Predictive maintenance method for NEG vacuum pump

Abstract

A performance monitoring method for predictive maintenance of an NEG pump in a vacuum system comprising at least one pressure sensor operably connected to a computer that calculates an estimated instantaneous operating parameter curve from pressure sensor readings, and comparing the evolution of the curve with a reference set of curves stored in a computer to obtain a value RUL of the pump remaining useful life.

Description

[0001] Due to the inherent advantages of getter technology applied to vacuum pumps, non-evaporable getter (NEG) pumps are currently widely used in getter pumping systems; in fact, NEG pumps are compact, have low energy consumption requirements, and do not cause vibration.

[0002] Given the above, the demand for reliable NEG pumps in vacuum systems is constantly increasing, thus requiring even greater optimization of pump performance and maintenance than ever before.

[0003] One way to achieve this is to monitor some pump parameters (such as pumping speed, remaining capacity, estimated reactivation time) and check whether the operating conditions meet the desired processing. US 2023 / 088465 discloses a method for regenerating a reactivatable NEG pump connected to a vacuum system, in which regeneration is controlled by determining the pressure in the vacuum system and by recording the current through an ion getter pump combined with the NEG pump.

[0004] EP 3751147 discloses a computer-implemented method for determining the operability of mechanical equipment. In particular, the method describes how to use machine learning algorithms to examine whether compressor damage jeopardizes the overall functionality of the system.

[0005] EP 1839151 describes a diagnostic analysis method for failure protection of general-purpose vacuum pumps. This method estimates the best-fit model parameters based on the minimum and maximum values ​​of the state variables of the evaluation system.

[0006] None of the aforementioned prior art documents address the problem of instantaneously estimating the remaining life of the system's main moving component—the getter pump—and therefore cannot provide users or customers with immediate information for forward-looking use and planned maintenance (such as NEG pump reactivation).

[0007] Monitoring of NEG pumps can be achieved using a digital controller according to the invention, connected to a vacuum system including at least one NEG pump, the controller being integrated into a computer for a user interface. When the controller successfully completes data processing, the computer screen displays a set of desired NEG pump parameters as output, such as pumping speed, remaining capacity, and estimated reactivation time. By increasing the networking capability of the digital controller, information regarding the use of the NEG pumps can be shared with pump suppliers to further refine the code behind the controller's logic and deepen understanding of the application systems using the NEG pumps, and more generally, a deeper understanding of vacuum products. A single digital controller can be used to control, collect, and process data from multiple NEG pumps connected to it. While not limited to a specific number of NEG pumps, a minimum configuration is 1:1 (one controller to one NEG pump).

[0008] In a preferred solution according to the invention, the controller is configured to execute the computer-implemented method as described in the following paragraphs via a supervised learning algorithm.

[0009] This invention provides a performance monitoring method for predictive maintenance of NEG pumps in a vacuum system, the vacuum system including at least one pressure sensor. In a preferred embodiment of the invention, the sensor is a Bayard-Alpert vacuum gauge, but other pressure sensors can be readily used, such as a separate extractor gauge, a cold cathode sensor, and a sensor for the current of an ion pump when an ion pump is also installed in the vacuum system.

[0010] The present invention is described below with reference to non-limiting examples and by means of the accompanying drawings, in which:

[0011] Figure 1 The diagram illustrates the logic tree representing the pump activation phase: the schematic shows the steps and conditions to be implemented before the pump monitoring phase is started.

[0012] Figure 2 This represents the linear evolution of the residual parameter (RP) defined in the following equation [5] from 1 to 0 during the life of the NEG pump; the remaining service life (RUL) of the pump is determined by the total failure time t. 失效 The difference between the current time and the timeline is given (the timeline should be understood as being expressed in any unit).

[0013] Figure 3 A flowchart of a performance monitoring method for predictive maintenance of NEG pumps in a vacuum system according to the present invention is shown; in particular, the flowchart illustrates method steps that need to be repeated cyclically.

[0014] Figure 4 The predicted RUL versus actual RUL of the NEG high vacuum pump is shown. This trend is based on the results of n tests performed on a single stack of NEG pumps under gas compositions of 20% CO, 20% O2 and 60% H2 in a vacuum system. (The time axis should be understood as being expressed in arbitrary units.)

[0015] Figure 5 This is a schematic diagram of a computer-readable storage device including a computer program according to the present invention, the computer-readable storage device being connected to a vacuum system comprising: a NEG high vacuum pump "NEG" including electronic components and a pressure gauge G1 including electronic components. These components are respectively connected via NC1 and GC1 to be operatively connected to electronic hardware (computer) having a user interface. Figure 5 The external interface is also shown.

[0016] Before initiating the monitoring phase via a supervised learning algorithm, the system checks for NEG pump activation; the logic tree for this function is in Figure 1 The diagram shows the following steps:

[0017] a) Set the pump model and operating parameters, such as: pump model specifications, operating voltage / current values, thermocouple readings (if present), and ramp time t to reach activation temperature. 上升 NEG maintains the activation temperature for a duration of t 保持 ;

[0018] b) Manually initiate pump activation or set pump activation at a specific time, such that when pump activation begins, the clock is activated, and once t... 时钟 >=t 上升 +t 保持 Once activated, the process ends.

[0019] c) Confirm successful activation and send a success signal S to the system, while updating the activation counter;

[0020] d) Wait for NEG to be ready before starting data collection. Readiness is determined by the following:

[0021] d1) After cooling time t 冷却 ,or

[0022] d2) Temperature equal to or below the threshold temperature T 阈值 .

[0023] The monitoring method includes repeatedly performing the following steps:

[0024] Store pressure readings from at least one pressure sensor;

[0025] Estimate and store operation parameters ( ) curve, the The curve from instantaneous Value obtained, at that instant The value is calculated based on the pressure sensor reading using the following equation [4];

[0026] Will The evolution of the curve up to the current instant and the n lines defined in the following equation [3] The curves are compared against a reference set, where each reference curve has an associated value for the pump's remaining useful life (RUL).

[0027] Select at least two A reference subset of the curve, wherein... The subset of curves includes those that are closest to the current value in a metric sense (meaning the distance calculated between the values ​​of the operating parameters). The curve of a curve;

[0028] By considering The median of the RUL of the curve reference subset is used to calculate the time required before reactivation.

[0029] To improve the predictive ability of this method, it is preferable to calculate the instantaneous time after reactivation of the reactivatable NEG pump. The curve is added to the reference set as an additional reference curve.

[0030] The training cycle is managed by a supervised learning algorithm, specifically a linear regression algorithm, which can typically be defined as follows: [1]

[0031] Where m indicates the number of sensors from which input signals are obtained, X is a vector (i.e., a set of values ​​of a specific quantity) of pressure values ​​(P1, P2, ..., Pn) obtained in dimensionless form, i.e., as pure numbers, and Y is a vector of RP values ​​calculated according to the following equation [5]. Linear regression makes it possible to calculate coefficients by minimizing the root mean square error (RMSE), which is used as a comprehensive metric for accuracy to measure model performance. The estimated value and error The estimated value RMSE is calculated as follows: [2]

[0032] Where N_train is the number of data points in the training dataset, and These are the predicted values ​​of the operating parameters, which can be calculated as follows: [3]

[0033] To prevent the model from performing poorly on new data, the available dataset is divided into complementary subsets: a training dataset and a validation dataset. The model is trained on a combination of some subsets and validated on the remaining subsets. This technique is called "cross-validation," and it is used to make the algorithm more accurate in predictions.

[0034] The monitoring cycle is managed by a supervised learning algorithm, specifically a similarity-based model based on instantaneous pressure readings P. t To transfer instantaneous operating parameters The prediction is: [4]

[0035] The residual parameters (RP) of the NEG pump are determined by RUL (i.e., total failure time t). 失效 The difference between the current time and the total failure time t 失效 The ratio between (i.e., the maximum operating time before the pump needs to be reactivated) and is given, such as Figure 2 The value represented therein, and its value is within the range including 0 and 1, as described in the equation: [5]

[0036] The remaining parameters are calculated for each pressure curve included in the training dataset. Such a training dataset for each pressure curve may include additional labels such as gas composition, pump reactivation count, total absorbed gas, auxiliary pumping rate, and NEG pump model (based on geometry).

[0037] exist Figure 3 The monitoring method described above not only has the advantage of instantaneously predicting RUL throughout the entire monitoring period, but also provides users with additional information (e.g., gas composition, remaining adsorption capacity, remaining pumping rate) by associating appropriate labels with the pressure curves included in the training dataset.

[0038] Predicting the trend of RUL versus actual RUL in Figure 4 The figure shows the time intervals during which the two values ​​are closest and therefore the predictions are most accurate. This figure is derived from experimental tests performed on a single-stack NEG pump with a gas composition of 20% CO, 20% O2, and 60% H2 in a vacuum system. Variations in the prediction curves are due to instantaneous... The curve is compared with the 5 closest curves in the training dataset, and the instantaneously predicted RUL is obtained from the median of the RUL of a subset of such curves.

[0039] It will be readily understood by anyone skilled in the art that, in order to reduce prediction error, linear regression fitting can be performed as in [1], but instead of fitting the entire training dataset, it can be fitted to C subsets of the dataset obtained using clustering techniques. This method will produce a set of estimated parameters. (j=1,…,C), these parameters will be used to estimate during the monitoring phase. Values. A suitable subset of parameters is selected based on clustering techniques.

[0040] In addition, other supervised learning algorithms can be used in the training and monitoring phases, such as logistic regression and support vector machines.

[0041] According to another aspect of the invention, a computer-implemented method is described. This method includes a machine learning algorithm with two consecutive operating phases: a training phase and a monitoring phase. The training phase uses a training dataset and a validation dataset as input to determine the operating parameters of a NEG pump in a high vacuum system. The dataset includes the evolution of pressure within the vacuum system over time, the number of NEG pump reactivations, the pumping rates for different gases, the evolution of the amount of absorbed gas over time, and the gas composition within the vacuum system. The dataset also considers the number of auxiliary pumps present in the system, the pumping rates of the auxiliary pumps, and the type of NEG pump.

[0042] The trained algorithm learns from the above data how to correlate the pressure evolution curve with the NEG pump status and its RUL. The monitoring phase aims to adjust the pressure curve based on the above operating parameters. To estimate this state in real time and the time required to reactivate the NEG pump.

[0043] On the other hand, the digital controller according to the present invention is capable of detecting and reporting system anomalies, such as:

[0044] NEG activation was unsuccessful (e.g., threshold power was not delivered for the entire duration).

[0045] The main body of NEG has reached saturation (the total capacity has been utilized based on the number of reactivations and the estimated gas absorption).

[0046] estimated The curve monotonically evolves to the training dataset Below the curve set indicates that the pumping performance degrades exceptionally rapidly, for example, due to leaks in the vacuum system.

[0047] Another aspect of the present invention is a computer-readable storage device comprising a computer program that performs an evolution monitoring and diagnostic analysis method for predictive maintenance of NEG high vacuum pumps. Figure 5 As indicated, the vacuum system is connected to electronic hardware with a user interface and external interfaces to display information and outputs from the monitoring method.

[0048] Electronic devices for acquisition and processing receive system signals as input (e.g., pressure signals from sensors). These signals can be analog or digital, or transmitted via a serial network or local area network (LAN). The electronic devices process the signals using mathematical methods or programmable logic according to the method of the present invention described above. The electronic devices may include memory, an ALU (arithmetic logic unit), a microprocessor, a PC, converters, interfaces, and power supplies for the pump power supply and / or its controller. The NEG pump (and its electronic power supply) installed on the system can be connected to the processing electronics via cable or wireless network.

[0049] The system can be equipped with a local or remote user interface that allows users to more easily monitor and manage the vacuum system. This electronic configuration, or one of its subsystems, can thus be considered part of a more general electronic system controlling other components of the vacuum system. During monitoring, experimental data can be collected, which can then be used to feed training datasets to this device, similar devices in the application, or systems with different applications. This is particularly significant because it enables communication between multiple systems and allows for the collection and exchange of information about the vacuum pump's status and performance, thus establishing a direct two-way communication channel with the customer.

Claims

1. A method for predicting the time required before the reactivation of a reactivatable NEG pump, the reactivatable NEG pump being connected to a vacuum system including at least one pressure sensor operatively connected to a computer, the method comprising cyclically repeating the following steps: a) Storing multiple pressure readings from the at least one pressure sensor in the computer. b) Estimate and store operating parameters in the computer The curve, the The curve starts from the instantaneous value calculated by the computer based on the pressure sensor readings. Value obtained, c) The The evolution of the curve up to the current instant and the n lines stored in the computer The reference sets of the curves are compared, where, Each of these reference curves has an associated value RUL for the pump's remaining service life. d) Select at least two A reference subset of the curve, wherein the reference subset includes curves that are closest in a metrical sense to the current curve. The curve of the curve, e) by the computer considering the The median of the RUL of the curve reference subset is used to calculate the time required before reactivation. f) Output a set of desired NEG pump parameters generated by the calculations via at least one local or remote user interface.

2. The method according to claim 1, wherein, The method has a preliminary stage of checking pump activation.

3. The method according to claim 2, wherein, The phase of checking pump activation includes the following steps: a) Set the pump model and its technical specifications, such as the pump model name, operating voltage / current, thermocouple reading, and ramp time t to reach activation temperature. 上升 The NEG is maintained at the activation temperature for a duration t. 保持 Operating parameters; b) Manually initiate the activation of the pump or activate the pump at a set specific time, such that when the pump activation begins, the clock is activated, and once t 时钟 >=t 上升 +t 保持 The activation process ends there. c) Confirm the activation is successful and send a success signal S to the system, while updating the activation counter; d) Wait for NEG to be ready before starting data collection. Readiness is determined by the following: d1) After cooling time t 冷却 ,or d2) Temperature equal to or below the threshold temperature T 阈值 .

4. The method according to any one of the preceding claims, wherein, The residual parameter RP is given by the ratio between the RUL of the NEG pump and the total failure time, and the residual parameter RP is in the range of 1 to 0.

5. The method according to any one of the preceding claims, wherein, The at least one pressure sensor is selected from: Bayard-Alpert sensors, discrete ionization vacuum gauges, cold cathode sensors, and sensors for the current of the ion pump when the vacuum system is also equipped with an ion pump.

6. The method according to any one of the preceding claims, wherein, The instantaneous time calculated after the reactivation of the NEG pump Curve added to the The reference set for the curve.

7. A computer-implemented method, comprising: A machine learning algorithm, preferably a linear regression algorithm, using a training dataset as input, is used to determine the operating parameters of an NEG pump in a high vacuum system, characterized in that: a) The training dataset includes the pressure evolution within the vacuum system over time, the number of reactivations of the NEG pump, the number and speed of the auxiliary pumps, the gas composition within the vacuum system, and the type of the NEG pump. b) The method provides the state of the NEG pump and an estimated time when the NEG pump needs to be reactivated as output from the machine learning algorithm.

8. The method according to claim 7, wherein, The operating parameters include one or more parameters selected from adsorption capacity and pumping speed.

9. The method according to claim 7 or 8, wherein, The output is given by determining, from a set of pre-labeled pressure distribution curves included in the training dataset, a pressure distribution curve that better represents the current state of the NEG pump to predict the maintenance of the NEG pump.

10. A computer-readable storage device, comprising: A computer program that executes an evolution monitoring and diagnostic analysis method according to any one of the preceding claims for failure protection and predictive maintenance of NEG high vacuum pumps.

11. A digital controller configured to perform the computer-implemented method of any one of claims 7 to 9.

12. The digital controller of claim 11, wherein the digital controller is capable of detecting and reporting anomalies, for example: The activation of NEG was unsuccessful; The main body of the NEG is saturated; The estimated The curve lies in the training dataset. The facts below the curve set.

Images