MEMS capacitor film vacuum gauge diaphragm life prediction system based on deep learning

Abstract

The invention is suitable for the field of health detection and life prediction of the whole life cycle of an MEMS capacitor film, and provides an MEMS capacitor film vacuum gauge diaphragm life prediction system based on deep learning. The device comprises a molecular pump set, a first vacuum valve, a second vacuum valve, a first cold cathode ionization vacuum gauge, a second cold cathode ionization vacuum gauge, a first vacuum chamber, a second vacuum chamber, a third vacuum chamber, a first LCR digital bridge, a second LCR digital bridge, a first MEMS diaphragm chamber, a second MEMS diaphragm chamber, a temperature control box, an upper computer, a controller and an electromagnetic valve. According to the method, more reliable life prediction can be realized, and better robustness is achieved. According to the method, test equipment does not need to be developed, the method can be realized through conventional equipment and methods, the evaluation accuracy is high, the efficiency is high, and the operation is convenient and simple.

Description

Technical Field

[0001] This invention relates to the field of health monitoring and lifespan prediction of MEMS capacitor films throughout their entire lifespan, and particularly to a deep learning-based MEMS capacitor film vacuum gauge diaphragm lifespan prediction system. Background Technology

[0002] This invention relates to the field of microelectromechanical systems (MEMS) technology, and more specifically, to a method for predicting the lifespan and providing predictable maintenance for key sensing components of MEMS thin-film vacuum gauges.

[0003] Microelectromechanical systems (MEMS) thin-film vacuum gauges are precision measuring devices manufactured using semiconductor micromachining technology. Due to their advantages such as miniaturization, high sensitivity, and low power consumption, they are widely used in fields with stringent vacuum requirements, including semiconductor manufacturing, vacuum deposition, scientific research, and aerospace. The core sensing component of this vacuum gauge is a MEMS thin film (or MEMS diaphragm), which senses changes in external pressure and undergoes minute deformation, converting this physical signal into a measurable electrical signal to achieve accurate measurement of the vacuum level.

[0004] The long-term stability and reliability of MEMS thin-film vacuum gauges directly depend on the service life of their core MEMS thin films. However, in actual operating conditions, the performance of MEMS thin films gradually degrades due to the coupled influence of multiple factors, eventually leading to failure. These factors include: the fatigue and creep characteristics of the material itself, microscopic defects introduced during manufacturing, and cyclic loading, temperature fluctuations, and corrosion from process gases under complex operating environments. The complexity of the failure mechanism and the coupled nature of multiple factors make accurately predicting the remaining service life of individual MEMS thin films a major challenge for current technologies.

[0005] Currently, the reliability assessment of MEMS thin films mainly relies on traditional experimental methods, such as accelerated life testing (ALT) and static / dynamic performance testing. These methods typically assess the mean time to failure (MTBF) of a batch of products by applying stresses far exceeding normal operating conditions in a laboratory environment. However, these methods have significant limitations: firstly, their test results are difficult to accurately reflect the performance degradation trajectory of individual devices under actual, variable service conditions; secondly, they cannot perform real-time health monitoring and lifespan prediction for equipment in service. Therefore, existing maintenance strategies are mostly passive post-failure repairs or fixed periodic replacements. The former may lead to production interruptions and economic losses, while the latter may result in unnecessary waste.

[0006] In summary, existing technologies have significant gaps in achieving "predictable maintenance" of key sensing components in MEMS thin-film vacuum gauges. Specifically, there is a lack of a technical means to accurately predict the health status and remaining lifespan of MEMS thin films in real time based on actual equipment operating data. This greatly limits the improvement of long-term reliability and operational efficiency of MEMS thin-film vacuum gauges in high-end applications. Therefore, developing an effective MEMS thin-film lifespan prediction model to support the implementation of predictive maintenance strategies has significant technical value and practical implications. Summary of the Invention

[0007] This invention provides a MEMS capacitive thin-film vacuum gauge diaphragm lifetime prediction system based on deep learning, aiming to solve the technical problems mentioned in the background art.

[0008] A deep learning-based MEMS capacitive thin-film vacuum gauge diaphragm lifetime prediction system includes a first mechanical pump, a second mechanical pump, a first molecular pump, a second molecular pump, a first vacuum valve, a second vacuum valve, a first cold cathode ionization vacuum gauge, a second cold cathode ionization vacuum gauge, a first vacuum chamber, a second vacuum chamber, a third vacuum chamber, a first LCR digital bridge, a second LCR digital bridge, a first MEMS diaphragm chamber, a second MEMS diaphragm chamber, a temperature control chamber, a host computer, a controller, and solenoid valves; the molecular pump groups 1 and 2 are respectively connected to the first vacuum valve and the second vacuum valve; the first vacuum chamber... One end of the chamber is connected to the first cold cathode ionization vacuum gauge, and the other end is connected to the first MEMS diaphragm chamber, and also connected to the second vacuum chamber; one end of the second vacuum chamber is connected to the first LCR digital bridge through an airtight connector, and the other end is connected to a solenoid valve; one end of the third vacuum chamber is connected to the second cold cathode ionization vacuum gauge, and the other end is connected to the second MEMS diaphragm chamber, and also connected to the second LCR digital bridge through an airtight connector; the solenoid valve is connected to a controller to control the solenoid valve's switching, and the controller is connected to a host computer; the first LCR digital bridge and the second LCR digital bridge are connected to the host computer to collect the diaphragm's capacitance value.

[0009] Preferably, both the first MEMS diaphragm chamber and the second MEMS diaphragm chamber are located in a temperature control chamber to control the temperature of the MEMS diaphragm during operation.

[0010] Preferably, the first vacuum chamber and the third vacuum chamber are actually KF25 four-way valves, and the second vacuum chamber is actually a KF25 three-way valve.

[0011] Preferably, the solenoid valve is connected to the silencer via a threaded connection.

[0012] Preferably, the implementation method of the system includes the following steps:

[0013] S1. Connect the MEMS in the first MEMS membrane chamber to the first LCR digital bridge, and connect the MEMS in the second MEMS membrane chamber to the second LCR digital bridge. Then, transmit the membrane operation data to the host computer through the serial port to record and store the data.

[0014] S2. Start the temperature control box and maintain the temperature at the preset temperature. Start the first mechanical pump, the second mechanical pump, the first molecular pump and the second molecular pump. Open the first vacuum valve and the second vacuum valve. Start the controller and the host computer. At the same time, let the controller control the solenoid valve to perform the cycle.

[0015] S3. Collect, clean, and preprocess membrane operation data, normalize / standardize it, and use the sliding window method to generate time series samples;

[0016] S4. Local Feature Extraction: Input the time series samples into the CNN network. Through convolution and activation operations, the CNN network extracts representative local patterns and structured features from the original sequence to generate a feature sequence.

[0017] S5. Temporal Dynamic Modeling: The feature sequences generated by CNN are fed into the LSTM network, which then models the temporal dynamics of local patterns, effectively capturing long-term dependencies and global contextual information in the sequence.

[0018] S6. Lifetime prediction and model evaluation: The new membrane is operated normally to obtain data, which is then input into the trained model to obtain the prediction results of the remaining lifetime of the membrane; and the predictive performance of the model is evaluated using an independent test set.

[0019] The beneficial effects achieved by this invention are as follows:

[0020] This invention collects and processes data from MEMS membranes and uses a CNN-LSTM algorithm to model the membranes based on a large amount of data to fit the MEMS membrane lifespan curve. The Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network achieve deep fusion of spatial and temporal information, effectively extracting and integrating spatial features and temporal dependencies from filament image sequences. This allows the algorithm to more comprehensively and accurately understand the complex failure process of MEMS membranes, resulting in more reliable lifespan prediction and better robustness. The model is trained using the CNN-LSTM algorithm, with a damage function R... 2 It can reach 98.79%.

[0021] This invention does not require the development of experimental equipment; all can be achieved using conventional equipment and methods. It offers high accuracy, high efficiency, and convenient and simple operation. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of a MEMS capacitive thin-film vacuum gauge diaphragm lifetime prediction system based on deep learning.

[0023] Figure 2 These are the CNN-LSTM prediction values ​​of this invention.

[0024] Figure 3 This is a diagram of the CNN-LSTM loss function of the present invention.

[0025] Figure reference numerals: 1-First mechanical pump, 2-First molecular pump, 3-First vacuum valve, 4-First cold cathode ionization vacuum gauge, 5-First vacuum chamber, 6-Second vacuum chamber, 7-First LCR digital bridge, 8-First MEMS diaphragm chamber, 9-Temperature control box, 10-Second MEMS diaphragm chamber, 11-Second LCR digital bridge, 12-Third vacuum chamber, 13-Second cold cathode ionization vacuum gauge, 14-Second vacuum valve, 15-Second molecular pump, 16-Second mechanical pump, 17-Host computer, 18-Controller, 19-Solenoid valve, 20-Silencer. Detailed Implementation

[0026] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0027] It should be understood that the terminology used in this invention is merely for describing particular embodiments and is not intended to limit the invention. Furthermore, with respect to numerical ranges in this invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Every smaller range between any stated value or intermediate value within a stated range, and any other stated value or intermediate value within said range, is also included in this invention. The upper and lower limits of these smaller ranges may be independently included or excluded from the range.

[0028] Unless otherwise stated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. While only preferred methods and materials have been described herein, any methods and materials similar or equivalent to those described herein may be used in the implementation or testing of this invention. All references to this specification are incorporated by way of citation to disclose and describe methods and / or materials associated with those references. In the event of any conflict with any incorporated reference, the content of this specification shall prevail.

[0029] Various modifications and variations can be made to the specific embodiments described in this specification without departing from the scope or spirit of the invention, as will be apparent to those skilled in the art. Other embodiments derived from this specification will also be apparent to those skilled in the art. This specification and embodiments are merely exemplary.

[0030] The terms “include,” “including,” “have,” “contain,” etc., used in this article are all open-ended terms, meaning that they include but are not limited to.

[0031] This invention provides a deep learning-based MEMS capacitive thin-film vacuum gauge diaphragm lifetime prediction system, comprising a first mechanical pump 1, a second mechanical pump 16, a first molecular pump 2, a second molecular pump 15, a first vacuum valve 3, a second vacuum valve 14, a first cold cathode ionization vacuum gauge 4, a second cold cathode ionization vacuum gauge 13, a first vacuum chamber 5, a second vacuum chamber 6, a third vacuum chamber 12, a first LCR digital bridge 7, a second LCR digital bridge 11, a first MEMS diaphragm chamber 8, a second MEMS diaphragm chamber 10, a temperature control chamber 9, a host computer 17, a controller 18, and a solenoid valve 19; the first molecular pump 2 and the second molecular pump 15 are respectively connected to the first vacuum valve 3 and the second vacuum valve 14. 4. Connections: One end of the first vacuum chamber 5 is connected to the first cold cathode ionization vacuum gauge 4, and the other end is connected to the first MEMS diaphragm chamber 8, and also connected to the second vacuum chamber 6; one end of the second vacuum chamber 6 is connected to the first LCR digital bridge 7 via a hermetically sealed air connector, and the other end is connected to the solenoid valve 19; one end of the third vacuum chamber 12 is connected to the second cold cathode ionization vacuum gauge 13, and the other end is connected to the second MEMS diaphragm chamber 10, and also connected to the second LCR digital bridge 11 via a hermetically sealed air connector; the solenoid valve 19 is connected to the controller 18 to control the solenoid valve's switch, and the controller 18 is connected to the host computer 17; the first LCR digital bridge 7 and the second LCR digital bridge 11 are connected to the host computer 17 to collect the diaphragm capacitance value. The solenoid valve 19 is used to control the pressure in the test chamber (specifically, the test chamber refers to the first MEMS diaphragm chamber 8 on the left, while the reference chamber refers to the second MEMS diaphragm chamber 10 on the right), so that the pressure in the test chamber can be repeatedly circulated and exposed to the atmosphere. The first LCR digital bridge 7 and the second LCR digital bridge 11 are used in the experiment to measure the capacitance value of the MEMS diaphragm and transmit it to the host computer 17 for recording. The first MEMS diaphragm chamber 8 and the second MEMS diaphragm chamber 10 are used to place the MEMS diaphragm. In order to control the pressure of the test chamber, the first LCR digital bridge 7 and the second LCR digital bridge 11 are connected by an airtight connector to obtain the capacitance value.

[0032] In this embodiment, the first MEMS diaphragm chamber 8 and the second MEMS diaphragm chamber 10 are both in the temperature control box 9 to control the temperature of the MEMS diaphragm during operation.

[0033] In this embodiment, the first vacuum chamber 5 and the third vacuum chamber 12 are actually KF25 four-way valves, and the second vacuum chamber 6 is actually a KF25 three-way valve.

[0034] In this embodiment, the solenoid valve 19 is connected to the silencer 20 via a threaded connection.

[0035] Please see Figure 2 Another objective of this invention is to provide an implementation method for a MEMS capacitive thin-film vacuum gauge diaphragm lifetime prediction system based on deep learning, the method comprising the following steps:

[0036] S1. After connecting the MEMS in the first MEMS diaphragm chamber 8 to the first LCR digital bridge 7 and the MEMS in the second MEMS diaphragm chamber 10 to the second LCR digital bridge 11, the diaphragm operation data (capacitance value, temperature value, pressure value) is transmitted to the host computer 17 through the serial port, so as to record and store the data.

[0037] S2. Start the temperature control box 9 and maintain the temperature at the preset temperature. Start the first mechanical pump 1, the second mechanical pump 16, the first molecular pump 2 and the second molecular pump 15. Open the first vacuum valve 3 and the second vacuum valve 14. Start the controller 18 and the host computer 17. At the same time, let the controller 18 control the solenoid valve 19 to perform the cycle.

[0038] S3. Collect, clean, and preprocess membrane operation data (capacitance, temperature, and pressure values), normalize / standardize them, and generate time series samples using the sliding window method.

[0039] S4. Local Feature Extraction: Input the time series samples into the CNN network. Through convolution and activation operations, the CNN network extracts representative local patterns and structured features from the original sequence to generate a feature sequence.

[0040] S5. Temporal Dynamic Modeling: The feature sequences generated by CNN are fed into the LSTM network, which then models the temporal dynamics of local patterns, effectively capturing long-term dependencies and global contextual information in the sequence.

[0041] S6. Lifetime Prediction and Model Evaluation: This involves taking the new diaphragm under normal operation, obtaining data, inputting it into the trained model, and obtaining the predicted lifetime of the diaphragm; then using an independent test set to evaluate the model's predictive performance.

[0042] Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs) work together in this model:

[0043] Convolutional Neural Networks (CNNs): Primarily responsible for feature extraction. In this application, a one-dimensional CNN acts as a feature extractor, automatically learning and extracting local, representative patterns (such as short-term trends and oscillations) from the raw time-series data. Its advantage lies in its ability to capture reusable key shape features independent of specific time points, thereby transforming the original sequence into a higher-dimensional feature sequence. Long Short-Term Memory Networks (LSTMs): Primarily responsible for temporal modeling. It receives the feature sequences generated by the CNN as input and analyzes the dynamic relationships and long-term dependencies between these local patterns over time. With its unique gating structure, LSTMs can effectively learn the contextual information of a sequence and are highly robust to noise and interference, ultimately used for lifetime prediction tasks.

[0044] The device developed in this invention (referring to...) Figure 1 This invention can effectively collect data on test chamber pressure, test chamber cycle count, temperature, and reference gauge (i.e., full-scale cold gauge) pressure to obtain a real-time dataset of measured sequence parameters of MEMS diaphragms under actual working conditions. Furthermore, the CNN-LSTM algorithm proposed in this invention, by combining the local feature extraction capability of convolutional neural networks and the long-range dependency modeling capability of long short-term memory networks, ensures that both training loss and validation loss converge after multiple rounds of training, thus accurately extracting diaphragm degradation patterns and remaining lifespan. To evaluate the predictive performance of CNN-LSTM, this invention calculates the following key indicators: Coefficient of Determination (R²), which measures the goodness of fit of the model (a value closer to 1 is better); Mean Absolute Error (MAE), which reflects the average absolute magnitude of the prediction error (a value smaller is better); Mean Bias Error (MBE), which judges the systematic bias of the model (overestimation or underestimation) (a value closer to 0 is better); Mean Squared Error (MSE), which measures the mean of the squares of the prediction error (more sensitive to large errors, a value smaller is better); Mean Absolute Percentage Error (MAPE), which reflects the relative magnitude of the prediction error (a value smaller is better). (Table 1) These metrics collectively constitute a comprehensive evaluation system for the model's performance on the training and test sets. To comprehensively evaluate the performance of the CNN-LSTM model, this invention not only calculates quantitative metrics such as R² and MAE, but also analyzes the model through visual charts. For example... Figure 2 As shown, on both the training and test sets (a) and (b) (training set data), the model's predicted values ​​(blue line) closely match the true values ​​(red line), intuitively demonstrating the model's excellent fitting and generalization prediction capabilities. Figure 3 The training loss of the model converges rapidly to a low level within 20 iterations, indicating that the model training process is stable and effective.

[0045] Table 1

[0046]

[0047] It should be noted that, in this document, the term "comprising" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0048] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A deep learning-based MEMS capacitance thin film vacuum gauge diaphragm lifetime prediction system, characterized in that, The system comprises a first mechanical pump (1), a second mechanical pump (16), a first molecular pump (2), a second molecular pump (15), a first vacuum valve (3), a second vacuum valve (14), a first cold cathode ionization vacuum gauge (4), a second cold cathode ionization vacuum gauge (13), a first vacuum chamber (5), a second vacuum chamber (6), a third vacuum chamber (12), a first LCR digital bridge (7), a second LCR digital bridge (11), a first MEMS diaphragm chamber (8), a second MEMS diaphragm chamber (10), a temperature control box (9), a host computer (17), a controller (18) and a solenoid valve (19). The first molecular pump (2) and the second molecular pump (15) are connected with the first vacuum valve (3) and the second vacuum valve (14) respectively; one end of the first vacuum chamber (5) is connected with the first cold cathode ionization vacuum gauge (4), the other end is connected with the first MEMS diaphragm chamber (8), and is also connected with the second vacuum chamber (6); one end of the second vacuum chamber (6) is connected with the first LCR digital bridge (7) through a gas-tight plug, the other end is connected with the solenoid valve (19); one end of the third vacuum chamber (12) is connected with the second cold cathode ionization vacuum gauge (13), the other end is connected with the second MEMS diaphragm chamber (10), and is also connected with the second LCR digital bridge (11) through a gas-tight plug; the solenoid valve (19) is connected with the controller (18) to control the opening and closing of the solenoid valve, and the controller (18) is connected with the host computer (17); the first LCR digital bridge (7) and the second LCR digital bridge (11) are connected with the host computer (17) to collect the capacitance values of the diaphragms.

2. The deep learning-based MEMS capacitance diaphragm vacuum gauge diaphragm lifetime prediction system of claim 1, wherein, The first MEMS diaphragm chamber (8) and the second MEMS diaphragm chamber (10) are in the temperature control box (9).

3. The deep learning-based MEMS capacitance thin film vacuum gauge membrane lifetime prediction system of claim 1, wherein, The first vacuum chamber (5) and the third vacuum chamber (12) are actually KF25 four-ways, and the second vacuum chamber (6) is actually a KF25 three-way.

4. The deep learning-based MEMS capacitance thin film vacuum gauge membrane lifetime prediction system of claim 1, wherein, The solenoid valve (19) is connected with a muffler (20) through threads.

5. The deep learning-based MEMS capacitance thin film vacuum gauge membrane lifetime prediction system of claim 1, wherein, The implementation method of the system comprises the following steps: S1, after connecting the MEMS in the first MEMS diaphragm chamber (8) with the first LCR digital bridge (7) and connecting the MEMS in the second MEMS diaphragm chamber (10) with the second LCR digital bridge (11), transmitting the diaphragm operation data to the host computer (17) through a serial port to record and store the data; S2, starting the temperature control box (9) to maintain the temperature at a preset temperature, starting the first mechanical pump (1), the second mechanical pump (16), the first molecular pump (2) and the second molecular pump (15), opening the first vacuum valve (3) and the second vacuum valve (14), starting the controller (18) and the host computer (17), and simultaneously controlling the solenoid valve (19) by the controller (18) to perform a cycle; S3, collecting, cleaning and pretreating the diaphragm operation data, performing normalization / standardization, and generating time series samples by using a sliding window method. S4, local feature extraction: input the time series sample into the CNN network, and through convolution and activation operation, the CNN network extracts representative local patterns and structured features from the original sequence to generate a feature sequence; S5, time series dynamic modeling: the feature sequence generated by the CNN is sent to the LSTM network, and the LSTM further models the time dynamics of the local pattern, effectively capturing the long-term dependence and global context information in the sequence; S6, life prediction and model evaluation, that is, the newly obtained data during normal operation of the membrane is input into the trained model to obtain the prediction result of the remaining life of the membrane; and the prediction performance of the model is evaluated using an independent test set.

Images