Method and system for monitoring health state of ultra-clean bellows pump cavity

Abstract

The application provides a super-clean corrugated pipe pump cavity health state monitoring method and system, comprising the following steps: S1: a marking step, a magnetic film is pasted at a preset position of the corrugated pipe as a magnetic film marking point; S2: a magnetic flux density value acquisition step, when the corrugated pipe is in a compressed state, a magnetic flux density signal generated by excitation of the magnetic film is synchronously collected through a magnetic sensor, and the magnetic flux density signal is processed to obtain a magnetic flux density value; S3: a strain prediction value acquisition step, the magnetic flux density value is taken as input, and a trained LSTM neural network prediction model is processed to output a corresponding strain prediction value; and S4: a health state judgment step, the strain prediction value is compared with a preset strain threshold value to judge the current health state of the corrugated pipe. Compared with the prior art, the application has the advantages of non-contact, high-precision monitoring and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of integrated circuit manufacturing equipment state monitoring, in particular to a super-clean bellows pump cavity health state monitoring method and system. BACKGROUND

[0002] In semiconductor integrated circuit manufacturing, PTFE bellows pump cavities used for conveying ultra-pure chemical fluids are prone to fatigue rupture due to long-term bearing of fluid-solid coupling cyclic load, and once leakage occurs, it will pollute the process environment and cause production interruption, so monitoring and early warning of the health state thereof is of great importance to guarantee production safety and reliability.

[0003] The existing monitoring methods for pump cavity state mainly include contact sensor measurement and non-contact analysis. The former directly or indirectly obtains state signals by installing strain gauges or acceleration sensors, but the sensor itself will change the structural characteristics and it is difficult to meet the ultra-clean environment requirements of semiconductor manufacturing; the latter such as finite element simulation is limited by model assumptions and has insufficient precision, and optical non-contact measurement may introduce pollution risk due to the equipment itself. Both methods have obvious limitations and cannot meet the early, accurate and clean diagnostic warning needs under complex working conditions.

[0004] In summary, the contact measurement method in the prior art will interfere with the structure and pollute the environment, and is not suitable for ultra-clean scenes; the simulation method of non-contact type has poor real-time performance, and the optical method is restricted by the cleanliness of the environment. At present, there is still a lack of an effective technical means that can realize high-precision monitoring of the dynamic strain and health state of the bellows without interfering with the work of the bellows and destroying the ultra-clean environment. This technical gap makes the state perception and reliability management of key components of the semiconductor ultra-pure fluid conveying system face great challenges. SUMMARY

[0005] The present application is carried out to solve the above problems, and aims to provide a super-clean bellows pump cavity health state monitoring method and system.

[0006] The present application provides a super-clean bellows pump cavity health state monitoring method, which has the following characteristics: S1: a marking step, a magnetic film is pasted at a predetermined position of the bellows as a magnetic film marking point; S2: a magnetic flux density value acquisition step, when the bellows is in a compressed state, a magnetic flux density signal generated by excitation of the magnetic film is synchronously collected by a magnetic sensor, and the magnetic flux density signal is processed to obtain a magnetic flux density value; S3: a strain prediction value acquisition step, the magnetic flux density value is taken as an input, and a trained LSTM neural network prediction model is used for processing to output a corresponding strain prediction value; S4: a health state judgment step, the strain prediction value is compared with a preset strain threshold value to judge the current health state of the bellows.

[0007] In the method for monitoring the health state of the ultra-clean bellows pump cavity provided by the application, the preset position in S1 can be the position of the bellows peak, and S1 specifically comprises the following sub-steps: S1-1: cleaning the outer surface of the bellows peak position; and S1-2: pasting the magnetic film on the bellows peak position, so that the magnetic film is tightly attached to the bellows peak, to ensure that the deformation behavior of the magnetic film is completely synchronized with the bellows body.

[0008] In the method for monitoring the health state of the ultra-clean bellows pump cavity provided by the application, the method for obtaining the magnetic flux density value in S2 can comprise the following steps: S2-1: determining the relative distance between the magnetic sensor and the magnetic film, to ensure that the magnetic sensor can effectively collect the magnetic flux density signal generated by the magnetic film excitation; S2-2: compressing the bellows, and synchronously collecting the magnetic flux density signal generated by the magnetic film excitation when the bellows is in the compressed state by the magnetic sensor, and transmitting the magnetic flux density signal to the processing unit; and S2-3: processing the magnetic flux density signal by the processing unit to filter out the environmental background magnetic field interference, to obtain the magnetic flux density value, wherein the processing unit comprises a single-chip microcomputer and a computer, the single-chip microcomputer is connected with the magnetic sensor and the computer, the single-chip microcomputer is used for receiving the magnetic flux density signal and transmitting the same to the computer, and the computer processes the magnetic flux density signal.

[0009] In the method for monitoring the health state of the ultra-clean bellows pump cavity provided by the application, the environmental background magnetic field interference can comprise the geomagnetic field.

[0010] In the method for monitoring the health state of the ultra-clean bellows pump cavity provided by the application, the input layer node number of the LSTM neural network prediction model in S3 corresponds to the number of the magnetic sensors, the output layer node number corresponds to the number of the magnetic film marker points, and the number of the hidden layers and the node number of each layer in the middle are selected and optimized according to the training effect of the LSTM neural network prediction.

[0011] In the method for monitoring the health state of the ultra-clean bellows pump cavity provided by the application, the training method of the LSTM neural network prediction model in S3 can comprise the following steps: S3-1: constructing a training data set comprising the corresponding relationship between the magnetic flux density value and the actual strain value of the bellows in different compressed states, and the training data set is obtained by simulation or experimental measurement; S3-2: training the LSTM neural network prediction model based on the training data set, to establish the mapping relationship from the magnetic flux density value to the strain value; and S3-3: calculating the relative error and the correlation coefficient between the strain prediction value output by the LSTM neural network prediction model and the actual strain value, to quantitatively evaluate the performance of the LSTM neural network prediction model.

[0012] In the super-clean bellows pump cavity health state monitoring method provided by the application, the training method of the LSTM neural network prediction model in S3 can further include the following steps: S3-4: The generalization ability of the LSTM neural network prediction model is verified by using a cross-dataset verification method, that is, an external dataset independent of the training dataset is input into the LSTM neural network prediction model, the relative error and the correlation coefficient between the output strain prediction value of the LSTM neural network prediction model and the actual strain value of the external dataset are calculated, and the performance of the LSTM neural network prediction model is quantitatively evaluated.

[0013] In the super-clean bellows pump cavity health state monitoring method provided by the application, the training dataset can be obtained by simulation, and Gaussian noise is added to the simulation data during the training process to simulate the uncertainty and interference in actual measurement.

[0014] In the super-clean bellows pump cavity health state monitoring method provided by the application, the judgment logic of the current health state of the bellows in S4 includes: state judgment: comparing the strain prediction value with the preset strain threshold value, if the strain prediction value is less than the preset strain threshold value, it is determined that the bellows is in a healthy state; if the strain prediction value is greater than or equal to the preset strain threshold value, it is determined that the bellows is in a near-failure state; risk assessment: according to the difference between the strain prediction value and the preset strain threshold value, the health risk level of the bellows is evaluated, the smaller the difference, the higher the health risk level, and the closer to the failure critical point.

[0015] The application further provides a super-clean bellows pump cavity health state monitoring system, which has the following characteristics: a marking module, which pastes a magnetic film at a preset position of the bellows as a magnetic film marking point; a magnetic flux density value acquisition module, which synchronously acquires the magnetic flux density signal generated by the magnetic film excitation through a magnetic sensor when the bellows is in a compressed state, and processes the magnetic flux density signal to obtain the magnetic flux density value; a strain prediction value acquisition module, which takes the magnetic flux density value as input, processes it through the trained LSTM neural network prediction model, and outputs the corresponding strain prediction value; and a health state judgment module, which compares the strain prediction value with a preset strain threshold value to judge the current health state of the bellows.

[0016] Compared with the prior art, the application has the following advantages:

[0017] The super-clean bellows pump cavity health state monitoring method and system according to the application have the following beneficial effects:

[0018] 1. Ultra-clean compatibility: The core monitoring device, the magnetic sensor, is located outside the pump cavity of the bellows. No electrical devices need to be installed on the bellows body, fundamentally eliminating the risk of pollution caused by the introduction of sensors, and perfectly meeting the requirements of ultra-clean environments.

[0019] 2. High-precision monitoring: The LSTM neural network prediction model is used to analyze the complex nonlinear relationship between the magnetic signal and the strain, enabling accurate inversion of the strain state and low monitoring result hysteresis.

[0020] 3. Strong early warning capability: Through continuous monitoring, the accumulation trend of plastic strain in the bellows during fatigue can be captured, and the performance degradation of the bellows can be identified before physical rupture occurs, enabling predictive maintenance.

[0021] 4. Good robustness: The magnetic signal is not easily disturbed by airflow, weak vibration, or light changes in the clean room, and has better anti-interference ability than optical methods, ensuring stable and reliable monitoring data in complex industrial environments and adapting to complex industrial site environments.

[0022] 5. Easy to integrate and deploy: The overall structure is compact, no modification of the existing pump body structure is required, and the installation of magnetic film marker points and sensor arrays is simple, allowing convenient integration into semiconductor fluid delivery systems.

[0023] 6. Provides quantitative risk rating: The invention compares the strain prediction value with the preset strain threshold value, determines the health status of the bellows according to the comparison result, and can evaluate the degree of approaching the breaking point based on the difference between the strain prediction value and the preset strain threshold value, thereby realizing the diagnosis and risk warning of the health status of the bellows structure, and providing direct and objective data support for maintenance decisions.

[0024] The invention provides an innovative and reliable technical means to solve the health monitoring problem of key components of semiconductor ultra-clean equipment, which is of great significance to improve the overall safety and reliability of the integrated circuit manufacturing process. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of the health status monitoring method of the ultra-clean bellows pump cavity in the embodiment of the invention.

[0026] Figure 2 is an architectural schematic diagram of the health status monitoring method of the ultra-clean bellows pump cavity in the embodiment of the invention.

[0027] Figure 3 is a layout schematic diagram of the magnetic film marker points and the magnetic sensor array of the bellows in the embodiment of the invention.

[0028] MAIN COMPONENT SYMBOL DESCRIPTION:

[0029] In the figure: 101, first magnetic film mark point; 102, second magnetic film mark point; 103, third magnetic film mark point; 104, fourth magnetic film mark point; 105, fifth magnetic film mark point; 106, sixth magnetic film mark point; 107, corrugated pipe; 2, magnetic sensor array; 201, first magnetic sensor; 202, second magnetic sensor; 203, third magnetic sensor; 204, fourth magnetic sensor; 205, fifth magnetic sensor; 206, sixth magnetic sensor; 3, single-chip microcomputer; 4, computer; 5, LSTM neural network prediction model. DETAILED DESCRIPTION

[0030] In order to make the technical means, creative features, purposes and effects realized by the present application easy to understand, the following embodiments will be described in detail in combination with the drawings.

[0031] The embodiment provides a super-clean corrugated pipe pump cavity health state monitoring method, and specifically comprises the following steps:

[0032] Figure 1 It is a flowchart of the super-clean corrugated pipe pump cavity health state monitoring method in the embodiment of the present application. Figure 2 It is an architecture schematic diagram of the super-clean corrugated pipe pump cavity health state monitoring method in the embodiment of the present application.

[0033] As shown in Figures 1-2 , step S1 is a marking step, and the magnetic film is pasted at a preset position of the corrugated pipe 107 as a magnetic film mark point, and specifically comprises:

[0034] S1-1: Before pasting the magnetic film, the outer surface of the wave crest position of the corrugated pipe 107 needs to be thoroughly cleaned, and the magnetic film pasting position is marked;

[0035] Figure 3 It is a layout schematic diagram of the magnetic film mark point of the corrugated pipe and the magnetic sensor array in the embodiment of the present application.

[0036] S1-2: as shown in Figure 3 , the magnetic film is pasted at the wave crest position (i.e. the magnetic film pasting position) of the corrugated pipe 107, and is respectively marked as the first magnetic film mark point 101, the second magnetic film mark point 102, the third magnetic film mark point 103, the fourth magnetic film mark point 104, the fifth magnetic film mark point 105 and the sixth magnetic film mark point 106, so that all the magnetic films are tightly attached to the wave crests of the corrugated pipe, to ensure that the deformation behavior of the magnetic film is completely synchronized with the corrugated pipe body.

[0037] Step S2 is a magnetic flux density value acquisition step. When the bellows 107 is in a compressed state, the magnetic flux density signals generated by the magnetic thin film excitation are synchronously collected by the magnetic sensor, and the magnetic flux density signals are processed to obtain the magnetic flux density value. Specifically, the following steps are performed:

[0038] S2-1: Determine the relative distance between the magnetic sensor and the magnetic thin film to ensure that the magnetic sensor can effectively collect the magnetic flux density signals generated by the magnetic thin film excitation. Specifically, the following steps are performed:

[0039] As shown in FIG. 1, the outer side of the bellows 107 with magnetic film marker points is provided with a magnetic sensor array 2 composed of a first magnetic sensor 201, a second magnetic sensor 202, a third magnetic sensor 203, a fourth magnetic sensor 204, a fifth magnetic sensor 205, and a sixth magnetic sensor 206. In this embodiment, the positions of all magnetic sensors correspond to the positions of all magnetic film marker points one by one, and the spatial relative distance between the magnetic sensor array 2 and the magnetic thin film is determined, so that the spatial magnetic field generated by the magnetic thin film excitation can be effectively detected. Figure 3

[0040] S2-2: Compress the bellows 107, and synchronously collect the magnetic flux density signals generated by the magnetic thin film excitation when the bellows 107 is in a compressed state by the magnetic sensor, and transmit the magnetic flux density signals to the processing unit. Specifically, the following steps are performed:

[0041] A linear motor is used to axially compress the bellows 107 provided with magnetic film marker points. Then, the magnetic flux density signals generated by the magnetic thin film excitation when the bellows 107 is in a compressed state are synchronously collected by the magnetic sensor array 2, and the magnetic flux density signals are transmitted to the processing unit.

[0042] S2-3: The processing unit processes the magnetic flux density signals to filter out the environmental background magnetic field interference to obtain the magnetic flux density value. Specifically, the following steps are performed:

[0043] The processing unit includes a single-chip microcomputer 3 and a computer 4, and the single-chip microcomputer 3 is connected with the magnetic sensor array 2 and the computer 4. The single-chip microcomputer 3 is used to receive the magnetic flux density signals and transmit them to the computer 4, and the computer 4 processes the magnetic flux density signals to filter out the environmental background magnetic field interference components including the geomagnetic field.

[0044] Step S3 is a strain prediction value acquisition step. The magnetic flux density value is taken as input, processed by the trained LSTM neural network prediction model 5, and the corresponding strain prediction value is output. Specifically, the following steps are performed:

[0045] ​The LSTM neural network prediction model 5 is built by a computer, with the input being the magnetic flux density value and the output being the corresponding strain prediction value. In this embodiment, the input of the LSTM neural network prediction model 5 is the three-axis magnetic flux density value collected by each magnetic sensor in the magnetic sensor array 2, and the output is the strain prediction value of each magnetic film marker point (i.e. the position of the wave peak).

[0046] The number of input layer nodes corresponds to the number of magnetic sensors; the number of output layer nodes corresponds to the number of magnetic film marker points; the number of hidden layers and the number of nodes in each layer are selected and optimized according to the training effect of the LSTM neural network prediction model 5.

[0047] Before applying the LSTM neural network prediction model 5, the LSTM neural network prediction model 5 needs to be trained, and the training method includes the following steps:

[0048] S3-1: A training data set containing the corresponding relationship between the magnetic flux density value and the actual strain value of the bellows under different compression states is constructed, and the training data set is obtained by simulation or experimental measurement.

[0049] S3-2: Based on the training data set, the LSTM neural network prediction model 5 is trained to establish a mapping relationship from the magnetic flux density value to the strain value.

[0050] S3-3: Calculate the relative error and correlation coefficient between the strain prediction value output by the LSTM neural network prediction model 5 and the actual strain value, and quantitatively evaluate the performance of the LSTM neural network prediction model 5.

[0051] In this embodiment, when measuring the strain of a bellows with a known fatigue degree, the corresponding magnetic flux density value is synchronously collected, thereby forming a complete training data set. 70% of the samples in this data set are divided into a training set for training the LSTM neural network prediction model 5; the remaining 30% of the samples are used as a validation set to evaluate the prediction performance of the LSTM neural network prediction model 5. Finally, the relative error and correlation coefficient between the strain prediction value output by the LSTM neural network prediction model 5 and the true measured value are calculated to quantitatively evaluate the prediction accuracy and consistency of the LSTM neural network prediction model 5.

[0052] S3-4: Further, to verify the generalization ability of the LSTM neural network prediction model 5, multiple groups of measured data independent of the training set and the validation set are used to perform cross-dataset verification on the LSTM neural network prediction model 5, specifically:

[0053] The relative error and correlation coefficient between the output strain prediction value of the LSTM neural network prediction model 5 and the actual strain value of the external measured data set are calculated using an external measured data set independent of the training data set, and the performance of the LSTM neural network prediction model 5 is quantitatively evaluated.

[0054] The verification process takes the relative error and correlation coefficient of the prediction result as the core evaluation index, aiming to ensure that the LSTM neural network prediction model 5 not only has good prediction accuracy, but also has excellent generalization ability.

[0055] In addition, if the training data set is completely obtained by simulation, Gaussian noise needs to be added to the simulation data during the training process to simulate the uncertainty and interference in actual measurement, and then the robustness of the LSTM neural network prediction model 5 in a noisy environment is trained and verified.

[0056] Step S4 is a health state judgment step, which compares the strain prediction value with the preset strain threshold to judge the current health state of the bellows 107, specifically:

[0057] The preset strain threshold needs to be determined in combination with a specific bellows. The strain at the peak position of the bellows is not the same due to the difference in structure and size.

[0058] In this embodiment, to determine the strain threshold at the peak position of the bellows 107 of the researched specification, the actual measured data of the strain corresponding to the same type of bellows that failed in the field application of semiconductor equipment is referred to. Based on the actual failure data, the preset strain threshold for health state evaluation is set as a quantitative basis for judging whether the bellows 107 is approaching fatigue failure and whether it needs to be warned or replaced, so as to convert the strain prediction result of the LSTM neural network prediction model 5 into a health state index that can directly guide the maintenance decision.

[0059] The specific judgment logic is as follows:

[0060] State judgment: compare the strain prediction value with the preset strain threshold, if the strain prediction value is less than the preset strain threshold, it is determined that the bellows 107 is in a healthy state; if the strain prediction value is greater than or equal to the preset strain threshold, it is determined that the bellows 107 is in a near failure state.

[0061] Risk assessment: considering that the fatigue accumulation of the bellows 107 during the cyclic compression process will cause the strain to gradually increase, further according to the difference between the strain prediction value and the preset strain threshold, the relative distance from the damage critical point is evaluated to evaluate the health risk level of the bellows 107. The smaller the difference, the higher the health risk level of the bellows 107, the closer to the failure critical point, and the shorter the remaining service life; on the contrary, it indicates that the structural integrity of the bellows 107 remains good and can continue to be safely served. This evaluation can provide a basis for preventive maintenance and replacement decisions.

[0062] The embodiment also provides a super-clean bellows pump cavity health state monitoring system, comprising:

[0063] The marking module is used to implement S1, that is, the magnetic film is pasted at a preset position of the bellows 107 as a magnetic film marking point.

[0064] The magnetic flux density value acquisition module is used to implement S2, that is, when the bellows 107 is in a compressed state, the magnetic flux density signal generated by the magnetic film excitation is synchronously collected by the magnetic sensor, and the magnetic flux density signal is processed to obtain the magnetic flux density value.

[0065] The strain prediction value acquisition module is used to implement S3, that is, the magnetic flux density value is taken as an input, and the trained LSTM neural network prediction model 5 is used for processing to output the corresponding strain prediction value.

[0066] The health state judgment module is used to implement S4, that is, the strain prediction value is compared with the preset strain threshold to judge the current health state of the bellows 107.

[0067] Effects of the embodiment

[0068] The super-clean bellows pump cavity health state monitoring method and system according to the application have the following beneficial effects:

[0069] The application realizes non-contact and high-precision measurement of the dynamic strain of the bellows 107 by combining magnetic film marking with external magnetic field detection, and further realizes the evaluation and early warning of the health state by combining intelligent analysis, thereby providing effective technical support for guaranteeing the reliability and safety of the semiconductor manufacturing process, including:

[0070] 1. High super-clean compatibility: the core monitoring device magnetic sensor is located outside the pump cavity of the bellows 107, without the need to install any electrical device on the bellows 107 body, which fundamentally eliminates the pollution risk caused by the introduction of the sensor, and perfectly meets the requirements of the super-clean environment.

[0071] 2. High-precision monitoring: The LSTM neural network prediction model 5 can analyze the complex nonlinear relationship between the magnetic signal and the strain, accurately invert the strain state, and reduce the monitoring result lag.

[0072] 3. Strong early warning capability: Through continuous monitoring, the accumulation trend of plastic strain in the corrugated pipe 107 during fatigue can be captured, and the performance degradation of the corrugated pipe 107 can be identified before physical rupture occurs, realizing predictive maintenance.

[0073] 4. Good robustness: The magnetic signal is not easily disturbed by airflow, weak vibration or light changes in the clean room, and has better anti-interference ability than optical methods, ensuring stable and reliable monitoring data in complex industrial environments and adapting to complex industrial environments.

[0074] 5. Easy to integrate and deploy: The overall structure is compact, and the existing pump body structure does not need to be modified. The installation of the magnetic film marker point and the magnetic sensor array 2 is simple, and it can be conveniently integrated into a semiconductor fluid conveying system.

[0075] 6. Provides quantitative risk rating: The application compares the strain prediction value with the preset strain threshold value, determines the health status of the corrugated pipe 107 according to the comparison result, and can evaluate the degree of approaching the damage critical point based on the difference between the strain prediction value and the preset strain threshold value, thereby realizing the diagnosis and risk warning of the health status of the corrugated pipe 107, and providing direct and objective data support for maintenance decisions.

[0076] In summary, the application provides an innovative and reliable technical means to solve the health monitoring problem of key components of semiconductor ultra-clean equipment, which is of great significance to improve the overall safety and reliability of the integrated circuit manufacturing process.

[0077] Those skilled in the art should understand that the application is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the application. Without departing from the spirit and scope of the application, various changes and improvements can be made to the application, and these changes and improvements all fall within the scope of the claimed application. The scope of protection of the application is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the health status of an ultra-clean corrugated pipe pump cavity, characterized in that, Includes the following steps: S1: Marking step, attach the magnetic film to the preset position of the bellows as magnetic film marking point; S2: Magnetic flux density value acquisition step: When the bellows is in a compressed state, the magnetic flux density signal generated by the excitation of the magnetic thin film is synchronously acquired by the magnetic sensor, and the magnetic flux density signal is processed to obtain the magnetic flux density value. S3: Strain prediction value acquisition step, taking the magnetic flux density value as input, processing it through the trained LSTM neural network prediction model, and outputting the corresponding strain prediction value. S4: Health status judgment step, compare the predicted strain value with the preset strain threshold to determine the current health status of the bellows.

2. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 1, characterized in that: in, The preset position mentioned in S1 is the position of the corrugated pipe's crest. S1 specifically includes the following sub-steps: S1-1: Clean the outer surface of the corrugated pipe at the crest position; S1-2: Adhere the magnetic film to the crest of the bellows, ensuring that the magnetic film fits tightly against the bellows crest to guarantee that the deformation behavior of the magnetic film is completely synchronized with the bellows body.

3. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 1, characterized in that: in, The method for obtaining the magnetic flux density value in S2 includes the following steps: S2-1: Determine the relative distance between the magnetic sensor and the magnetic thin film to ensure that the magnetic sensor can effectively acquire the magnetic flux density signal generated by the excitation of the magnetic thin film; S2-2: The bellows is compressed, and the magnetic sensor synchronously collects the magnetic flux density signal generated by the magnetic thin film excitation when the bellows is in a compressed state, and transmits the magnetic flux density signal to the processing unit. S2-3: The processing unit processes the magnetic flux density signal to filter out ambient background magnetic field interference and obtain the magnetic flux density value. The processing unit includes a microcontroller and a computer. The microcontroller is connected to the magnetic sensor and the computer. The microcontroller is used to receive the magnetic flux density signal and transmit it to the computer. The computer processes the magnetic flux density signal.

4. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 3, characterized in that: in, The environmental background magnetic field interference includes geomagnetism.

5. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 1, characterized in that: in, In S3, the number of nodes in the input layer of the LSTM neural network prediction model corresponds to the number of magnetic sensors; the number of nodes in the output layer corresponds to the number of magnetic film markers; and the number of hidden layers and nodes in each layer are selected and optimized based on the training effect of the LSTM neural network prediction.

6. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 1, characterized in that: in, The training method for the LSTM neural network prediction model in S3 includes the following steps: S3-1: Construct a training dataset containing the correspondence between magnetic flux density values ​​and actual strain values ​​of bellows under different compression states. The training dataset is obtained through simulation or experimental measurement. S3-2: Based on the training dataset, train the LSTM neural network prediction model to establish a mapping relationship from magnetic flux density values ​​to strain values; S3-3: Calculate the relative error and correlation coefficient between the strain prediction value output by the LSTM neural network prediction model and the actual strain value to quantitatively evaluate the performance of the LSTM neural network prediction model.

7. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 6, characterized in that: in, The training method for the LSTM neural network prediction model in S3 further includes the following steps: S3-4: The generalization ability of the LSTM neural network prediction model is verified by using a cross-dataset verification method. The cross-dataset verification method is as follows: an external dataset independent of the training dataset is input into the LSTM neural network prediction model, and the relative error and correlation coefficient between the output strain prediction value and the actual strain value of the LSTM neural network prediction model on the external dataset are calculated to quantitatively evaluate the performance of the LSTM neural network prediction model.

8. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 6, characterized in that: in, The training dataset is obtained through the simulation method. Gaussian noise is added to the simulation data during the training process to simulate the uncertainties and interference in actual measurements.

9. The method for monitoring the health status of an ultra-clean corrugated pipe pump cavity according to claim 8, Its features are: The logic for determining the current health status of the bellows in S4 includes: Status judgment: The strain prediction value is compared with the preset strain threshold. If the strain prediction value is less than the preset strain threshold, the bellows is determined to be in a healthy state; if the strain prediction value is greater than or equal to the preset strain threshold, the bellows is determined to be in a near-damage state. Risk assessment: The health risk level of the bellows is assessed based on the difference between the predicted strain value and the preset strain threshold. The smaller the difference, the higher the health risk level and the closer it is to the failure threshold.

10. A health status monitoring system for ultra-clean corrugated pipe pump chambers, characterized in that, include: The marking module attaches a magnetic film to a preset position on the bellows as a magnetic film marking point; The magnetic flux density acquisition module acquires the magnetic flux density signal generated by the excitation of the magnetic thin film through a magnetic sensor when the bellows is in a compressed state, and processes the magnetic flux density signal to obtain the magnetic flux density value. The strain prediction value acquisition module takes the magnetic flux density value as input, processes it through the trained LSTM neural network prediction model, and outputs the corresponding strain prediction value. The health status judgment module compares the predicted strain value with a preset strain threshold to determine the current health status of the bellows.

Images