, includes: Temperature data is collected from temperature sensors installed at the evaporation source, substrate, and chamber wall. The original temperature distribution data is obtained through local regression model combined with sliding window analysis. The vacuum degree data is collected from the pressure sensors installed at different positions in the cavity, and the original pressure gradient data is obtained through multi-point average sampling and outlier elimination. The power input data recorded in real time by the power monitor is sampled at a fixed frequency and marked with mutation points to obtain the original power fluctuation data. Applying wavelet transform filtering technology to the temperature distribution raw data, pressure gradient raw data, and power fluctuation raw data to retain signal characteristics while removing high-frequency noise to obtain filtered parameter data; Performing zero-mean normalization processing on the filtered parameter data, converting parameters of different dimensions into a standard distribution format, and obtaining standardized parameter data; The sensor data of different sampling frequencies in the standardized parameter data are aligned by time stamps to establish a unified time base, and the first-order derivative and moving average features are extracted to obtain the evaporation process parameter data.

[0008] In a second embodiment of the first aspect, the time-frequency analysis and parameter correlation calculation are performed on the evaporation process parameter data to extract the variation characteristics of the evaporation source temperature, chamber pressure, and deposition rate to obtain key parameter characteristics of the evaporation process, including: The evaporation process parameter data is divided into fixed windows, and the statistical moment characteristics and trend characteristics of the evaporation source temperature, chamber pressure and deposition rate in each window are calculated to obtain the parameter time domain characteristics; Performing a fast Fourier transform on the evaporation process parameter data, extracting a power spectral density curve, identifying periodic fluctuation components of the evaporation source temperature, chamber pressure, and deposition rate, and obtaining parameter frequency domain characteristics; Applying wavelet transform or Hilbert transform to the evaporation process parameter data to capture the time-varying frequency characteristics of the evaporation source temperature, chamber pressure and deposition rate to obtain parameter time-frequency characteristics; Calculating sample entropy and multi-scale entropy for the parameter time domain features, parameter frequency domain features, and parameter time-frequency features, quantifying the complexity of parameter fluctuations, and obtaining parameter complexity features; Based on the parameter time domain characteristics, parameter frequency domain characteristics, parameter time-frequency characteristics and parameter complexity characteristics, the Pearson correlation coefficient and mutual information value between the evaporation source temperature, the chamber pressure and the deposition rate are calculated to obtain a parameter correlation matrix; Partial correlation analysis and conditional independence test are applied to the parameter correlation matrix to eliminate indirect correlation effects, determine the causal relationship and time delay between parameters, and obtain the key parameter characteristics of the evaporation process.

[0009] In a third embodiment of the first aspect, the deviation analysis is performed by setting a dynamic threshold based on the key parameter characteristics of the evaporation process, and the degree of deviation between the real-time value of the parameter and the standard value is calculated to obtain the abnormality index of the evaporation process, including: Performing cluster analysis on key parameter characteristics of the evaporation process, dividing process operation modes using a Gaussian mixture model, and obtaining a standard mode library for the evaporation process; For each process mode in the standard pattern library of the evaporation process, characteristic multidimensional distribution parameters are calculated, including a mean vector and a covariance matrix, to obtain a statistical representation of the mode; For the key parameter characteristics of the evaporation process collected in real time, the Mahalanobis distance and Hotelling statistics of the pattern statistical representation are calculated to obtain the parameter deviation; Based on the parameter deviation, a deep autoencoder network is constructed to encode and decode the input features, calculate the reconstruction error, and obtain the abnormal reconstruction loss; For the abnormal reconstruction loss, an adaptive threshold is set according to the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations, and obtain an abnormal level classification; Based on the abnormality level classification, the abnormality start time, duration, severity score and related parameter contribution are calculated to obtain the abnormality index of the evaporation process.

[0010] In a fourth implementation of the first aspect, constructing a deep autoencoder network based on the parameter deviation, encoding and decoding the input features, calculating the reconstruction error, and obtaining the abnormal reconstruction loss include: Performing feature concatenation on the parameter deviation and key parameter features of the evaporation process to form an input vector, which is fed into the input layer of the deep autoencoder network to obtain features to be processed; Compress the features to be processed through a four-layer convolutional encoder, where the convolution kernel size of each layer is 3×3, the number of convolutional layer channels is 32, 64, 128, and 256, respectively, and activation function and pooling operation are applied after each convolution layer to obtain compressed features; The compressed features are further reduced in dimension through a fully connected layer, with the number of nodes being 512, 256, 128, and 64, respectively, to obtain a latent space feature representation; The latent space feature representation is restored through a four-layer transposed convolutional decoder. The transposed convolutional decoder adopts a structure symmetrical to the encoder, and the number of channels is 256, 128, 64, and 32, respectively, to obtain the reconstructed features; Calculating the Euclidean distance and cosine similarity between the reconstructed features and the original input features, constructing a weighted loss function, and obtaining a feature reconstruction loss; The feature reconstruction loss is compared and analyzed with the reconstruction loss of continuous time series samples to identify abnormal mutation points and gradual trends, thereby obtaining the abnormal reconstruction loss.

[0011] In a fifth implementation of the first aspect, the abnormal reconstruction loss is adaptively thresholded according to the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations to obtain abnormal level classification, including: Perform probability distribution fitting on the historically accumulated abnormal reconstruction loss data, use the kernel density estimation method to construct a non-parametric probability density function, and obtain the loss distribution model; Based on the loss distribution model, the reconstruction loss values ​​of multiple quantiles are calculated and set as judgment criteria of different levels to obtain a static threshold set; Applying piecewise linear transformation to the static threshold set in combination with parameter distributions of the start-up phase, the stable phase, and the end phase of the evaporation process to obtain a stage-adaptive threshold; For the adaptive threshold of the stage, a sliding window average and exponential smoothing method is applied to eliminate the influence of short-term fluctuations to obtain a smooth adaptive threshold; Comparing the real-time anomaly reconstruction loss with the smooth adaptive threshold, and grading the loss according to the degree of exceeding the quantile thresholds of different levels to obtain an initial anomaly level; The initial abnormality level is subjected to time series processing, and the abnormality level is updated according to the abnormality duration and the accumulated abnormality value to obtain the abnormality level classification.

[0012] In a sixth embodiment of the first aspect, the step of adjusting the RF power input and substrate temperature control in real time based on the abnormality indicator of the evaporation process, generating a correction control instruction, and obtaining a target control scheme for thin film growth in the evaporation chamber includes: Integrating the abnormal indicators of the evaporation process with the historical operation data of the evaporation process to construct a digital twin model of the evaporation process, simulate the effects of different control parameters on thin film growth, and obtain a control parameter influence map; Based on the control parameter influence mapping, a policy gradient algorithm is used to train a control strategy network, taking the abnormality index of the evaporation process as input, outputting the adjustment amount of the RF power and substrate temperature, and obtaining a parameter control strategy; Performing physical constraint checks on the parameter control strategy to limit the range and rate of change of RF power and substrate temperature to prevent system instability caused by sudden adjustments, thereby obtaining a safe control strategy; Based on the safety control strategy, large-scale parameter adjustments are decomposed into a sequence of multiple small steps to form a smooth transition control trajectory and obtain a progressive control sequence; Converting the progressive control sequence into a command signal in a standard industrial control interface format, and writing the command signal into a programmable logic controller of the evaporation device via a communication interface to obtain an execution control command; The execution result of the execution control instruction is monitored in real time, and the abnormal indicator changes before and after the parameter adjustment are recorded to form control effect evaluation data, and obtain the target control plan for the thin film growth in the evaporation chamber.

[0013] In a second aspect, the present application provides a data analysis-based evaporation process monitoring system, the data analysis-based evaporation process monitoring system comprising: The acquisition module is used to collect raw data from the temperature sensor, pressure sensor, power monitor and thickness monitoring system installed in the evaporation chamber, and obtain the evaporation process parameter data through filtering and standardization; A calculation module is used to perform time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extract the variation characteristics of the evaporation source temperature, chamber pressure and deposition rate, and obtain the key parameter characteristics of the evaporation process; A setting module is used to perform deviation analysis based on the key parameter characteristics of the evaporation process by setting dynamic thresholds, calculate the degree of deviation between the real-time value of the parameter and the standard value, and obtain an abnormality indicator of the evaporation process; The adjustment module is used to adjust the radio frequency power input and substrate temperature control in real time according to the abnormal indicators of the evaporation process, generate correction control instructions, and obtain a target control plan for thin film growth in the evaporation chamber.

[0014] In a third aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned evaporation process monitoring method based on data analysis.

[0015] In the technical solution provided in the present application, by filtering and standardizing the raw data collected by various sensors in the evaporation chamber, the problems of inconsistent data formats and serious noise interference in traditional monitoring are effectively solved, and the data quality and availability are improved; time-frequency analysis and parameter correlation calculation are performed on the evaporation process parameter data, not only extracting parameter characteristics from multiple dimensions such as time domain, frequency domain and time-frequency domain, but also analyzing the correlation and causal relationship between parameters, breaking through the limitation of traditional methods that only focus on a single parameter and ignore the complex relationship between parameters, enabling the system to identify complex process state changes; setting dynamic thresholds based on the key parameter characteristics of the evaporation process for deviation analysis overcomes the defect that fixed thresholds are not suitable for different process stages, realizes the adaptability of anomaly detection, and greatly reduces the false alarm rate and missed alarm rate; the RF power input and substrate temperature control are adjusted in real time according to the abnormal indicators of the evaporation process to form a closed-loop control system, which solves the problem of the lack of automatic control mechanism after the traditional monitoring system finds an anomaly, and significantly shortens the response time of the abnormal state. In the specific application field of evaporation processes, the deep learning and reinforcement learning algorithms adopted by this invention contribute intelligent perception and autonomous decision-making capabilities to the overall solution: the deep autoencoder network can accurately capture subtle abnormal patterns by learning low-dimensional representations of normal process parameters, and is extremely sensitive to subtle fluctuations in deposition rate and film thickness uniformity; the policy gradient algorithm autonomously learns the optimal control strategy based on historical optimization experience, and automatically adjusts the RF power and substrate temperature parameters according to abnormal indicators, achieving human-like intelligent control capabilities; the kernel density estimation and piecewise linear transformation algorithms in the dynamic threshold calculation adapt to the changing characteristics of the parameter distribution at each stage of the evaporation process, enabling the system to accurately distinguish normal fluctuations from abnormal deviations; the wavelet transform in the time-frequency analysis captures the time-varying frequency characteristics of the evaporation parameters, revealing dynamic patterns that are difficult to detect with traditional spectrum analysis. These algorithmic features work together to significantly improve the stability of the evaporation process and the consistency of product quality, while reducing the waste of materials and energy caused by abnormal conditions and lowering production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 Schematic diagram of an embodiment of a method for monitoring an evaporation process based on data analysis in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a evaporation process monitoring system based on data analysis in an embodiment of the present application. DETAILED DESCRIPTION

[0018] The embodiments of the present application provide a method, system and medium for monitoring an evaporation process based on data analysis. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the evaporation process monitoring method based on data analysis includes: Step S101: collecting raw data from the temperature sensor, pressure sensor, power monitor, and thickness monitoring system installed in the evaporation chamber, and obtaining evaporation process parameter data through filtering and standardization; Step S102: performing time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extracting the variation characteristics of the evaporation source temperature, chamber pressure, and deposition rate, and obtaining key parameter characteristics of the evaporation process; Step S103: Based on the key parameter characteristics of the evaporation process, a deviation analysis is performed by setting a dynamic threshold value, and the deviation degree between the real-time value of the parameter and the standard value is calculated to obtain an abnormality index of the evaporation process; Step S104 : According to the abnormal indicators of the evaporation process, the RF power input and the substrate temperature control are adjusted in real time, and a correction control instruction is generated to obtain a target control scheme for the thin film growth in the evaporation chamber.

[0020] It is understandable that the execution subject of this application can be a evaporation process monitoring system based on data analysis, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0021] Specifically, the process of collecting raw data from the temperature sensors, pressure sensors, power monitors, and thickness monitoring system installed within the evaporation chamber and filtering and normalizing it to obtain evaporation process parameter data is as follows: thermocouple temperature sensors are installed at the evaporation source, substrate, and chamber wall of the evaporation equipment; pressure sensors are installed at various locations within the chamber; and a power monitor and a real-time film thickness monitoring system based on ellipsometers are configured. The temperature data collected by the temperature sensors is processed using a local regression model combined with sliding window analysis. The sliding window length is 30 seconds and the window overlap is 50%. The local regression model uses a quadratic polynomial fit to identify and mark sudden changes. The vacuum data from the pressure sensor is processed using multi-point averaging and outlier rejection. The multi-point averaging takes the average of ten sampling points. Outlier rejection uses the median absolute deviation method, marking data points that deviate more than three times the median as outliers and removing them. The power input data from the power monitor is processed using fixed-frequency sampling at 100 Hz and sudden change point marking. A cumulative sum control chart is used to detect sudden power changes. The raw data is processed using wavelet transform filtering technology. Using the Daubechies wavelet basis, the wavelet transform uses a three-layer decomposition process to preserve signal characteristics while removing high-frequency noise. The filtered parameter data undergoes zero-mean normalization using the formula (x-μ) / σ, where x is the raw data, μ is the mean, and σ is the standard deviation. This converts parameters of varying dimensions into a standardized distribution format. Finally, sensor data with varying sampling frequencies are aligned using timestamps to establish a unified time base. First-order derivative and moving average features are then extracted to obtain the evaporation process parameter data.

[0022] Time-frequency analysis and parameter correlation calculations were performed on the evaporation process parameter data to extract the variation characteristics of the evaporation source temperature, chamber pressure, and deposition rate, thereby obtaining the key parameter characteristics of the evaporation process. The detailed process is as follows: The evaporation process parameter data was segmented into fixed windows with a window length of 30 seconds. The statistical moment characteristics (mean, standard deviation, skewness, kurtosis) and trend characteristics (monotonicity, periodicity, and inflection point density) of the evaporation source temperature, chamber pressure, and deposition rate within each window were calculated to obtain the parameter time-domain characteristics. The evaporation process parameter data was then subjected to a fast Fourier transform (FFT) and the power spectral density curve was analyzed to identify the periodic fluctuation components of the evaporation source temperature, chamber pressure, and deposition rate, thereby obtaining the parameter frequency-domain characteristics. A wavelet transform or Hilbert transform was then applied to capture the time-varying frequency characteristics, obtaining the parameter time-frequency characteristics. Sample entropy and multiscale entropy were calculated for these characteristics to quantify the complexity of the parameter fluctuations. The sample entropy was calculated using an embedding dimension of 2 and a tolerance of 0.2 standard deviations. The multiscale entropy was calculated using a scaling factor ranging from 1 to 20 to obtain the parameter complexity characteristics. Based on these characteristics, the Pearson correlation coefficient and mutual information value between the evaporation source temperature, chamber pressure, and deposition rate were calculated to generate a parameter correlation matrix. Partial correlation analysis and conditional independence tests were applied to the parameter correlation matrix to eliminate indirect correlation effects, determine the causal relationship and time delay between parameters, and ultimately obtain the key parameter characteristics of the evaporation process.

[0023] Based on the key parameter characteristics of the evaporation process, deviation analysis is performed by setting dynamic thresholds. The degree of deviation between the real-time parameter values ​​and the standard values ​​is calculated to obtain anomaly indicators for the evaporation process. The specific process is as follows: First, cluster analysis is performed on the key parameter characteristics of the evaporation process. The process operation modes are divided using a Gaussian mixture model. The number of clusters is automatically determined using the Bayesian Information Criterion to obtain a standard pattern library for the evaporation process. For each process mode in the standard pattern library, the characteristic multidimensional distribution parameters, including the mean vector and covariance matrix, are calculated to obtain a pattern statistical representation. The Mahalanobis distance and Hotelling statistic are calculated for the real-time key parameter characteristics of the evaporation process with the pattern statistical representation to obtain the parameter deviation. Based on the parameter deviation, a deep autoencoder network is constructed to encode and decode the input features, calculate the reconstruction error, and obtain the anomaly reconstruction loss. The deep autoencoder network architecture consists of a four-layer convolutional encoder for compression. Each layer has a 3×3 convolution kernel size and the number of channels is 32, 64, 128, and 256, respectively. LeakyReLU activation function and maximum pooling are applied after each convolution layer to obtain compressed features. The compressed features are further reduced in dimension through a fully connected layer to obtain a latent space feature representation. The latent space features are restored through a four-layer transposed convolutional decoder to obtain reconstructed features. The Euclidean distance and cosine similarity between the reconstructed features and the original input features are calculated to construct a weighted loss function to obtain the feature reconstruction loss. Anomalous mutation points and gradient trends are identified through comparative analysis to obtain the abnormal reconstruction loss.

[0024] Adaptive thresholds are set for abnormal reconstruction loss based on the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations, resulting in an anomaly classification. The specific process involves fitting the probability distribution of historically accumulated abnormal reconstruction loss data, constructing a nonparametric probability density function using kernel density estimation, and calculating reconstruction loss values ​​at multiple quantiles as the criteria for determining different levels, resulting in a set of static thresholds. Based on the parameter distribution characteristics of different stages of the evaporation process, a piecewise linear transformation is applied to the stage-by-stage adaptive thresholds. Sliding window averaging and exponential smoothing are applied to the stage-by-stage adaptive thresholds to eliminate the influence of short-term fluctuations, resulting in a smoothed adaptive threshold. The real-time abnormal reconstruction loss is compared with the smoothed adaptive thresholds and classified according to the degree to which it exceeds the quantile thresholds at different levels. The anomaly level is updated based on the duration of the anomaly and the cumulative anomaly value, ultimately resulting in an abnormality indicator for the evaporation process. Based on the abnormality indicator for the evaporation process, real-time adjustments are made to the RF power input and substrate temperature control, generating corrective control instructions and deriving a target control solution for thin film growth within the evaporation chamber. The detailed process is as follows: the abnormality indicator for the evaporation process is integrated with historical evaporation process data to construct a digital twin model of the evaporation process. The effects of different control parameters on thin film growth are simulated, resulting in a control parameter impact map. Based on the control parameter impact mapping, a policy gradient algorithm is used to train the control strategy network. This network takes evaporation process anomaly indicators as input and outputs adjustments for RF power and substrate temperature, resulting in a parameter control strategy. Physical constraints are then applied to the control strategy to limit the range and rate of change of RF power and substrate temperature, preventing system instability caused by abrupt adjustments. This results in a safe control strategy.

[0025] Based on a safe control strategy, large parameter adjustments are broken down into a sequence of smaller steps, forming a smooth control trajectory and resulting in a progressive control sequence. This progressive control sequence is converted into a command signal in a standard industrial control interface format and written to the evaporation equipment's programmable logic controller via a communication interface to execute the control command. The execution results of the control command are monitored in real time, and abnormal changes in indicators before and after parameter adjustments are recorded to generate control effect evaluation data, thereby determining a targeted control solution for thin film growth within the evaporation chamber.

[0026] In a specific embodiment, the process of executing step S101 may specifically include the following steps: Temperature data is collected from temperature sensors installed at the evaporation source, substrate, and chamber wall. The original temperature distribution data is obtained through local regression model combined with sliding window analysis. The vacuum degree data is collected from the pressure sensors installed at different positions in the cavity, and the original pressure gradient data is obtained through multi-point average sampling and outlier elimination. The power input data recorded in real time by the power monitor is sampled at a fixed frequency and marked with mutation points to obtain the original power fluctuation data. Applying wavelet transform filtering technology to the temperature distribution raw data, pressure gradient raw data, and power fluctuation raw data to retain signal characteristics while removing high-frequency noise to obtain filtered parameter data; Performing zero-mean normalization processing on the filtered parameter data, converting parameters of different dimensions into a standard distribution format, and obtaining standardized parameter data; The sensor data of different sampling frequencies in the standardized parameter data are aligned by time stamps to establish a unified time base, and the first-order derivative and moving average features are extracted to obtain the evaporation process parameter data.

[0027] Specifically, temperature data is collected from temperature sensors installed at the evaporation source, substrate, and chamber walls. Raw temperature distribution data is obtained using a local regression model combined with sliding window analysis. The temperature sensors include K-type thermocouples with a measurement range of 20°C to 1200°C and a sampling frequency of 10 Hz. The local regression model is a non-parametric regression method that generates a smooth curve by weighted fitting of the local region around each data point. In the implementation, the temperature data is processed using LOESS (Local Weighted Scatter Point Smoothing), with a sliding window length of 30 seconds and a window overlap of 50%. The LOESS algorithm first selects a quadratic polynomial fit function within each window. Then, a cubic kernel function is applied to the data points within the window, assigning weights that decrease with increasing distance from the center point. This smoothes the local data and identifies sudden changes. Vacuum data is collected from pressure sensors installed at different locations within the chamber. Raw pressure gradient data is obtained through multi-point averaging and outlier removal. The pressure sensors are capacitance diaphragm vacuum gauges with a sampling frequency of 5 Hz. Multi-point averaging takes the arithmetic mean of multiple consecutive sampling points as a single data point, reducing the impact of random fluctuations. In actual processing, every 10 original sampling points are combined into one data point. Outlier removal uses the median absolute deviation (MAD) method. The calculation process is as follows: first, the median of all data points is calculated, then the absolute deviation of each data point from the median is calculated, and finally the median of these deviations (i.e., the MAD value) is calculated. If a data point deviates from the median by more than three times the MAD value, it is marked as an outlier and removed, and replaced with the average of the preceding and following data points.

[0028] The power input data recorded in real time by the power monitor is sampled at a fixed frequency and marked as sudden changes to obtain raw power fluctuation data. The power monitor records RF power input values ​​with a range of 0-5kW and an original sampling frequency of 1kHz. Fixed-frequency sampling resamples the raw data collected at a fixed time interval, downsampling to 100Hz in actual processing. A sudden change point is marked using the CUSUM (Cumulative Sum) control chart method. This calculates the cumulative deviation of the data sequence and marks a sudden change point when the cumulative deviation exceeds a preset threshold. The CUSUM algorithm consists of two parts: rise detection and fall detection. They calculate the functions S+(t)=max[0, S+(t-1)+(x(t)-μ0-k)] and S-(t)=max[0, S-(t-1)-(x(t)-μ0+k)], respectively. μ0 is the baseline mean, and k is the sensitivity parameter, set to half the standard deviation. When S+ or S- exceeds the threshold h (set to 5 times the standard deviation), a sudden change point is marked.

[0029] Wavelet transform filtering was applied to the raw data of temperature distribution, pressure gradient, and power fluctuation, preserving signal characteristics while removing high-frequency noise, resulting in filtered parameter data. Wavelet transform is a time-frequency analysis method suitable for processing non-stationary signals, providing both time and frequency domain information. The Daubechies wavelet basis (db4) was used in the actual processing, with a decomposition layer of three. Wavelet decomposition decomposes the signal into approximate coefficients and detail coefficients. The detail coefficients are thresholded (retaining large coefficients and setting small coefficients to zero) to remove noise, and then the signal is reconstructed. The threshold selection method uses a universal thresholding method. This method effectively removes noise caused by high-frequency electromagnetic interference and equipment vibration during the evaporation process, while preserving the true characteristics of temperature, pressure, and power variations. The filtered parameter data was then subjected to zero-mean normalization, converting the parameters of different dimensions into a standard distribution format to obtain standardized parameter data. Zero-mean normalization, also known as Z-score normalization, is calculated by subtracting the mean from the raw data and dividing it by the standard deviation, resulting in a mean of 0 and a standard deviation of 1 for the processed data. In actual processing, the mean and standard deviation are calculated for each parameter series separately, and then a normalization formula is applied. This step normalizes parameters with different dimensions and numerical ranges, such as temperature (unit: °C), pressure (unit: Pa), and power (unit: W), to the same scale, facilitating subsequent unified processing and comparative analysis.

[0030] By aligning sensor data with different sampling frequencies within the standardized parameter data through timestamps, a unified time base is established, and first-order derivative and moving average features are extracted to obtain the evaporation process parameter data. Because different sensors have different sampling frequencies (temperature 10Hz, pressure 5Hz, power 100Hz), a unified time base is required for data alignment. In actual processing, the lowest sampling frequency (5Hz) is selected as the benchmark, and other sensor data are adjusted to the same time point through linear interpolation or downsampling. The first-order derivative feature is obtained by calculating the difference between adjacent time points and reflects the rate of parameter change. The moving average feature uses a 10-point sliding window to calculate the mean and reflects the smooth trend of the parameter. These time series features enrich the data dimension and provide a foundation for subsequent pattern recognition and anomaly detection.

[0031] In a specific embodiment, the process of executing step S102 may specifically include the following steps: The evaporation process parameter data is divided into fixed windows, and the statistical moment characteristics and trend characteristics of the evaporation source temperature, chamber pressure and deposition rate in each window are calculated to obtain the parameter time domain characteristics; Performing a fast Fourier transform on the evaporation process parameter data, extracting a power spectral density curve, identifying periodic fluctuation components of the evaporation source temperature, chamber pressure, and deposition rate, and obtaining parameter frequency domain characteristics; Applying wavelet transform or Hilbert transform to the evaporation process parameter data to capture the time-varying frequency characteristics of the evaporation source temperature, chamber pressure and deposition rate to obtain parameter time-frequency characteristics; Calculating sample entropy and multi-scale entropy for the parameter time domain features, parameter frequency domain features, and parameter time-frequency features, quantifying the complexity of parameter fluctuations, and obtaining parameter complexity features; Based on the parameter time domain characteristics, parameter frequency domain characteristics, parameter time-frequency characteristics and parameter complexity characteristics, the Pearson correlation coefficient and mutual information value between the evaporation source temperature, the chamber pressure and the deposition rate are calculated to obtain a parameter correlation matrix; Partial correlation analysis and conditional independence test are applied to the parameter correlation matrix to eliminate indirect correlation effects, determine the causal relationship and time delay between parameters, and obtain the key parameter characteristics of the evaporation process.

[0032] Specifically, the evaporation process parameter data is segmented using fixed windows. The statistical moment and trend characteristics of the evaporation source temperature, chamber pressure, and deposition rate within each window are calculated to obtain the time-domain characteristics of the parameters. Fixed window segmentation divides the time series data into multiple continuous data segments of fixed length. Each window length is set to 30 seconds, and adjacent windows overlap by 50% to ensure that the continuous process is captured. Within each window, four statistical moment characteristics are calculated for the evaporation source temperature, chamber pressure, and deposition rate: mean (average level of the data), standard deviation (data dispersion), skewness (asymmetry of the data distribution), and kurtosis (peaked or flattened distribution). Trend characteristics are also calculated, including monotonicity (using the Mann-Kendall trend test to determine whether the data has an upward or downward trend), periodicity (detecting repetitive patterns in the signal through the autocorrelation function), and inflection point density (calculating the number of points where the first-order derivative of the curve changes sign, reflecting the frequency of fluctuations). A fast Fourier transform (FFT) is performed on the evaporation process parameter data to extract the power spectral density curve, identify the periodic fluctuation components of the evaporation source temperature, chamber pressure, and deposition rate, and obtain the frequency domain characteristics of the parameters. The fast Fourier transform (FFT) is an effective algorithm for converting time-domain signals to the frequency domain, revealing the implicit periodic components in the signal. The processing process first preprocesses the data, including detrending (subtracting the linear fitting curve) and windowing (using a Hanning window to reduce spectral leakage). The preprocessed data is then converted to the frequency domain using the FFT algorithm, and the power spectral density is calculated to obtain the energy distribution of different frequency components. Features such as the main frequency (the frequency corresponding to the energy peak), frequency bandwidth (full width at half maximum), and energy distribution characteristics (the proportion of energy in different frequency intervals) are extracted from the power spectral density curve to identify the periodic components and frequency distribution characteristics in the data.

[0033] Wavelet transform or Hilbert transform is applied to the evaporation process parameter data to capture the time-varying frequency characteristics of the evaporation source temperature, chamber pressure, and deposition rate, and obtain the parameter time-frequency characteristics. Wavelet transform is a time-frequency analysis tool suitable for analyzing non-stationary signals and can provide both time and frequency domain information. In actual processing, continuous wavelet transform is used, and the Morlet wavelet is selected as the wavelet basis function, which has good time-frequency positioning capabilities. By calculating the wavelet coefficients at different scales, a wavelet spectrum is formed to display the frequency distribution of the signal at different time points. Time-varying frequency characteristics are extracted from the wavelet spectrum, including instantaneous frequency (the dominant frequency that changes with time), energy concentration area, frequency change slope, and other features, to describe the dynamic frequency characteristics of the parameters. The Hilbert transform can also be used for similar purposes to extract instantaneous frequency and instantaneous amplitude information by calculating the analytical signal of the signal.

[0034] Sample entropy and multiscale entropy are calculated for the time-domain, frequency-domain, and time-frequency characteristics of the parameters to quantify the complexity of parameter fluctuations and obtain parameter complexity characteristics. Sample entropy is a measure of the complexity or irregularity of a time series; higher values ​​indicate a more complex or random sequence. The calculation process involves setting the embedding dimension m (usually 2 or 3) and the similarity tolerance r (usually 0.2 times the standard deviation), counting the number of patterns in the sequence that meet similarity conditions, and then calculating the negative logarithm of the conditional probability. Multiscale entropy reflects the complexity of the signal at different time scales by coarsening the original sequence at different scales and then calculating the sample entropy at each scale. By calculating the sample entropy and multiscale entropy for each parameter feature (usually with a scaling factor of 1 to 20), a quantitative indicator of the complexity of parameter fluctuations is obtained, which helps identify the complexity and regularity of system state changes.

[0035] Based on the parameter time-domain characteristics, frequency-domain characteristics, time-frequency characteristics, and complexity characteristics, the Pearson correlation coefficient and mutual information value between the evaporation source temperature, chamber pressure, and deposition rate were calculated to obtain a parameter correlation matrix. The Pearson correlation coefficient measures the degree of linear correlation between two variables and ranges from [-1 to 1]. Positive values ​​indicate positive correlation, while negative values ​​indicate negative correlation. Larger absolute values ​​indicate stronger correlation. The calculation process is to calculate the covariance of each pair of parameter features divided by the product of their respective standard deviations. The mutual information value measures the nonlinear dependency between two variables by calculating the expected logarithm of the ratio of the joint probability distribution of the two variables to the marginal probability distribution. A larger mutual information value indicates that the two variables contain more common information. By calculating the Pearson correlation coefficient and mutual information value between all pairs of parameter features, a parameter correlation matrix is ​​constructed to display the strength of the linear and nonlinear correlations between the parameters.

[0036] Partial correlation analysis and conditional independence tests are applied to the parameter correlation matrix to eliminate indirect correlations, identify causal relationships and time delays between parameters, and characterize key parameters of the evaporation process. Partial correlation analysis calculates the direct correlation between two variables by controlling for the influence of other variables, eliminating indirect correlations caused by intermediate variables. The calculation process is based on conditional probability theory and calculates the correlation coefficient between two variables after controlling for other variables. Conditional independence tests use statistical hypothesis testing methods to determine whether two variables are independent given a third variable. Common methods include the G² test based on conditional mutual information or the Fisher's Z-transform test. Furthermore, to identify time delay relationships between parameters, cross-correlation functions and Granger causality tests are applied. The cross-correlation function calculates the correlation between two time series at different time lags, identifying the time lag corresponding to the maximum correlation, which represents the time delay between the parameters. Granger causality tests construct autoregressive models to determine whether one time series helps predict the future values ​​of another, thereby determining the direction of causality between the variables.

[0037] Through this analysis, key parameter characteristics of the evaporation process are derived, including important eigenvalues ​​of each parameter, direct correlations between parameters, causal directions, and time delay information, providing a foundation for subsequent anomaly detection and control. Taking the evaporation process of a high-precision optical filter film as an example, the evaporation process parameter data from a single production run were analyzed. The data was first segmented into 30-second windows. The calculated mean of the evaporation source temperature was 1080°C, with a standard deviation of 8.5°C, a skewness of 0.12 (slightly right-skewed), and a kurtosis of 2.8 (close to a normal distribution). The mean of the chamber pressure was 2.3×10⁻¹⁴Pa, with a standard deviation of 3.7×10⁻¹⁵Pa. The mean of the deposition rate was 0.5nm / s, with a standard deviation of 0.08nm / s. FFT analysis revealed a main fluctuation component in the temperature data with a period of approximately 120 seconds, characterized by a significant peak in the power spectral density. Wavelet transform analysis revealed that this periodic fluctuation was most pronounced within 20-30 seconds after the evaporation source power was adjusted and then gradually decayed. Sample entropy calculations show that the sample entropy of the temperature series under normal operating conditions is 0.62, but increases to 0.85 during abnormal fluctuations, indicating increased system complexity. The parameter correlation matrix reveals that the Pearson correlation coefficient between the evaporation source temperature and the deposition rate is 0.83, with a mutual information value of 0.76, indicating a strong correlation. The correlation coefficient between the evaporation source temperature and the chamber pressure is 0.65, with a mutual information value of 0.58, indicating a moderate correlation. Further partial correlation analysis eliminated the indirect influence of the correlation and found that after controlling the evaporation source temperature, the partial correlation coefficient between the chamber pressure and the deposition rate dropped to 0.23, indicating that their correlation is primarily mediated by temperature changes. Cross-correlation analysis determined a 5-second time delay between power adjustment and temperature change, and a 3-second time delay between temperature change and deposition rate change. This multi-dimensional feature extraction and correlation analysis overcomes the limitations of traditional methods in uncovering deep relationships between parameters.

[0038] In a specific embodiment, the process of executing step S103 may specifically include the following steps: Performing cluster analysis on key parameter characteristics of the evaporation process, dividing process operation modes using a Gaussian mixture model, and obtaining a standard mode library for the evaporation process; For each process mode in the standard pattern library of the evaporation process, characteristic multidimensional distribution parameters are calculated, including a mean vector and a covariance matrix, to obtain a statistical representation of the mode; For the key parameter characteristics of the evaporation process collected in real time, the Mahalanobis distance and Hotelling statistics of the pattern statistical representation are calculated to obtain the parameter deviation; Based on the parameter deviation, a deep autoencoder network is constructed to encode and decode the input features, calculate the reconstruction error, and obtain the abnormal reconstruction loss; For the abnormal reconstruction loss, an adaptive threshold is set according to the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations, and obtain an abnormal level classification; Based on the abnormality level classification, the abnormality start time, duration, severity score and related parameter contribution are calculated to obtain the abnormality index of the evaporation process.

[0039] Specifically, a cluster analysis is performed on the key parameter characteristics of the evaporation process, and the Gaussian mixture model is used to divide the process operation modes, thereby generating a library of standard evaporation process patterns. The Gaussian mixture model is a probabilistic model that assumes that the data is generated by a mixture of multiple Gaussian distributions, each representing a data pattern. In its implementation, the appropriate number of clusters is first determined, and the Bayesian Information Criterion is automatically used to determine the optimal number of clusters. The Bayesian Information Criterion comprehensively considers model fit and complexity, finding a balance between the two to avoid overfitting. The model parameters are then initialized, including the weight, mean vector, and covariance matrix of each Gaussian distribution. The initialization method uses the results of the K-means++ algorithm. The model parameters are then iteratively optimized using the expectation-maximization algorithm. This algorithm consists of two steps: the expectation step calculates the posterior probability that each data point belongs to each Gaussian distribution, and the maximization step updates the parameters of each Gaussian distribution. This iterative process continues until the parameters converge or the maximum number of iterations is reached. Through Gaussian mixture model clustering, the key parameter characteristics of the evaporation process are divided into multiple process operation modes, such as the startup phase, the stable deposition phase, and the power adjustment phase, thereby forming a library of standard evaporation process patterns. For each process pattern in the standard evaporation process pattern library, characteristic multidimensional distribution parameters, including the mean vector and covariance matrix, are calculated to obtain a statistical representation of the pattern. The mean vector represents the average level of each parameter feature and is calculated by taking the arithmetic mean of all data points within each pattern. The covariance matrix describes the correlation structure and degree of variation between parameter features. It is calculated by calculating the covariance for each pair of features to form a covariance matrix. Covariance indicates the degree to which two variables vary together; a positive value indicates a positive correlation, a negative value indicates a negative correlation, and zero indicates no correlation. The diagonal elements represent the variance of each feature. By calculating the mean vector and covariance matrix for each process pattern, a statistical representation of the pattern is established, providing a benchmark for subsequent anomaly detection. Furthermore, expert knowledge is combined to provide physical interpretation and semantic annotation for each pattern, such as identifying a pattern as a "film formation stabilization period" or a "temperature adjustment period," enhancing the interpretability of the pattern.

[0040] For key parameter features of the evaporation process collected in real time, the Mahalanobis distance and Hotelling statistic are calculated based on the pattern statistical representation to determine parameter deviation. The Mahalanobis distance measures the distance between a data point and the center of the distribution, taking into account inter-feature correlation and making it suitable for multidimensional data analysis. The calculation method is to multiply the difference vector between the data point and the mean by the inverse of the covariance matrix, then multiply it by the transpose of the difference vector, and then take the square root to obtain the Mahalanobis distance. The Hotelling T² statistic is the square of the Mahalanobis distance multiplied by the number of samples. It is used in multivariate hypothesis testing to assess whether a data point deviates significantly from the center of the distribution. In actual processing, the most likely process mode is first identified by calculating the probability density between the real-time data and each pattern. The pattern with the highest probability density is selected as the current mode. The Mahalanobis distance and Hotelling T² statistic are then calculated based on the real-time data and the current pattern statistical representation. The results are compared with an adaptive threshold to determine whether there is significant deviation. The threshold setting takes into account the multidimensionality of the data and the pattern covariance structure. A critical value based on the chi-square distribution or F distribution is commonly used, with a confidence level of 95%. In this way, parameter deviation is obtained, reflecting the degree to which the real-time data deviates from the normal operating mode. Based on the parameter deviation, a deep autoencoder network is constructed to encode and decode the input features, calculate the reconstruction error, and obtain the anomaly reconstruction loss. A deep autoencoder is an unsupervised learning neural network consisting of an encoder and a decoder. Its goal is to learn a latent space representation of the input data and reconstruct the original input. The encoder uses a four-layer convolutional neural network, with 32, 64, 128, and 256 convolution kernels per layer, respectively, with a kernel size of 3×3. The activation function uses LeakyReLU to prevent gradient vanishing. Max pooling is applied after each convolution layer to reduce the feature dimension and extract key features. The decoder uses a symmetrical structure to the encoder and reconstructs the original input through transposed convolution (also known as deconvolution). The training process uses normal data and aims to minimize the reconstruction loss function, typically the mean squared error. After training, the network can reconstruct normal data well, but has poor reconstruction performance on anomaly data. In the anomaly detection stage, real-time data is input into the trained autoencoder, and the error between the input and the reconstructed output is calculated as the anomaly reconstruction loss. Calculation methods include the mean or weighted sum of point-to-point errors, taking into account the importance of different features. Since abnormal data often leads to large reconstruction errors, monitoring the size of the reconstruction error can effectively detect abnormal conditions.

[0041] For abnormal reconstruction loss, adaptive thresholds are set based on the distribution characteristics of historical data to distinguish between normal fluctuations and abnormal deviations, thereby achieving anomaly classification. First, a probability distribution fit is performed on the historically accumulated abnormal reconstruction loss data. A nonparametric probability density function is constructed using kernel density estimation to obtain a loss distribution model. Kernel density estimation is a nonparametric statistical method that estimates probability density by superimposing kernel functions (such as Gaussian kernels), avoiding strong assumptions about the data distribution shape. Based on the fitted loss distribution model, reconstruction loss values ​​at different probability quantiles, such as the 75th, 90th, 95th, and 99th percentiles, are calculated as the criteria for determining different levels, resulting in a set of static thresholds. To adapt to the different stages of the evaporation process, a piecewise linear transformation is applied to the static threshold set, combining the parameter distribution characteristics of the startup, stabilization, and finalization stages, to obtain stage-adaptive thresholds. This piecewise linear transformation accounts for the differences in normal fluctuation amplitudes across different stages, making the thresholds more targeted. To eliminate the impact of short-term fluctuations, a sliding window averaging and exponential smoothing method are applied to the stage-adaptive thresholds to obtain smoothed adaptive thresholds. The real-time anomaly reconstruction loss is compared with a smooth adaptive threshold and graded according to the degree to which it exceeds the thresholds for different quantiles. For example, exceeding the 90th quantile is considered a minor anomaly, exceeding the 95th quantile is considered a moderate anomaly, and exceeding the 99th quantile is considered a severe anomaly. This results in an initial anomaly level. Finally, considering the temporal characteristics of the anomaly, the initial anomaly level is processed temporally and the level is updated based on the duration and cumulative anomaly value, resulting in the final anomaly level. Based on the anomaly level classification, the anomaly onset time, duration, severity score, and associated parameter contributions are calculated to obtain an anomaly index for the vapor deposition process. The anomaly onset time is determined by backtracking the anomaly reconstruction loss sequence to find the time when the minor anomaly threshold is first exceeded. The duration is calculated from the onset of the anomaly until the anomaly disappears (the loss value returns to the normal range). The severity score is a comprehensive indicator that takes into account the anomaly level, duration, and impact range. A weighted algorithm is used to calculate the score, typically ranging from 1 to 10. Associated parameter contribution analysis aims to identify the key parameters that cause the anomaly. This is achieved by calculating the reconstruction error contribution of each original parameter or using SHAP values ​​to analyze the contribution of each parameter to the anomaly. The SHAP value, based on the Shapley value in game theory, fairly evaluates the contribution of each feature to the model output. By calculating these abnormal feature indicators, comprehensive evaporation process anomaly indicators are obtained, providing a basis for process control and adjustment.

[0042] In a specific embodiment, the process of performing the steps of encoding and decoding the input features may specifically include the following steps: Performing feature concatenation on the parameter deviation and key parameter features of the evaporation process to form an input vector, which is fed into the input layer of the deep autoencoder network to obtain features to be processed; Compress the features to be processed through a four-layer convolutional encoder, where the convolution kernel size of each layer is 3×3, the number of convolutional layer channels is 32, 64, 128, and 256, respectively, and activation function and pooling operation are applied after each convolution layer to obtain compressed features; The compressed features are further reduced in dimension through a fully connected layer, with the number of nodes being 512, 256, 128, and 64, respectively, to obtain a latent space feature representation; The latent space feature representation is restored through a four-layer transposed convolutional decoder. The transposed convolutional decoder adopts a structure symmetrical to the encoder, and the number of channels is 256, 128, 64, and 32, respectively, to obtain the reconstructed features; Calculating the Euclidean distance and cosine similarity between the reconstructed features and the original input features, constructing a weighted loss function, and obtaining a feature reconstruction loss; The feature reconstruction loss is compared and analyzed with the reconstruction loss of continuous time series samples to identify abnormal mutation points and gradual trends, thereby obtaining the abnormal reconstruction loss.

[0043] Specifically, the parameter deviation and key parameter features of the evaporation process are concatenated to form an input vector, which is fed into the input layer of a deep autoencoder network to obtain the features to be processed. Feature concatenation is the process of combining multiple related features into a unified vector. In this method, the parameter deviation calculated using the Mahalanobis distance and Hotelling statistics is combined with the key parameter features of the evaporation process. In specific implementation, the parameter deviation is directly concatenated as a scalar value to the key parameter feature vector to form an augmented feature vector. For example, the evaporation source temperature feature set, the chamber pressure feature set, and the deposition rate feature set each contain multiple time-domain, frequency-domain, and time-frequency features. These features, together with the deviation metric, form the complete input vector. To ensure that the numerical scales of the input vector dimensions are similar, normalization is applied to map each eigenvalue to the interval [0, 1]. The input vector is typically reorganized into a two-dimensional tensor for easier processing by a convolutional neural network. For example, a one-dimensional vector can be reshaped into a matrix to preserve the spatial or temporal relationships of the original features and obtain the features to be processed. The features to be processed are compressed through a four-layer convolutional encoder. Each convolutional encoder layer has a 3×3 kernel size and the number of channels in each convolutional layer is 32, 64, 128, and 256, respectively. Activation functions and pooling operations are applied after each convolution layer to obtain compressed features. The convolutional encoder is the encoding portion of a deep autoencoder, extracting high-level representations of the input features through convolution operations. The first convolutional layer convolves the input features with 32 3×3 kernels, with a stride of 1 and "same" padding to maintain the size of the feature map. The convolution operation performs a dot product of a sliding window with the input features to capture local feature patterns. After the convolution, the LeakyReLU activation function is applied, with the form f(x)=max(0.01x,x). Compared to the traditional ReLU activation function, LeakyReLU has a smaller slope in negative regions, avoiding the "dead neuron" problem. After the activation, a 2×2 max pooling operation with a stride of 2 is applied to halve the size of the feature map while preserving significant features. The second convolution layer uses 64 3×3 convolution kernels to process the feature maps output by the first layer. The convolution, activation, and pooling settings are the same as the first layer. The third layer uses 128 convolution kernels, and the fourth layer uses 256 convolution kernels. The number of feature channels increases layer by layer to extract more abstract feature representations. After four layers of convolution and pooling, the feature map size is significantly reduced while the number of channels increases, achieving information compression and abstract representation, resulting in compressed features.

[0044] The compressed features are further reduced in dimension through fully connected layers, with the number of nodes increasing to 512, 256, 128, and 64, respectively, to obtain the latent space feature representation. The fully connected layers further compress the spatial features extracted by the convolutional layers into a compact one-dimensional vector representation. The compressed features after the four convolutional layers are flattened into a one-dimensional vector and then connected to the first fully connected layer, which contains 512 neurons. Each neuron is fully connected to all elements of the input vector, and a weighted sum is calculated before applying the LeakyReLU activation function. To prevent overfitting, Dropout regularization is applied after the fully connected layer, randomly dropping some neuron outputs with a dropout ratio of 0.3. The second fully connected layer contains 256 neurons and also applies the LeakyReLU activation function and Dropout. The third and fourth layers contain 128 and 64 neurons, respectively, forming a "funnel" structure with decreasing number of neurons. The final 64-dimensional vector is the latent space feature representation, also known as the "latent code" or "bottleneck layer," which is a highly compressed and abstract representation of the entire input feature. The latent space feature dimension is much smaller than the original input, forcing the network to learn the intrinsic structure of the data and discard redundant and noisy information. Due to the inherent regularity of normal operating data, its latent space representation is typically more compact and consistent. However, abnormal data is difficult to compress effectively, resulting in significantly different latent space features. The latent space feature representation is restored through a four-layer transposed convolutional decoder. The transposed convolutional decoder adopts a symmetrical structure to the encoder, with the number of channels being 256, 128, 64, and 32, respectively, to obtain reconstructed features. Transposed convolution is the inverse of convolution, mapping low-dimensional features back to a higher-dimensional space. This is also known as deconvolution or deconvolution. First, the 64-dimensional latent space features are reshaped into a small feature map and connected to the first transposed convolution layer. This layer uses 256 3×3 transposed convolution kernels with a stride of 2. The feature map size is increased by inserting blank pixels and applying convolution operations. The LeakyReLU activation function is also applied after the transposed convolution. The second layer uses 128 transposed convolution kernels, the third layer uses 64, and the fourth layer uses 32. The number of channels decreases with each layer, symmetrical with the encoder. The four layers of transposed convolution operations gradually restore the spatial dimensions of the feature map, ultimately outputting reconstructed features with the same size as the original input. The decoder's structural design mirrors the encoder's principle to ensure balanced information flow. For example, while pooling downsamples in the encoder, transposed convolution performs upsampling in the decoder; while the number of channels increases in the encoder, it decreases in the decoder. This symmetrical design facilitates efficient information reconstruction, and ideally, the reconstructed features output should be close to the original input features.

[0045] Euclidean distance and cosine similarity are calculated between the reconstructed features and the original input features to construct a weighted loss function, resulting in the feature reconstruction loss. Euclidean distance measures the linear distance between two vectors in Euclidean space and is calculated by taking the square root of the sum of the squared differences between the corresponding elements. In implementation, the difference between each corresponding element of the reconstructed and original input features is calculated, squared, summed, and the square root is taken to obtain the overall reconstruction error. Cosine similarity measures the directional similarity between two vectors and is calculated by dividing the dot product of the vectors by the product of their respective moduli. Cosine similarity has a range of [-1, 1], with values ​​closer to 1 indicating greater similarity. Both metrics have their own advantages: Euclidean distance is sensitive to amplitude differences, while cosine similarity focuses on directional consistency. The two metrics are combined to construct a weighted loss function: L = α × Euclidean distance - β × cosine similarity, where α and β are weighting coefficients that are adjusted based on the specific application; typically α = 0.7 and β = 0.3. This combined loss function considers both the accuracy of feature reconstruction and the structural similarity of feature vectors, improving the sensitivity and accuracy of anomaly detection. The calculated feature reconstruction loss reflects the degree of deviation of the input data from the normal pattern, with larger loss values ​​indicating more significant deviation.

[0046] The feature reconstruction loss is compared with the reconstruction loss of continuous time series samples to identify abnormal breakpoints and gradual trends, resulting in the abnormal reconstruction loss. Continuous time series samples refer to multiple samples collected continuously in chronological order. The comparative analysis involves comparing the reconstruction loss of the current sample with the reconstruction loss of historical samples. First, a time series of reconstruction loss is constructed, recording and storing the feature reconstruction loss at each time point. Then, a sliding window technique is applied, typically with a window length of 10-20 time points, to calculate the statistical characteristics of the reconstruction loss within the window, such as the mean, standard deviation, and trend slope. To identify abnormal breakpoints, a change point detection algorithm is used to calculate significant changes in reconstruction loss between adjacent windows. Breakpoints are marked if the difference in the mean loss between adjacent windows exceeds three standard deviations. To identify gradual trends, linear regression is used to estimate the slope of the loss within the window. A slope significantly greater than zero indicates increasing anomaly, while a slope significantly less than zero indicates recovery. By comprehensively considering the absolute value, relative change, and trend of the reconstruction loss, an abnormal reconstruction loss index is constructed. This index not only reflects the current anomaly level but also includes information on the anomaly's evolving trend. The anomaly reconstruction loss metric is a comprehensive score that takes into account both point anomalies (instantaneous deviations) and contextual anomalies (abnormal change patterns), enhancing the robustness and accuracy of anomaly detection.

[0047] In a specific embodiment, the process of performing the abnormal reconstruction loss step may specifically include the following steps: Perform probability distribution fitting on the historically accumulated abnormal reconstruction loss data, use the kernel density estimation method to construct a non-parametric probability density function, and obtain the loss distribution model; Based on the loss distribution model, the reconstruction loss values ​​of multiple quantiles are calculated and set as judgment criteria of different levels to obtain a static threshold set; Applying piecewise linear transformation to the static threshold set in combination with parameter distributions of the start-up phase, the stable phase, and the end phase of the evaporation process to obtain a stage-adaptive threshold; For the adaptive threshold of the stage, a sliding window average and exponential smoothing method is applied to eliminate the influence of short-term fluctuations to obtain a smooth adaptive threshold; Comparing the real-time anomaly reconstruction loss with the smooth adaptive threshold, and grading the loss according to the degree of exceeding the quantile thresholds of different levels to obtain an initial anomaly level; The initial abnormality level is subjected to time series processing, and the abnormality level is updated according to the abnormality duration and the accumulated abnormality value to obtain the abnormality level classification.

[0048] Specifically, a probability distribution fit is performed on historically accumulated abnormal reconstruction loss data. Kernel density estimation is used to construct a nonparametric probability density function (PDF) to obtain a loss distribution model. Kernel density estimation is a nonparametric statistical method that does not presuppose a data distribution form but instead directly learns distribution characteristics from the data itself. During implementation, accumulated abnormal reconstruction loss data under normal operating conditions is extracted from a historical database, typically containing thousands to tens of thousands of data points. The basic idea of ​​kernel density estimation is to treat each observation point as a source of probability mass and smooth these masses using a kernel function. During the calculation, an appropriate kernel function (usually a Gaussian kernel) and bandwidth parameter are selected. The bandwidth controls the degree of smoothing; a too small value leads to overfitting, while a too large value leads to underfitting. The Silverman rule is generally used to automatically determine the optimal bandwidth. By placing the kernel function at each data point and summing the values, an overall probability density function is obtained, forming a continuous and smooth distribution curve that accurately reflects the distribution characteristics of abnormal reconstruction loss, including multimodal characteristics and tail behavior. Based on the loss distribution model, reconstruction loss values ​​at multiple quantiles are calculated and set as criteria for different levels of judgment to obtain a set of static thresholds. Percentiles are points that divide a probability distribution into regions of equal probability. Typical percentiles include 75% (minor anomaly threshold), 90% (moderate anomaly threshold), 95% (severe anomaly threshold), and 99% (extreme anomaly threshold). The calculation process is based on a reverse lookup of the cumulative distribution function (CDF). This involves integrating the probability density function of the loss distribution model to obtain the CDF, then finding the corresponding loss value when the cumulative probability reaches a specified percentage. For example, the 95% quantile indicates that 95% of normal data has a loss value below this threshold, and data points with a loss value exceeding this threshold have a 95% probability of being anomalies. By calculating different quantiles, a multi-level anomaly determination standard is established, forming a set of static thresholds. This probability distribution-based threshold setting method is more statistically based than a simple fixed threshold and can automatically adjust the determination criteria based on the actual data distribution characteristics, reducing human intervention.

[0049] A piecewise linear transformation is applied to the static threshold set, combined with the parameter distributions during the startup, stabilization, and final stages of the evaporation process, to obtain stage-adaptive thresholds. The evaporation process is typically divided into three main stages: the startup phase (initial heating and evaporation stabilization), the stabilization phase (primary film formation), and the final stage (power reduction and cooling). Parameter fluctuation characteristics vary significantly between stages, with significant parameter fluctuations during the startup and final stages, while high stability is required during the stabilization phase. Piecewise linear transformation is a method for adjusting static thresholds based on the characteristics of the process stages. The basic principle is to apply different transformation coefficients to different stages. In implementation, the current process stage is first identified based on the state of key parameters such as evaporation power, chamber temperature, and deposition rate. Threshold adjustment coefficients are then set for each stage. During the startup phase, due to significant fluctuations, the threshold is relaxed by 20%-30%. During the stabilization phase, strict control is required, and the threshold remains unchanged or slightly tightened by 5%-10%. Similarly, the threshold is appropriately relaxed during the final stage. Using a stage identification function and a linear transformation function, the static threshold is mapped to a dynamic threshold that matches the process stage, ensuring that anomaly determination is more consistent with actual process characteristics.

[0050] Sliding window averaging and exponential smoothing are applied to the stage-adaptive threshold to eliminate the effects of short-term fluctuations, resulting in a smoothed adaptive threshold. Sliding window averaging is a commonly used smoothing method for time series data. It replaces the original value with the arithmetic mean of the data within a fixed-length window. In implementation, an appropriate window length is set (typically 10-20 data points), and sliding window averaging is applied to the stage-adaptive threshold sequence. For the smoothed threshold at time point t, the average of the thresholds within half the window length before and after t is calculated. Exponential smoothing assigns exponentially decreasing weights to data at different time points, with the most recent data receiving the highest weight and gradually decreasing weights over time. In implementation, an appropriate smoothing coefficient α is selected (typically between 0.1 and 0.3), with the current smoothed value being a weighted combination of the previous smoothed value and the current observed value. Exponential smoothing is particularly well-suited for data with trending changes, smoothing noise while preserving the data's trend. Combining these two smoothing methods: sliding window averaging addresses local fluctuations, while exponential smoothing tracks long-term trends. The resulting smoothed adaptive threshold eliminates the effects of short-term fluctuations while adapting to gradual changes in process parameters.

[0051] The real-time anomaly reconstruction loss is compared with the smoothing adaptive threshold and graded according to the degree to which it exceeds different quantile thresholds to obtain an initial anomaly level. The real-time anomaly reconstruction loss is the current-time loss value calculated by a deep autoencoder network, while the smoothing adaptive threshold is a multi-level threshold sequence obtained through the aforementioned steps. The comparison process first calculates the normalized deviation, which is the ratio of the real-time loss value to the current threshold. The initial anomaly level is then determined based on the deviation and the threshold level exceeded. Anomalies are generally categorized into three levels: minor (loss value exceeds the 75th percentile threshold but does not exceed the 90th percentile threshold), moderate (exceeds the 90th percentile threshold but does not exceed the 95th percentile threshold), and severe (exceeds the 95th percentile threshold). Extreme cases, such as loss values ​​exceeding the 99th percentile threshold, are marked as extreme anomalies. The initial anomaly level is an instantaneous assessment result calculated at a single time point, reflecting the degree of anomaly in the current state and providing a basis for subsequent time series processing.

[0052] The initial anomaly level is processed through time series analysis. The anomaly level is updated based on the duration of the anomaly and the cumulative anomaly value, resulting in an anomaly classification. This time series processing takes into account the persistence and cumulative effects of anomaly states, avoiding over-alarming of isolated anomalies and under-reporting of persistent anomalies. During implementation, an anomaly status buffer is maintained to record the anomaly level sequence for a recent period (e.g., the past 30 minutes). Anomaly levels within the buffer are analyzed through time series analysis, including duration statistics and cumulative anomaly value calculation. Duration statistics calculate the duration of consecutive anomaly states. If moderate anomalies occur continuously for a specified time (e.g., 60 seconds), the anomaly is upgraded to a severe anomaly. Cumulative anomaly value calculation considers the cumulative effect of anomalies by weighting the anomaly values ​​in the buffer, with the weights decaying over time. When the cumulative anomaly value exceeds a set threshold, the anomaly level is raised, even if the individual anomalies were not severe. Furthermore, the trend of anomalies is considered, with trend analysis performed on the anomaly level sequence. If the anomaly severity continues to increase, the warning is upgraded; if the anomaly severity continues to decrease, the warning is downgraded. Through these time series processing rules, more accurate abnormality level classification can be obtained, false alarms and missed alarms can be reduced, and the reliability and practicality of the monitoring system can be improved.

[0053] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Integrating the abnormal indicators of the evaporation process with the historical operation data of the evaporation process to construct a digital twin model of the evaporation process, simulate the effects of different control parameters on thin film growth, and obtain a control parameter influence map; Based on the control parameter influence mapping, a policy gradient algorithm is used to train a control strategy network, taking the abnormality index of the evaporation process as input, outputting the adjustment amount of the RF power and substrate temperature, and obtaining a parameter control strategy; Performing physical constraint checks on the parameter control strategy to limit the range and rate of change of RF power and substrate temperature to prevent system instability caused by sudden adjustments, thereby obtaining a safe control strategy; Based on the safety control strategy, large-scale parameter adjustments are decomposed into a sequence of multiple small steps to form a smooth transition control trajectory and obtain a progressive control sequence; Converting the progressive control sequence into a command signal in a standard industrial control interface format, and writing the command signal into a programmable logic controller of the evaporation device via a communication interface to obtain an execution control command; The execution result of the execution control instruction is monitored in real time, and the abnormal indicator changes before and after the parameter adjustment are recorded to form control effect evaluation data, and obtain the target control plan for the thin film growth in the evaporation chamber.

[0054] Specifically, abnormal indicators of the evaporation process are integrated with historical evaporation process data to construct a digital twin model of the evaporation process. This model simulates the impact of different control parameters on thin film growth and generates a control parameter impact map. A digital twin model is a virtual replica of a physical system, capable of reflecting state changes in real time. The construction process first extracts historical operational data from a database, including process parameters (such as evaporation source temperature, chamber pressure, and substrate temperature), control parameters (such as RF power and gas flow), and quality indicators (such as film thickness uniformity and optical properties). Data preprocessing includes removing outliers, filling in missing values, and normalizing the data. A hybrid model architecture is then constructed, integrating a physical model with a data-driven model. The physical model describes the theoretical relationships between parameters based on evaporation dynamics, particle transport, and film formation theory. The data-driven model uses a deep neural network to learn the complex nonlinear relationships in the data. Model training uses a hybrid loss function that simultaneously considers prediction error and the degree of violation of physical constraints, achieving a deep fusion of physical information and data. The resulting digital twin model accurately simulates the dynamic behavior of the evaporation process. By inputting different combinations of control parameters (such as RF power and substrate temperature), it predicts the film growth process and final quality, generating a control parameter impact map. Based on this control parameter impact map, a policy gradient algorithm is used to train a control policy network. This network takes evaporation process anomaly indicators as input and outputs adjustments to RF power and substrate temperature, resulting in a parameter control policy. The policy gradient algorithm is a reinforcement learning method that directly optimizes the control policy rather than a value function. The evaporation process is first modeled as a Markov decision process, with the state space consisting of evaporation process anomaly indicators and key process parameters, and the action space consisting of adjustments to RF power and substrate temperature. The reward function comprehensively considers film quality, energy consumption, and device stability. This implementation utilizes the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, which comprises a policy network and a dual-Q network. The policy network consists of four fully connected layers. The input layer receives the current state (abnormality indicators and process parameters). The number of nodes in the intermediate hidden layers is 400, 300, 200, and 100, respectively. The activation function is ReLU. The output layer uses a tanh activation function to map the output to the range [-1, 1], and then a linear transformation to the actual action range. The Q network has a similar structure, but the input includes both the state and the action, and the output is a single Q value. The training process uses experience replay and target network techniques to reduce sample correlation and improve stability. The experience pool capacity is set to 1 million, with 128 batches sampled each time, and a discount factor of 0.99. To enhance exploration capabilities, adaptive Gaussian noise is added to the policy network parameters. Training iterations continue until performance converges or the preset number of iterations is reached. Ultimately, a parameter control strategy is obtained that outputs optimal control adjustments for different abnormal conditions.

[0055] Physical constraint checking is performed on the parameter control strategy to limit the range and rate of change of RF power and substrate temperature, preventing system instability caused by abrupt adjustments and ultimately resulting in a safe control strategy. Physical constraint checking is a critical step in ensuring that the control strategy complies with actual process constraints. First, a safe operating range for RF power and substrate temperature is defined, typically based on device specifications and process requirements. For example, the RF power range is 10%-90% of the rated power, and the substrate temperature range is from room temperature to the device's maximum temperature tolerance. Then, parameter change rate limits are defined to prevent system oscillation caused by abrupt adjustments. Typical limits include a per-second change rate of RF power exceeding 5% of the rated power and a per-minute change rate of substrate temperature exceeding 20°C. The constraint checking process consists of two steps: range checking and rate checking. The range check ensures that the adjusted target value of the parameter output by the control strategy is within a safe range; if it exceeds, it is truncated to the boundary value. The rate check ensures that the parameter change rate does not exceed the limit; if it does, it is adjusted to the maximum allowable rate. Furthermore, parameter interaction constraint checking is required, as certain parameter combinations may lead to unsafe conditions, such as high power at high temperatures, which may damage the device. By applying these constraints to the policy network output, a safe control strategy is obtained that meets both control objectives and physical constraints.

[0056] Based on a safe control strategy, large parameter adjustments are decomposed into a sequence of small steps, forming a smooth transition control trajectory, resulting in a progressive control sequence. The progressive control sequence aims to achieve smooth parameter adjustments and avoid sudden changes and oscillations in system response. In specific implementation, a large adjustment is first determined by determining whether it exceeds a preset threshold, such as when the RF power adjustment exceeds 10% of the rated power or when the substrate temperature adjustment exceeds 30°C. For large adjustments, trajectory planning is used to decompose them into multiple small steps. Trajectory planning considers the starting point (current parameter value), the target point (target parameter value), and the system dynamics to design an appropriate transition trajectory. Common methods include linear interpolation, polynomial interpolation, and S-curve interpolation. Linear interpolation is the simplest, evenly distributing the adjustment amount across multiple time steps. Polynomial interpolation achieves a smooth transition between the start and end points, reducing sudden acceleration changes. S-curve interpolation further limits jerk, providing the smoothest transition. The time step size and total transition time are determined based on the adjustment magnitude and system response characteristics. Large adjustments are typically decomposed into 5-10 time steps. For example, a 500W RF power adjustment can be broken down into five small 100W adjustments, with 10-second intervals between each step, creating a smooth 50-second transition trajectory. Small adjustments are executed directly as planned without breaking them down. This ultimately results in a series of control commands that are continuous in time and smooth in amplitude, known as a progressive control sequence. The progressive control sequence is converted into command signals in a standard industrial control interface format and written to the programmable logic controller of the evaporation equipment via a communication interface to generate execution control instructions. The standard industrial control interface format is a universal data exchange format for industrial control systems. Common standards include OPC UA (OPC Unified Architecture), Modbus, and Profinet. The conversion process first identifies the control system type and interface protocol used by the evaporation equipment. Then, the control parameters are encapsulated into standard data packets according to the protocol specifications. For example, for systems using the OPC UA protocol, the RF power and substrate temperature adjustment values ​​must be encapsulated as OPC UA data nodes, with the correct data type, range, and unit specified. For Modbus protocols, these must be mapped to specific register addresses. Once the data packet is prepared, it is sent to the evaporation equipment's programmable logic controller (PLC) via a physical layer communication interface such as Ethernet, serial port, or fieldbus. The communication process must address issues such as timeout retransmission, data verification, and error recovery to ensure reliable transmission of the instructions. Upon receiving the instructions, the PLC converts the digital instructions into analog control signals, driving the power control unit and temperature control unit to adjust the actual parameters. The entire communication and execution process requires strict timing control and status feedback, typically using a handshake mechanism to confirm the execution status of the instructions, forming a complete execution control flow.

[0057] The execution results of control instructions are monitored in real time, and changes in abnormal indicators before and after parameter adjustments are recorded. This data is used to evaluate control effectiveness and determine the target control plan for thin film growth within the evaporation chamber. Real-time monitoring requires the acquisition of multi-source data, including actual control parameter values ​​(such as actual readings of RF power and substrate temperature), process parameter responses (such as changes in evaporation source temperature and chamber pressure), and changes in abnormal indicators (such as real-time values ​​of reconstruction loss and deviation). The data acquisition frequency is determined by the control cycle, typically 1-10 times per second. The monitored data is stored in a time series database after preprocessing, which includes noise filtering, outlier detection, and data alignment. The effectiveness of parameter adjustments is then evaluated, with key metrics calculated, including response time (the delay from instruction issuance to the start of a significant parameter change), stabilization time (the time required to reach the new steady state after parameter adjustment), overshoot (the extent to which a parameter briefly exceeds the target value), and improvement in abnormal indicators (the magnitude of the change in abnormal indicators before and after adjustment). The evaluation results are used to verify the effectiveness of the control strategy and adjust its parameters. If an abnormal indicator doesn't significantly improve after a specific adjustment, the current control strategy is ineffective and requires increasing the adjustment range or trying other control parameters. If the abnormal indicator quickly returns to normal, the control strategy is effective. Through continuous monitoring and evaluation, we continuously optimize control parameter combinations and adjustment strategies, ultimately forming a library of target control solutions suitable for different abnormal situations to guide automatic control decisions in subsequent production processes.

[0058] The above describes the evaporation process monitoring method based on data analysis in the embodiment of the present application. The following describes the evaporation process monitoring system based on data analysis in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the evaporation process monitoring system based on data analysis includes: The acquisition module 201 is used to collect raw data from the temperature sensor, pressure sensor, power monitor and thickness monitoring system installed in the evaporation chamber, and obtain evaporation process parameter data through filtering and standardization; A calculation module 202 is used to perform time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extract the variation characteristics of the evaporation source temperature, chamber pressure and deposition rate, and obtain the key parameter characteristics of the evaporation process; A setting module 203 is configured to perform deviation analysis based on the key parameter characteristics of the evaporation process by setting dynamic thresholds, calculate the degree of deviation between the real-time value of the parameter and the standard value, and obtain an abnormality index of the evaporation process; The adjustment module 204 is used to adjust the RF power input and substrate temperature control in real time according to the abnormal indicators of the evaporation process, generate correction control instructions, and obtain a target control solution for thin film growth in the evaporation chamber.

[0059] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, and when the instructions are run on a computer, the computer executes the steps of the evaporation process monitoring method based on data analysis.

[0060] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for monitoring an evaporation process based on data analysis, characterized in that: The method comprises: Collect raw data from the temperature sensor, pressure sensor, power monitor, and thickness monitoring system installed in the evaporation chamber, and obtain evaporation process parameter data through filtering and standardization; Performing time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extracting the variation characteristics of the evaporation source temperature, chamber pressure and deposition rate, and obtaining the key parameter characteristics of the evaporation process; Based on the key parameter characteristics of the evaporation process, deviation analysis is performed by setting dynamic thresholds, calculating the degree of deviation between the real-time value of the parameter and the standard value, and obtaining an abnormality index of the evaporation process; According to the abnormal indicators of the evaporation process, the radio frequency power input and substrate temperature control are adjusted in real time, and correction control instructions are generated to obtain a target control plan for thin film growth in the evaporation chamber.

2. The method for monitoring the evaporation process based on data analysis according to claim 1, characterized in that: The raw data of the temperature sensor, pressure sensor, power monitor and thickness monitoring system installed in the evaporation chamber are collected, and the evaporation process parameter data is obtained through filtering and standardization, including: Temperature data is collected from temperature sensors installed at the evaporation source, substrate, and chamber wall. The original temperature distribution data is obtained through local regression model combined with sliding window analysis. The vacuum degree data is collected from the pressure sensors installed at different positions in the cavity, and the original pressure gradient data is obtained through multi-point average sampling and outlier elimination. The power input data recorded in real time by the power monitor is sampled at a fixed frequency and marked with mutation points to obtain the original power fluctuation data. Applying wavelet transform filtering technology to the temperature distribution raw data, pressure gradient raw data, and power fluctuation raw data to retain signal characteristics while removing high-frequency noise to obtain filtered parameter data; Performing zero-mean normalization processing on the filtered parameter data, converting parameters of different dimensions into a standard distribution format, and obtaining standardized parameter data; The sensor data of different sampling frequencies in the standardized parameter data are aligned by time stamps to establish a unified time base, and the first-order derivative and moving average features are extracted to obtain the evaporation process parameter data.

3. The method for monitoring the evaporation process based on data analysis according to claim 1, characterized in that: The time-frequency analysis and parameter correlation calculation of the evaporation process parameter data are performed to extract the variation characteristics of the evaporation source temperature, chamber pressure and deposition rate, and obtain the key parameter characteristics of the evaporation process, including: The evaporation process parameter data is divided into fixed windows, and the statistical moment characteristics and trend characteristics of the evaporation source temperature, chamber pressure and deposition rate in each window are calculated to obtain the parameter time domain characteristics; Performing a fast Fourier transform on the evaporation process parameter data, extracting a power spectral density curve, identifying periodic fluctuation components of the evaporation source temperature, chamber pressure, and deposition rate, and obtaining parameter frequency domain characteristics; Applying wavelet transform or Hilbert transform to the evaporation process parameter data to capture the time-varying frequency characteristics of the evaporation source temperature, chamber pressure and deposition rate to obtain parameter time-frequency characteristics; Calculating sample entropy and multi-scale entropy for the parameter time domain features, parameter frequency domain features, and parameter time-frequency features, quantifying the complexity of parameter fluctuations, and obtaining parameter complexity features; Based on the parameter time domain characteristics, parameter frequency domain characteristics, parameter time-frequency characteristics and parameter complexity characteristics, the Pearson correlation coefficient and mutual information value between the evaporation source temperature, the chamber pressure and the deposition rate are calculated to obtain a parameter correlation matrix; Partial correlation analysis and conditional independence test are applied to the parameter correlation matrix to eliminate indirect correlation effects, determine the causal relationship and time delay between parameters, and obtain the key parameter characteristics of the evaporation process.

4. The method for monitoring the evaporation process based on data analysis according to claim 1, characterized in that: Based on the key parameter characteristics of the evaporation process, the deviation analysis is performed by setting a dynamic threshold, calculating the degree of deviation between the real-time value of the parameter and the standard value, and obtaining the abnormal index of the evaporation process, including: Performing cluster analysis on key parameter characteristics of the evaporation process, dividing process operation modes using a Gaussian mixture model, and obtaining a standard mode library for the evaporation process; For each process mode in the standard pattern library of the evaporation process, characteristic multidimensional distribution parameters are calculated, including a mean vector and a covariance matrix, to obtain a statistical representation of the mode; For the key parameter characteristics of the evaporation process collected in real time, the Mahalanobis distance and Hotelling statistics of the pattern statistical representation are calculated to obtain the parameter deviation; Based on the parameter deviation, a deep autoencoder network is constructed to encode and decode the input features, calculate the reconstruction error, and obtain the abnormal reconstruction loss; For the abnormal reconstruction loss, an adaptive threshold is set according to the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations, and obtain an abnormal level classification; Based on the abnormality level classification, the abnormality start time, duration, severity score and related parameter contribution are calculated to obtain the abnormality index of the evaporation process.

5. The method for monitoring the evaporation process based on data analysis according to claim 4, characterized in that: The method of constructing a deep autoencoder network based on the parameter deviation, encoding and decoding the input features, calculating the reconstruction error, and obtaining the abnormal reconstruction loss includes: Performing feature concatenation on the parameter deviation and key parameter features of the evaporation process to form an input vector, which is fed into the input layer of the deep autoencoder network to obtain features to be processed; Compress the features to be processed through a four-layer convolutional encoder, where the convolution kernel size of each layer is 3×3, the number of convolutional layer channels is 32, 64, 128, and 256, respectively, and activation function and pooling operation are applied after each convolution layer to obtain compressed features; The compressed features are further reduced in dimension through a fully connected layer, with the number of nodes being 512, 256, 128, and 64, respectively, to obtain a latent space feature representation; The latent space feature representation is restored through a four-layer transposed convolutional decoder. The transposed convolutional decoder adopts a structure symmetrical to the encoder, and the number of channels is 256, 128, 64, and 32, respectively, to obtain the reconstructed features; Calculating the Euclidean distance and cosine similarity between the reconstructed features and the original input features, constructing a weighted loss function, and obtaining a feature reconstruction loss; The feature reconstruction loss is compared and analyzed with the reconstruction loss of continuous time series samples to identify abnormal mutation points and gradual trends, thereby obtaining the abnormal reconstruction loss.

6. The method for monitoring the evaporation process based on data analysis according to claim 4, characterized in that: The abnormal reconstruction loss is configured by setting an adaptive threshold based on the distribution characteristics of historical data to distinguish normal fluctuations from abnormal deviations, thereby obtaining an abnormal level classification, including: Perform probability distribution fitting on the historically accumulated abnormal reconstruction loss data, use the kernel density estimation method to construct a non-parametric probability density function, and obtain the loss distribution model; Based on the loss distribution model, the reconstruction loss values ​​of multiple quantiles are calculated and set as judgment criteria of different levels to obtain a static threshold set; Applying piecewise linear transformation to the static threshold set in combination with parameter distributions of the start-up phase, the stable phase, and the end phase of the evaporation process to obtain a stage-adaptive threshold; For the adaptive threshold of the stage, a sliding window average and exponential smoothing method is applied to eliminate the influence of short-term fluctuations to obtain a smooth adaptive threshold; Comparing the real-time anomaly reconstruction loss with the smooth adaptive threshold, and grading the loss according to the degree of exceeding the quantile thresholds of different levels to obtain an initial anomaly level; The initial abnormality level is subjected to time series processing, and the abnormality level is updated according to the abnormality duration and the accumulated abnormality value to obtain the abnormality level classification.

7. The method for monitoring the evaporation process based on data analysis according to claim 1, characterized in that: According to the abnormal indicators of the evaporation process, the RF power input and substrate temperature control are adjusted in real time to generate a correction control instruction to obtain a target control scheme for thin film growth in the evaporation chamber, including: Integrating the abnormal indicators of the evaporation process with the historical operation data of the evaporation process to construct a digital twin model of the evaporation process, simulate the effects of different control parameters on thin film growth, and obtain a control parameter influence map; Based on the control parameter influence mapping, a policy gradient algorithm is used to train a control strategy network, taking the abnormality index of the evaporation process as input, outputting the adjustment amount of the RF power and substrate temperature, and obtaining a parameter control strategy; Performing physical constraint checks on the parameter control strategy to limit the range and rate of change of RF power and substrate temperature to prevent system instability caused by sudden adjustments, thereby obtaining a safe control strategy; Based on the safety control strategy, large-scale parameter adjustments are decomposed into a sequence of multiple small steps to form a smooth transition control trajectory and obtain a progressive control sequence; Converting the progressive control sequence into a command signal in a standard industrial control interface format, and writing the command signal into a programmable logic controller of the evaporation device via a communication interface to obtain an execution control command; The execution result of the execution control instruction is monitored in real time, and the abnormal indicator changes before and after the parameter adjustment are recorded to form control effect evaluation data, and obtain the target control plan for the thin film growth in the evaporation chamber.

8. A evaporation process monitoring system based on data analysis, characterized in that: For implementing the evaporation process monitoring method based on data analysis according to any one of claims 1 to 7, the evaporation process monitoring system based on data analysis comprises: The acquisition module is used to collect raw data from the temperature sensor, pressure sensor, power monitor and thickness monitoring system installed in the evaporation chamber, and obtain the evaporation process parameter data through filtering and standardization; A calculation module is used to perform time-frequency analysis and parameter correlation calculation on the evaporation process parameter data, extract the variation characteristics of the evaporation source temperature, chamber pressure and deposition rate, and obtain the key parameter characteristics of the evaporation process; A setting module is used to perform deviation analysis based on the key parameter characteristics of the evaporation process by setting dynamic thresholds, calculate the degree of deviation between the real-time value of the parameter and the standard value, and obtain an abnormality indicator of the evaporation process; The adjustment module is used to adjust the radio frequency power input and substrate temperature control in real time according to the abnormal indicators of the evaporation process, generate correction control instructions, and obtain a target control plan for thin film growth in the evaporation chamber.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the method for monitoring a vapor deposition process based on data analysis according to any one of claims 1 to 7 .

Images