General molecular pump life prediction method based on neural network and related equipment
By finely disassembling the molecular pump components and using a Transformer neural network to build a lifetime prediction model, the problems of scarce molecular pump data and insufficient memory capacity of traditional models are solved, achieving high-confidence fault warning and lifetime prediction.
Patent Information
- Application Number
- CN202511041276.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-28
- Publication Date
- 2025-11-07
AI Technical Summary
Existing technologies cannot provide high-confidence fault warnings and remaining life predictions for molecular pumps under conditions of limited data and single operating conditions. Furthermore, traditional neural network models have limited memory capacity for ultra-long sequences, resulting in delayed or drifting prediction results.
By classifying the molecular pump components through fine-grained decomposition, multiple local fault periodic data are constructed. A lifetime prediction model is built using a Transformer neural network to capture the coupling relationship between vibration, temperature, and current ripple, thereby realizing the construction of large-scale, multi-condition datasets and lifetime prediction.
It achieves high-confidence molecular pump failure early warning and remaining lifetime prediction, alleviates the problem of data scarcity, avoids the gradient vanishing and computational explosion problems of traditional models, and improves prediction accuracy.
Smart Images
Figure CN120911279A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of molecular pump life prediction, in particular to a universal molecular pump life prediction method based on a neural network and related equipment. BACKGROUND
[0002] A molecular pump (Molecular Pump) is a kind of dry vacuum pump that uses high-speed rotating blades or rotor-stator micro-gap to form molecular-level pumping channel and relies on molecular momentum transfer to achieve ultimate vacuum. Compared with conventional volumetric pumps, diffusion pumps or turbo molecular pumps, the "high-speed rotation + micro-gap" of the molecular pump makes its fault tolerance to the environment and operation failure extremely low, while the volumetric or vapor flow design of other pumps allows more relaxed working conditions. Therefore, compared with the life prediction of conventional volumetric pumps, diffusion pumps or turbo molecular pumps, the life prediction of molecular pumps requires higher accuracy.
[0003] The fault and life prediction method is common in modern intelligent equipment, which can effectively reduce the cost and enhance the reliability, but the current life prediction method for molecular pumps has limited prediction ability, and the reasons are that it is difficult to extract strong correlation indicators, the data set is not enough, the working condition of the molecular pump is single, and the mathematical model is selected.
[0004] Among them, the cost of establishing the condition of the whole life cycle of the molecular pump and large-scale monitoring is high, which limits the ability to obtain large-scale data sets, and for the users of the molecular pump, the lack of large-scale long-time data will lead to insufficient comprehensive extraction of key indicators of molecular pump life and failure; the existing public data set is insufficient in size and incomplete in label, and it is difficult to support deep learning model training. At the same time, the fault evolution of the molecular pump often shows early weak vibration, temperature drift, current ripple and other multi-source heterogeneous characteristics, and these characteristics change dynamically with vacuum degree, speed and load. Traditional feature engineering relies on expert experience and is difficult to automatically capture cross-scale and cross-modal potential correlations. The life degradation period of the molecular pump can be up to thousands of hours, and there is a long interval between early weak abnormal signals and final failure. The existing convolutional neural network (CNN) or traditional recurrent neural network (RNN) has limited memory capacity for long sequences, which is prone to gradient disappearance or calculation explosion, resulting in lag or drift of the prediction result. SUMMARY
[0005] Based on the problems raised in the above background technology, the purpose of the present application is to provide a universal molecular pump life prediction method based on a neural network and related equipment, which solves the problem that the prior art cannot provide high-confidence fault warning and remaining life prediction for 63-aperture small molecular pumps under data-limited and single-working-condition conditions.
[0006] The application is achieved by the following technical solutions: The application provides a universal molecular pump life prediction method based on a neural network, comprising the following steps: Classify the molecular pump fault types according to the component assemblies of the molecular pump; Perform correlation analysis on the molecular pump full life cycle operating state and the molecular pump fault types to obtain a molecular pump full life cycle time sequence; Construct a life prediction model based on a Transformer neural network, input the molecular pump full life cycle time sequence into the life prediction model for life prediction, and obtain a life consumption percentage.
[0007] In the above technical solution, the molecular pump is classified according to the component assemblies, the molecular pump is converted into multiple local fault periodic data through a fine-grained splitting method, the high-correlation data is screened out as strong correlation indicators through correlation analysis, and the molecular pump full life cycle time sequence is formed, so that a large-scale, multi-condition and weakly-labeled data set is constructed at a very low cost, the training and prediction of the subsequent life prediction model are supported, and the problem of molecular pump full life cycle data scarcity is significantly alleviated.
[0008] The life prediction model based on the Transformer neural network can capture the coupling relationship among the vibration high-frequency component, the temperature low-frequency drift and the current ripple, avoids the limitation of the traditional CNN / RNN that requires manual design of the convolution kernel or the gating structure, inputs the molecular pump full life cycle time sequence into the life prediction model for training, and uses the trained life prediction model for life prediction, so that the life consumption percentage can be obtained, the problem that the existing convolutional neural network (CNN) or the traditional recurrent neural network (RNN) has limited memory capacity for an ultra-long sequence and is prone to gradient vanishing or computational explosion, and the problem that the prediction result is lagging or drifting are solved, and the life prediction of the molecular pump is realized.
[0009] In an alternative embodiment, the component assemblies of the molecular pump include a main shaft impeller, a driving motor, a supporting system, a lubricating system, a peripheral seal and a controller; The classification of the molecular pump fault types according to the component assemblies of the molecular pump comprises: Traverse the main shaft impeller, the driving motor, the supporting system, the lubricating system, the peripheral seal and the controller, and perform failure mode and effects analysis to obtain the molecular pump fault type classification.
[0010] In an alternative embodiment, the correlation analysis on the molecular pump full life cycle operating state and the molecular pump fault types comprises the following steps: extract normal equipment operation data from the molecular pump full life cycle operation state; calculate the correlation of the normal operation data and the molecular pump fault type, and screen out strong correlation indicators according to the results of the correlation; construct a molecular pump full life cycle time sequence by using the strong correlation indicators.
[0011] In an optional embodiment, calculating the correlation of the normal operation data and the molecular pump fault type comprises:
[0012] In the above formula, is the correlation, is the normal operation data of the i th sample, is the molecular pump fault type corresponding to the i th sample, is the mean value of all normal operation data samples, is the mean value of all fault type samples. In an optional embodiment, the molecular pump full life cycle time sequence is constructed as follows:
[0013] In the above formula,
[0014] is a time sequence matrix, is the sample category, is a time node, is a molecular pump different life stage point.
[0015] In an optional embodiment, inputting the molecular pump full cycle life time sequence into the life prediction model for life prediction comprises: dividing the molecular pump full cycle life time sequence into a training set and a validation set; labeling the training set, the label including: current fault state degree; wherein the current fault state degree is divided into 20 labels from the initial state to the failure state; selecting 10 consecutive time node molecular pump full life cycle data from the labeled training set and inputting the data into the life prediction model for training, and verifying the trained life prediction model by using the validation set; using the verified life prediction model to perform life prediction.
[0016] In an optional embodiment, the life prediction model further comprises: using cross-entropy loss as a loss function, wherein the loss function is as follows:
[0017] In the above formula, is a cross-entropy loss, is a label of an i-th sample, is a predicted probability of the i-th sample.
[0018] The second aspect of the present application provides a neural network-based universal molecular pump life prediction system, comprising: a fault classification module configured to classify molecular pump fault types according to constituent components of the molecular pump; a time sequence module configured to analyze the correlation between the molecular pump full life cycle operating state and the molecular pump fault type, and obtain a molecular pump full life cycle time sequence; a prediction module configured to construct a life prediction model based on a Transformer neural network, input the molecular pump full cycle life time sequence into the life prediction model for life prediction, and obtain a life consumption percentage.
[0019] The third aspect of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the neural network-based universal molecular pump life prediction method.
[0020] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the neural network-based universal molecular pump life prediction method.
[0021] Compared with the prior art, the present application has the following advantages and beneficial effects: 1. The molecular pump is classified according to constituent components, and the molecular pump is converted into multiple local fault periodic data through a fine-grained splitting method. Then, the correlation analysis is used to screen out high correlation as strong correlation indicators, and the molecular pump full life cycle time sequence is formed, so that a large-scale, multi-condition, and weakly-labeled data set is constructed at a very low cost, thereby supporting the training and prediction of the subsequent life prediction model, and significantly alleviating the problem of scarcity of molecular pump full life cycle data; 2. The life prediction model is constructed based on the Transformer neural network, and the trained life prediction model is used for life prediction, so that the life consumption percentage can be obtained, thereby solving the problem that the existing convolutional neural network (CNN) or traditional recurrent neural network (RNN) has limited memory capacity for long sequences, which is prone to gradient disappearance or calculation explosion, resulting in lag or drift of the prediction result, and realizing the life prediction of the molecular pump. BRIEF DESCRIPTION OF DRAWINGS
[0022] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be considered as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor. In the drawings: Figure 1 A flowchart of a neural network-based universal molecular pump life prediction method provided for Embodiment 1 of the present application; Figure 2 A schematic diagram of failure mode and effects analysis provided for Embodiment 1 of the present application; Figure 3 An example diagram of a life prediction model provided for Embodiment 1 of the present application; Figure 4 A structural schematic diagram of an electronic device provided for Embodiment 3 of the present application. DETAILED DESCRIPTION
[0023] In order to make the objects, technical solutions and advantages of the present application clearer, the following will further describe the present application in detail with reference to the embodiments and drawings, the illustrative embodiments of the present application and the description thereof are only used to explain the present application, and should not be considered as a limitation on the present application.
[0024] Embodiment 1 of the present application provides a neural network-based universal molecular pump life prediction method, as shown in Figure 1 The neural network-based universal molecular pump life prediction method comprises the following steps: Classifying the molecular pump failure types according to the constituent components of the molecular pump; Performing correlation analysis on the molecular pump full life cycle operating state and the molecular pump failure types to obtain a molecular pump full life cycle time sequence; Constructing a life prediction model based on a Transformer neural network, inputting the molecular pump full cycle life time sequence into the life prediction model for life prediction to obtain a life consumption percentage.
[0025] It should be noted that the molecular pump is classified according to the constituent components, the molecular pump is converted into multiple local fault periodic data through a fine-grained splitting method, and then the correlation analysis is used to screen out high correlation as strong correlation indicators, and the molecular pump full life cycle time sequence is formed, so that a large-scale, multi-condition, weakly-labeled data set is constructed at a very low cost, to support the training and prediction of the subsequent life prediction model, and significantly alleviate the problem of scarcity of molecular pump full life cycle data.
[0026] The life prediction model is constructed based on a Transformer neural network, which can capture the coupling relationship between high-frequency components of vibration, low-frequency drift of temperature and current ripple, and avoid the limitation of traditional CNN / RNN which needs to manually design convolution kernel or gating structure. The time sequence of the full life cycle of the molecular pump is input into the life prediction model for training, and the trained life prediction model is used for life prediction, so as to obtain the life consumption percentage, thereby solving the problem that the existing convolutional neural network (CNN) or traditional recurrent neural network (RNN) has limited memory capacity for super-long sequence, and is prone to gradient vanishing or explosion, resulting in lagging or drifting of the prediction result, and realizing the life prediction of the molecular pump.
[0027] In an alternative embodiment, the constituent components of the molecular pump include a main shaft impeller, a drive motor, a support system, a lubrication system, a peripheral seal and a controller; The molecular pump fault type classification according to the constituent components of the molecular pump includes: The main shaft impeller, the drive motor, the support system, the lubrication system, the peripheral seal and the controller are traversed, and a failure mode and effect analysis is performed to obtain the molecular pump fault type classification.
[0028] It should be noted that according to the structure and performance of the molecular pump, the constituent components can be divided into six parts, including a main shaft impeller, a drive motor, a support system, a lubrication system, a peripheral seal and a controller. In this embodiment, the six parts are traversed and a failure mode and effect analysis is performed to obtain all the fault types of the molecular pump, wherein the failure mode and effect analysis is described in Figure 2 which reflects the fault types and causes of all parts of the molecular pump, for example, the function of the rotor is to provide kinetic energy of gas molecules, and the failure mode is support wear and failure. The six parts are traversed and analyzed in turn to obtain the molecular pump fault type classification: lubrication system degradation and failure, support wear and failure, vacuum seal degradation and failure, motor degradation and failure, and controller failure.
[0029] Further, based on the failure mode and effect analysis of Figure 2 , nine collectable indexes related to the molecular pump fault and life condition can be obtained, including current, voltage, noise during steady operation, maximum vibration frequency of the molecular pump, historical operation time of the molecular pump, vacuum degree, operation temperature, current maximum speed of the molecular pump and acceleration to steady speed time (acceleration time). The selection of the nine indexes is obtained through correlation analysis.
[0030] In an alternative embodiment, the correlation analysis of the full life cycle operation state of the molecular pump and the molecular pump fault type includes the following steps: Normal equipment operation data is extracted from the full life cycle operation state of the molecular pump; correlation between the normal operation data and the molecular pump fault type, and screening a strong correlation index according to a result of the correlation; constructing a molecular pump full life cycle time sequence by using the strong correlation index.
[0031] In an optional embodiment, the correlation between the normal operation data and the molecular pump fault type is calculated, including:
[0032] In the above formula, is the correlation, is the normal operation data of the i-th sample, is the molecular pump fault type corresponding to the i-th sample, is the mean of all normal operation data samples, is the mean of all fault type samples. In this embodiment, the Pearson correlation coefficient is used to calculate the correlation between the normal operation data and the molecular pump fault type, to obtain the correlation of the two types of data, and the correlation calculation result is taken as a strong correlation object.
[0033] The strong correlation object parameters are collected by the molecular pump controller internal sensors, decibel meters, vacuum gauges, vibration test benches, and non-contact tachometers, to form a molecular pump life cycle time sequence.
[0034] In an optional embodiment, the molecular pump full life cycle time sequence is constructed as follows:
[0035]
[0036] In the above formula, is a time sequence matrix, and is a sample category, is a time node, is a molecular pump different life stage point It should be noted that the time sequence matrix includes current time sequence, voltage time sequence, molecular pump vibration frequency time sequence, vacuum degree time sequence, running temperature time sequence, running noise time sequence, molecular pump running time sequence, molecular pump current maximum speed time sequence, and speed-up time sequence types. The full life cycle of the molecular pump can be represented by the time sequence matrix.
[0037] In an optional embodiment, the molecular pump full cycle life time sequence is input into the life prediction model for life prediction, including: dividing the molecular pump full cycle life time sequence into a training set and a validation set; labeling the training set, the label including: current failure state degree; wherein the current failure state degree is divided into 20 labels from an initial state to a failure state; The molecular pump full life cycle data of 10 continuous time nodes are selected from the labeled training set and input into the life prediction model for training, and the trained life prediction model is verified by using the verification set; The life prediction model is used for life prediction.
[0038] It should be noted that the molecular pump full cycle life time series is subjected to data cleaning, abnormal processing, data normalization and other preprocessing operations, and after preprocessing, it is divided into a training set and a verification set The data in the training set are labeled, wherein the label records the state of each component and the degree of the current failure state, and the degree of the current failure state is divided into 20 stages and 20 degrees from an initial state to a failure state, specifically 0%, 5%, 10%, 15%, 20%... 95%, 100%. Each label collects at least 2000 data, and the data is taken from multiple molecular pumps.
[0039] In the construction of the life prediction model based on the Transformer neural network, the molecular pump full life cycle time series is subjected to position encoding to obtain:
[0040] As shown in Figure 3 The molecular pump full life cycle data of 10 continuous time nodes are selected from the labeled training set and input into the life prediction model for training, wherein the feature value matrix of the time node is input into the encoder of the life prediction model for encoding, and the labeled feature value matrix is input into the decoder for decoding. The output of the final decoder is the percentage of life consumption; finally, the trained life prediction model is verified by using the verification set.
[0041] Further, the life prediction model can be used for life prediction and fault prediction. The fault wear degree feature vector in the molecular pump full life cycle data is input into the life prediction model for training, and various key faults of the molecular pump, such as lubrication system degradation and failure, bearing wear and failure, vacuum seal degradation and failure, motor degradation and failure, and controller failure, can be obtained.
[0042] Further, in this embodiment, the life prediction model constructs 6 different parallel models according to different training targets, which are used to output the life remaining degree prediction of the molecular pump and various key faults of the molecular pump.
[0043] In an optional embodiment, the lifetime prediction model further includes: using cross-entropy loss as a loss function, wherein the loss function is as follows:
[0044] In the above formula, For cross-entropy loss, For the first The label of each sample For the first The predicted probability of a sample.
[0045] in, .
[0046] In this embodiment, on a production line using 63-caliber molecular pumps, the current data status of each molecular pump on the production line is first saved, and the time series of the status of the molecular pumps is saved for 1 hour. The prediction model of the present invention is used to predict the failure and lifespan of each molecular pump, which can predict the remaining status and failure status of each molecular pump.
[0047] This method allows for monitoring the lifespan and operating conditions of molecular pumps. If a pump is about to fail or reach the end of its lifespan, it can prevent production stoppages caused by pump malfunctions, thus saving significant funds on maintenance, testing, and labor costs. Compared to manual warnings or shutdowns for repairs, this method greatly assists in production line status management and cost control.
[0048] Embodiment 2 of the present invention provides a universal molecular pump lifetime prediction system based on neural networks, comprising: The fault classification module is used to classify the fault types of the molecular pump based on its constituent components. The time series module is used to perform correlation analysis on the operating status of the molecular pump throughout its entire life cycle and the failure type of the molecular pump to obtain the time series of the molecular pump's entire life cycle. The prediction module is used to construct a lifetime prediction model based on the Transformer neural network. The full-cycle life time series of the molecular pump is input into the lifetime prediction model to predict the lifetime and obtain the percentage of lifetime consumption.
[0049] Embodiment 3 of the present invention provides an electronic device, such as... Figure 4 As shown, the electronic device includes a processor 21, a memory 22, an input device 23, and an output device 24; the number of processors 21 in the computer device can be one or more. Figure 4 Taking a processor 21 as an example; the processor 21, memory 22, input device 23, and output device 24 in an electronic device can be connected via a bus or other means. Figure 4For example, the connection by bus.
[0050] The memory 22 can be used to store software programs, computer executable programs and modules as a computer readable storage medium. The processor 21 executes various functions and data processing of the electronic device by running the software programs, instructions and modules stored in the memory 22, that is, implements the neural network-based universal molecular pump life prediction method of embodiment 1.
[0051] The memory 22 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application program required by a function; the data storage area can store data created according to the use of the terminal, etc. In addition, the memory 22 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state memory device. In some examples, the memory 22 can further include a memory remotely arranged with respect to the processor 21, which can be connected to the electronic device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0052] The input device 23 can be used to receive the id and password input by the user, etc. The output device 24 is used to output the network configuration page.
[0053] The embodiment 4 of the present application also provides a computer readable storage medium, the computer executable instructions of which, when executed by a computer processor, are used to implement the neural network-based universal molecular pump life prediction method provided in embodiment 1.
[0054] The storage medium provided by the embodiment of the present application contains computer executable instructions, which are not limited to the method operations provided in embodiment 1, but can also perform related operations in the neural network-based universal molecular pump life prediction method provided by any embodiment of the present application.
[0055] The above specific embodiments further illustrate the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A neural network-based universal molecular pump lifetime prediction method, characterized by, Comprising the following steps: According to the composition components of the molecular pump, the molecular pump fault type is classified; Correlation analysis is performed on the molecular pump full life cycle operation state and the molecular pump fault type, and a molecular pump full life cycle time sequence is obtained; A life prediction model is constructed based on a Transformer neural network, and the molecular pump full cycle life time sequence is input into the life prediction model for life prediction to obtain a life consumption percentage. 2.The neural network-based universalized molecular pump lifetime prediction method of claim 1, wherein, The composition components of the molecular pump include a main shaft impeller, a driving motor, a support system, a lubrication system, a peripheral seal, and a controller; According to the composition components of the molecular pump, the molecular pump fault type is classified, which comprises: Traverse the main shaft impeller, the driving motor, the support system, the lubrication system, the peripheral seal, and the controller, and perform failure mode and effects analysis to obtain a molecular pump fault type classification. 3.The neural network-based universal molecular pump life prediction method according to claim 1, wherein, Correlation analysis is performed on the molecular pump full life cycle operation state and the molecular pump fault type, which comprises the following steps: Normal equipment operation data is extracted from the molecular pump full life cycle operation state; The correlation between the normal operation data and the molecular pump fault type is calculated, and strong correlation indicators are selected according to the correlation results; The strong correlation indicators are used to construct a molecular pump full life cycle time sequence. 4.The neural network-based universal molecular pump life prediction method according to claim 3, wherein, The correlation between the normal operation data and the molecular pump fault type is calculated, which comprises: In the above formula, is the correlation, is the normal operation data of the first sample, is the molecular pump fault type corresponding to the first sample, is the mean of all normal operation data samples, is the mean of all fault type samples. 5.The neural network-based universal molecular pump life prediction method according to claim 3, wherein, The molecular pump full life cycle time sequence is constructed as follows: In the above formula, is a time series matrix, is a sample category, is a time node, is a different life stage point of the molecular pump. 6.The neural network-based universal molecular pump life prediction method according to claim 1, wherein, The molecular pump full cycle life time sequence is input into the life prediction model for life prediction, which comprises: The molecular pump full cycle life time sequence is divided into a training set and a validation set; The training set is labeled, and the label includes: current fault state degree; wherein the current fault state degree is divided into 20 labels from the initial state to the failure state; Ten consecutive time node molecular pump full life cycle data are selected from the labeled training set and input into the life prediction model for training, and the validation set is used to verify the trained life prediction model; The life prediction model is used for life prediction. 7.The neural network-based universal molecular pump life prediction method of claim 1, wherein, The life prediction model also includes: using cross-entropy loss as a loss function, wherein the loss function is as follows: In the above formula, is the cross-entropy loss, is the label of the th sample, is the predicted probability of the th sample.
8. A neural network-based universal molecular pump life prediction system, characterized by, Comprising: A fault classification module for classifying the molecular pump fault type according to the composition components of the molecular pump; A time sequence module for correlation analysis of the molecular pump full life cycle operation state and the molecular pump fault type, and obtaining a molecular pump full life cycle time sequence; A prediction module for constructing a life prediction model based on a Transformer neural network, inputting the molecular pump full cycle life time sequence into the life prediction model for life prediction, and obtaining a life consumption percentage.
9. An electronic device, comprising: A computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the neural network-based universal molecular pump life prediction method of any one of claims 1-7.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the neural network-based universal molecular pump life prediction method according to any one of claims 1 to 7.