Vacuum pump life prediction device and program

Abstract

To provide a device for predicting the life of a vacuum pump which can easily predict the life of a vacuum pump.SOLUTION: The present invention includes: a data acquisition unit 11 for acquiring data detected by a vibration acceleration sensor for detecting vibrations during the operation of a vacuum pump 10B; a shift amount calculation unit 13 for calculating a shift amount (a change amount) from the reference value of the unique vibration frequency of the vacuum pump 10B on the basis of the data acquired by the acquisition unit 11; and an inference unit 16 for inferring the life of the vacuum pump 10B on the basis of the amount of change calculated by the shift amount calculation unit 13.SELECTED DRAWING: Figure 1

Description

Technical Field

[0006]

[0001] The present invention relates to a vacuum pump life prediction device and a program.

Background Art

[0002] Since a vacuum pump deteriorates during use, replacement or overhaul is required. The period until this replacement is necessary, or the period until overhaul is necessary, is called "life". However, an operator cannot grasp this life in advance.

[0003] Therefore, Patent Document 1 below discloses a device for predicting the life of a vacuum pump. In the device described in this Patent Document 1, reference index values for predicting the life of a vacuum pump are determined in advance for each process condition, and the life of the vacuum pump is predicted based on the difference between the index value of the life obtained from the state quantity of the vacuum pump and the reference index value corresponding to the process condition.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In Patent Document 1, reference index values corresponding to each of innumerable process conditions must be determined in advance. In addition, it is also necessary to consider that the state quantity of the vacuum pump changes due to the accumulation of deposits. Thus, there is a problem that it is not easy to predict the life of the vacuum pump in the device described in Patent Document 1.

[0006] Therefore, an object of the present invention is to provide a vacuum pump life prediction device and a program that can easily predict the life of a vacuum pump. [Means for solving the problem]

[0007] To solve the above problems, a vacuum pump life prediction device according to a first aspect of the present invention comprises: an acquisition unit that acquires data detected by a vibration acceleration sensor that detects vibrations during the operation of the vacuum pump; a calculation unit that calculates the amount of change from a reference value of the natural vibration frequency of the vacuum pump based on the data acquired by the acquisition unit; and an inference unit that infers the life of the vacuum pump based on the amount of change calculated by the calculation unit.

[0008] Furthermore, in a second aspect of the present invention, a learning unit is provided that performs learning to predict the lifespan based on the amount of change from a reference value of the natural vibration frequency of the learning vacuum pump and state information of the learning vacuum pump, and the inference unit infers the lifespan based on the learning result by the learning unit and the amount of change calculated by the calculation unit.

[0009] Furthermore, in a third aspect of the present invention, the system further includes a reception unit that receives the state information from a user, the learning unit performs the learning each time the reception unit receives the state information, and the inference unit infers the lifespan each time the learning is performed by the learning unit, based on the learning result by the learning unit and the amount of change calculated by the calculation unit.

[0010] Furthermore, in a fourth aspect of the present invention, an output unit is provided that outputs the lifespan inferred by the inference unit.

[0011] Furthermore, the program according to the fifth aspect of the present invention causes a computer capable of communicating with a vibration acceleration sensor that detects vibrations during the operation of a vacuum pump to function as an acquisition unit that acquires data detected by the vibration acceleration sensor, a calculation unit that calculates the amount of change from a reference value of the natural vibration frequency of the vacuum pump based on the data acquired by the acquisition unit, and an inference unit that infers the lifespan of the vacuum pump based on the amount of change calculated by the calculation unit. [Effects of the Invention]

[0012] According to the vacuum pump life prediction device and program of the present invention, the lifespan of a vacuum pump can be easily predicted. [Brief explanation of the drawing]

[0013] [Figure 1] This figure schematically shows an example of the functional configuration of a vacuum pump life prediction device according to the first embodiment of the present invention. [Figure 2] Figure 1 is a schematic block diagram showing an example of the hardware configuration of the vacuum pump life prediction device. [Figure 3] Figure 1 is a graph showing an example of the results of FFT processing performed by the frequency analysis unit. [Figure 4] This graph shows the results learned by the learning unit in Figure 1. [Figure 5] This is a flowchart illustrating the operation during learning by the vacuum pump life prediction device according to the first embodiment of the present invention. [Figure 6] This is a flowchart illustrating the operation during inference by the vacuum pump life prediction device according to the first embodiment of the present invention. [Figure 7] This graph shows an example of a statistical regression model stored in a vacuum pump life prediction device according to the second embodiment of the present invention. [Modes for carrying out the invention]

[0014] Hereinafter, several embodiments of the present invention will be described with reference to the attached drawings. To facilitate understanding of the description, the same reference numerals are used for identical components in each drawing as much as possible, and redundant explanations are omitted as much as possible.

[0015] ===First Embodiment=== First, let me describe the first embodiment.

[0016] "composition" First, the schematic configuration of the vacuum pump life prediction device according to the first embodiment will be described. FIG. 1 is a block diagram schematically showing an example of the functional configuration of the vacuum pump life prediction device according to the first embodiment.

[0017] As shown in FIG. 1, the vacuum pump life prediction device 1 predicts the life of the vacuum pump 10 by performing predetermined processing based on the data detected by the vibration acceleration sensors 21, 22, and 23 installed in the vacuum pump 10. In the first embodiment, the vacuum pump life prediction device 1 performs learning for predicting the life of the vacuum pump 10 and makes an inference about the life of the vacuum pump 10 based on the learning results. Hereinafter, in order to distinguish between the vacuum pump 10 used during learning and the vacuum pump 10 whose life is to be predicted during inference, if necessary, the vacuum pump 10 used during learning will be referred to as the "learning vacuum pump 10A", and the vacuum pump 10 used during inference will be referred to as the "prediction target vacuum pump 10B" for explanation. The learning vacuum pump 10A and the prediction target vacuum pump 10B are, for example, of the same type.

[0018] The vacuum pump life prediction device 1 is an information processing device that is communicably connected to the vibration acceleration sensors 21, 22, 23, the PC 41, the display device 31, etc., and controls the operations of these respective components. FIG. 2 is a block diagram schematically showing an example of the hardware configuration of the vacuum pump life prediction device 1 in FIG. 1.

[0019] As shown in FIG. 2, the vacuum pump life prediction device 1 includes a control device 2, a communication device 3, and a storage device 4. The control device 2 is mainly composed of a CPU (Central Processing Unit) 5 and a memory (main storage device) 6. The CPU 5 controls various components of the vacuum pump life prediction device 1. The memory 6 stores, for example, various programs necessary for the execution of processing in the vacuum pump life prediction device 1.

[0020] The control device 2 functions in various functional configurations by having the CPU 5 execute a predetermined program stored in memory 6 or storage device 4, etc. Details of these functional configurations will be described later. The program may be stored in storage device 4 located inside the vacuum pump life prediction device 1, or it may be stored in storage device, etc., outside the vacuum pump life prediction device 1. Furthermore, the program may be provided on a computer-readable recording medium, or it may be provided in a form that can be installed via a network such as the Internet.

[0021] The communication device 3 consists of a communication interface and the like for communicating with external devices. The communication device 3 transmits and receives various types of information with, for example, the vibration acceleration sensors 21, 22, and 23, the PC 41, and the display device 31.

[0022] The storage device 4 is an auxiliary storage device consisting of a hard disk or the like. This storage device 4 stores various programs, various information, and processing result information necessary for executing processing in the control device 2.

[0023] Furthermore, the vacuum pump life prediction device 1 can be implemented using a dedicated or general-purpose computer. Also, Figure 2 only shows a part of the main hardware configuration of the vacuum pump life prediction device 1, and the vacuum pump life prediction device 1 may have other configurations that are generally found in computers.

[0024] Returning to Figure 1, the vacuum pump life prediction device 1 has the following functional configuration: a data acquisition unit 11, a frequency analysis unit 12, a shift amount calculation unit 13, a learning unit 14, a learning and storage unit 15, a data insertion unit 15A, an inference unit 16, a threshold setting unit 17, a threshold storage unit 18, and a display command unit 19.

[0025] The data acquisition unit 11 (acquisition unit) acquires data detected by vibration acceleration sensors 21, 22, and 23 that detect vibrations during the operation of the vacuum pumps 10 (learning vacuum pump 10A and prediction target vacuum pump 10B). The vibration acceleration sensors 21, 22, and 23 are installed on the vacuum pumps 10 by an operator or the like. The vibration acceleration sensors 21, 22, and 23 each detect vibration accelerations in the mutually orthogonal three-axis directions (X-axis, Y-axis, Z-axis), and detect vibrations during the operation of the vacuum pumps 10 in each of the three axis directions.

[0026] The data acquisition unit 11 acquires vibration acceleration data from the vacuum pump 10 during operation, detected by vibration acceleration sensors 21, 22, and 23, and performs A / D conversion on these data. The A / D converted vibration acceleration data is written to the data register of a PLC (Programmable Logic Controller) (not shown). Furthermore, it may also be stored as a CSV file in the data storage unit 42 (hard disk) of an external PC 41. The vibration acceleration data acquired and A / D converted by the data acquisition unit 11 is output to the frequency analysis unit 12.

[0027] The frequency analysis unit 12 performs frequency analysis on the vibration acceleration data acquired by the data acquisition unit 11. Specifically, the frequency analysis unit 12 performs frequency analysis, i.e., FFT (Fast Fourier Transform) processing, on each vibration acceleration data that has been A / D converted by the data acquisition unit 11 to obtain amplitude spectrum values ​​or power spectrum values ​​(hereinafter simply referred to as "spectrum values"). The frequency analysis unit 12 performs the same processing when vibration acceleration data stored in the data storage unit 42 of the PC 41 is input, as when vibration acceleration data is input from the data acquisition unit 11. The analysis results from the frequency analysis unit 12 are output to the shift amount calculation unit 13.

[0028] Figure 3 is a graph showing an example of the results of FFT processing by the frequency analysis unit 12 in Figure 1. In the example in Figure 3, an example of the results of FFT processing performed by the frequency analysis unit 12 on vibration acceleration data acquired from the vibration acceleration sensor 22 in the Y-axis direction is shown. The horizontal axis in Figure 3 represents frequency [Hz], and the vertical axis in Figure 3 represents spectral values ​​(in this case, amplitude spectral values). Graphs A to D in Figure 3 show data for different usage years of the vacuum pump 10. Graph A shows the case where the vacuum pump 10 is newly installed, i.e., has been in use for 0 years. Graph B shows the case where the usage year of the vacuum pump 10 is greater than 0, a first predetermined number of years. Graph C shows the case where the vacuum pump 10 has been in use for a second predetermined number of years (however, the first predetermined number of years < the second predetermined number of years). Graph D shows the case where the vacuum pump 10 is just before the end of its life (the usage year is greater than the second predetermined number of years, a third predetermined number of years).

[0029] As shown in Figure 3, the peak frequency around 412 Hz differs depending on the number of years of use of the vacuum pump 10, with the peak frequency shifting upward (to the right on the page) for pumps with longer service life. Here, the peak frequency indicates the frequency at which the amplitude spectral value appears as a peak above a predetermined value, and the peak frequency around 412 Hz is the intrinsic vibration frequency of a particular vacuum pump (hereinafter referred to as the "intrinsic vibration frequency").

[0030] The shift amount calculation unit 13 (calculation unit) calculates the amount of change from the reference value of the natural vibration frequency of the vacuum pump 10 based on the data acquired by the data acquisition unit 11. Specifically, the shift amount calculation unit 13 calculates the amount of change from the reference value of the natural vibration frequency of the vacuum pump 10 (for example, the value when it is newly installed) (hereinafter referred to as "shift amount") based on the spectral value obtained by the frequency analysis unit 12 based on the vibration acceleration data acquired by the data acquisition unit 11.

[0031] An example of this calculation procedure is described below. First, the shift amount calculation unit 13 obtains the top N peak frequencies of the vacuum pump 10 that are near the peak frequency corresponding to the natural vibration frequency of the newly installed vacuum pump 10 (for example, within ± a few Hz of the natural vibration frequency) based on the output from the frequency analysis unit 12. Next, the shift amount calculation unit 13 calculates a weighted average value for these top N peak frequencies, using the spectral values ​​as weights. Then, the shift amount calculation unit 13 calculates the difference between the weighted average value obtained in this way and the natural vibration frequency of the newly installed vacuum pump 10 as the shift amount. If the top N number is 0, the learning unit 14 and the inference unit 16 determine that the vacuum pump 10 has reached the end of its lifespan.

[0032] Furthermore, peak frequencies may occur near the harmonics or subharmonics of the natural vibration frequency. In such cases, as described above when peak frequencies occur near the natural vibration frequency (412 Hz), the shift amount calculation unit 13 obtains the top N peak frequencies near these harmonics, calculates a weighted average of these top N peak frequencies using the spectral values ​​as weights, and calculates the difference between the weighted average obtained in this way and the value of these harmonics as the shift amount.

[0033] The shift amount calculation unit 13 then outputs this calculation result to the learning unit 14 during learning and to the inference unit 16 during inference.

[0034] For illustrative purposes, Figure 3 shows the analysis results based on vibration acceleration data in the Y-axis direction detected by the vibration acceleration sensor 22 in the Y-axis direction. However, in reality, the shift amount calculation unit 13 calculates the shift amount based on vibration acceleration data in each direction detected by the vibration acceleration sensors 21 and 23 in the X-axis and Z-axis directions. The shift amount calculation unit 13 calculates a single shift amount by calculating the three-dimensional geometric distance for each of these shift amounts in the X-axis, Y-axis, and Z-axis directions.

[0035] This three-dimensional geometric distance can be calculated using, for example, the Euclidean distance, Manhattan distance, Mahalanobis distance, Chebyshev distance, or Minkowski distance.

[0036] The learning unit 14 performs learning to predict the lifespan of the vacuum pump 10B based on the change in the natural vibration frequency of the learning vacuum pump 10A from a reference value and the state information of the learning vacuum pump 10A. Specifically, the learning unit 14 uses a threshold set by the threshold setting unit 17 (a threshold set based on the state information of the vacuum pump 10A) as a learning label (target variable) and the shift amount output from the shift amount calculation unit 13 as an explanatory variable to perform machine learning on the relationship between the degree of deterioration of the vacuum pump 10A and the shift amount. Alternatively, the learning unit 14 may perform this machine learning using the learning data output to the learning memory unit 15 by the data insertion unit 15A.

[0037] Figure 4 is a graph showing an example of the results learned by the learning unit 14 in Figure 1. The horizontal axis of the graph in Figure 4 represents the shift amount [Hz] calculated using three-dimensional geometric distance, and the vertical axis represents the degree of aging degradation [years]. This degree of aging degradation indicates the equivalent number of years of use corresponding to the degradation state of the vacuum pump 10. For example, "degree of aging degradation of 5 years" means "a degree of degradation equivalent to 5 years of use." In Figure 4, the curved graph indicated by the dashed circle represents the machine learning regression model as a result of learning by the learning unit 14, and is used for inference by the inference unit 16. The solid circle in Figure 4 indicates measured values.

[0038] Furthermore, the learning unit 14 performs the above learning whenever a threshold is set by the threshold setting unit 17 (when it receives state information input from the user). That is, when the threshold is reset by the threshold setting unit 17, the learning unit 14 changes the original threshold to the reset threshold and retrains. The learning unit 14 may also perform the above learning whenever the learning data output to the learning memory unit 15 by the data insertion unit 15A is changed. The learning results from the learning unit 14 are output to the learning memory unit 15.

[0039] The learning memory unit 15 stores the learning results from the learning unit 14. The learning memory unit 15 updates the contents of the stored learning results each time the learning unit 14 performs retraining. The learning memory unit 15 also stores the machine learning training data output (imported) by the data insertion unit 15A.

[0040] The data insertion unit 15A outputs learning data collected from outside the vacuum pump life prediction device 1 to the learning memory unit 15. This learning data is collected from multiple learning vacuum pumps 10A in new condition or various deterioration conditions (years of use), and includes information such as the years of use and shift amount associated with each other. The measuring device for collecting this learning data is preferably a portable measuring device.

[0041] The inference unit 16 infers the lifespan of the vacuum pump 10B based on the shift amount calculated by the shift amount calculation unit 13. More specifically, the inference unit 16 infers the degree of aging degradation or lifespan of the vacuum pump 10B based on the learning results from the learning unit 14 and the shift amount calculated by the shift amount calculation unit 13 for the vacuum pump 10B to be predicted.

[0042] Here, we will explain the meaning of inferring the degree of deterioration over time. Users can determine the actual number of years of use of the vacuum pump 10B themselves. If the vacuum pump 10B continues to operate in a normal state, the degree of deterioration over time and the actual number of years of use will be approximately the same, so the user can predict the approximate lifespan of the vacuum pump 10B based on the actual number of years of use without having to determine the degree of deterioration over time. However, in reality, the vacuum pump 10B may not continue to operate in a normal state, and in this case, the degree of deterioration over time and the actual number of years of use will be different, making it difficult for the user to accurately predict the lifespan of the vacuum pump 10B from the actual number of years of use. In the first embodiment, the lifespan of the vacuum pump 10B can be predicted with high accuracy by determining the degree of deterioration over time using the inference unit 16.

[0043] Specifically, the inference unit 16 can infer the lifespan of the vacuum pump 10B based on a machine learning regression model (see Figure 4) stored in the learning memory unit 15, for example. In this machine learning regression model, the inference unit 16 defines the degree of aging degradation just before the graph converges as the critical lifespan M, and the shift amount corresponding to this critical lifespan M as the critical shift amount L. The critical lifespan M is the maximum lifespan of the vacuum pump 10B in a newly installed state with zero years of use. Next, in this machine learning regression model, the inference unit 16 identifies the degree of aging degradation corresponding to the shift amount calculated by the shift amount calculation unit 13 for the vacuum pump 10B. Then, based on the critical lifespan M and the identified degree of aging degradation, the inference unit 16 determines the current lifespan of the vacuum pump 10B (the period until replacement or overhaul becomes necessary, based on the current time).

[0044] The inference unit 16 can calculate the current lifespan of the vacuum pump 10B by, for example, the difference between the critical lifespan M and the degree of aging degradation. For example, if the critical lifespan M is 22 years and the degree of aging degradation is 10 years, the current lifespan of the vacuum pump 10B is calculated to be 12 years. The lifespan calculated by this inference unit 16 is shorter than the lifespan expected from the actual years of use (for example, if the actual years of use are 3 years and the critical lifespan M is 22 years, it would be expected to be 19 years) if the aging degradation has progressed further than the actual years of use. The inference unit 16 also determines that the vacuum pump 10B has reached the end of its lifespan if the shift amount is greater than or equal to the critical shift amount L (i.e., if the specified degree of aging degradation is greater than or equal to the critical lifespan M).

[0045] Furthermore, each time the learning results stored in the learning memory unit 15, which have been learned by the learning unit 14, are updated, the inference unit 16 infers the degree of aging deterioration or lifespan of the vacuum pump 10B to be predicted (by recalculating retrospectively to past inferences) based on the learned results and the shift amount calculated by the shift amount calculation unit 13, and updates the inference results. The inference results from the inference unit 16 are output to the display command unit 19.

[0046] The threshold setting unit 17 (reception unit) receives input of state information for the learning vacuum pump 10A from the user. The threshold setting unit 17 has a user interface that allows the user to input state information for the vacuum pump 10A. Based on the state information input by the user through the user interface, the threshold setting unit 17 automatically calculates and sets a threshold as a condition that will serve as the basis for learning by the learning unit 14, and outputs it to the threshold storage unit 18. The state information is information that indicates the state of the vacuum pump 10A over time, and examples include the number of years the vacuum pump 10A has been in use, the date of manufacture, the date of overhaul, etc. The threshold as a condition that will serve as the basis for learning is, for example, the number of years the vacuum pump 10A has been in use. For example, if the user inputs the number of years in use as state information, the input number of years in use is used as the threshold. Also, for example, if the user inputs the date of manufacture as state information, the number of years in use obtained by calculating the difference between that date and the actual date is used as the threshold. The threshold setting unit 17 may also accept the input of the threshold from the user themselves and set the received threshold as is.

[0047] The threshold setting unit 17 changes the threshold each time it receives status information input from the user. Examples of situations in which the user may re-enter status information include adding the number of vacuum pumps 10A to be trained, or re-entering status information if it is found that the actual deterioration state is worse than the previously entered status information.

[0048] The threshold storage unit 18 stores the threshold calculated by the threshold setting unit 17.

[0049] The display command unit 19 (output unit) outputs a display command to the display device 31 that displays the degree of aging deterioration or lifespan of the vacuum pump 10B being measured, which has been inferred by the inference unit 16. The display command unit 19 outputs a display command to display the updated inference result whenever the inference result (degree of aging deterioration or lifespan) by the inference unit 16 is updated. The display device 31 has a graphical user interface (screen) and displays the degree of aging deterioration or lifespan according to the display command output from the display command unit 19. The display device 31 updates the display content each time a display command is output (each time the inference result is updated). The display command unit 19 may also output sound output commands to a sound output device (not shown), such as a speaker, to output the inference result as sound, rather than just display commands.

[0050] 《Processing during learning》 Next, the learning process by the vacuum pump life prediction device 1 will be explained using the flowchart in Figure 5.

[0051] (Step SP1) The threshold setting unit 17 sets a threshold (years of use) as a condition for learning by the learning unit 14, based on the status information of the vacuum pump 10A entered by the user. After that, the process moves on to step SP2.

[0052] (Step SP2) The measurement begins when the user presses the measurement start button (not shown) of the vacuum pump life prediction device 1. The process then proceeds to step SP3.

[0053] (Step SP3) The data acquisition unit 11 acquires vibration acceleration data from the vacuum pump 10A during operation, as detected by vibration acceleration sensors 21, 22, and 23, and performs A / D conversion on these data. After that, the process proceeds to step SP4.

[0054] (Step SP4) The frequency analysis unit 12 performs FFT processing on each vibration acceleration data acquired and A / D converted by the data acquisition unit 11 in step SP3 to obtain spectral values. After that, the process moves on to step SP5.

[0055] (Step SP5) The shift amount calculation unit 13 calculates the shift amount for the vacuum pump 10A based on the spectral values ​​obtained in step SP4. The process then proceeds to step SP6.

[0056] (Step SP6) The learning unit 14 uses machine learning to determine the relationship between the degree of aging deterioration of the vacuum pump 10A and the shift amount, based on the shift amount calculated in step SP5 and the threshold (years of use) set in step SP1. After that, the process moves on to step SP7.

[0057] (Step SP7) The learning unit 14 determines whether or not a condition has been changed by user operation. A change in condition includes the threshold being reset by the threshold setting unit 17 based on user input. If the determination is positive, the process proceeds to step SP8; if the determination is negative, the process proceeds to step SP9.

[0058] (Step SP8) The learning unit 14 retrains based on the conditions changed in step SP7. Afterward, the process moves on to step SP9.

[0059] (Step SP9) The learning unit 14 determines whether the user has pressed the measurement end button (not shown) of the vacuum pump life prediction device 1. If the determination is positive, the process in the flowchart of Figure 5 is completed. If the determination is negative, the process proceeds to step SP3.

[0060] 《Inference process》 Next, the processing during inference by the vacuum pump life prediction device 1 will be explained using the flowchart in Figure 6.

[0061] (Step SP11) The user activates the measurement start button (not shown) of the vacuum pump life prediction device 1, initiating the measurement. The process then proceeds to step SP12.

[0062] (Step SP12) The data acquisition unit 11 acquires vibration acceleration data from the vacuum pump 10B during operation, as detected by vibration acceleration sensors 21, 22, and 23, and performs A / D conversion on these data. After that, the process proceeds to step SP13.

[0063] (Step SP13) The frequency analysis unit 12 performs FFT processing on each vibration acceleration data acquired and A / D converted by the data acquisition unit 11 in step SP12 to obtain spectral values. After that, the process moves on to step SP14.

[0064] (Step SP14) The shift amount calculation unit 13 calculates the shift amount for the vacuum pump 10B based on the spectral values ​​obtained in step SP13. After that, the process proceeds to step SP15.

[0065] (Step SP15) The inference unit 16 infers the degree of aging degradation and lifespan of the vacuum pump 10B based on the shift amount calculated in step SP14 and the learning results from the learning unit 14 in the processing of step SP6 in Figure 5. The process then proceeds to the processing of step SP16.

[0066] (Step SP16) The display command unit 19 outputs a display command to the display device 31 that indicates the degree of aging deterioration or lifespan of the vacuum pump 10B, which was inferred in the processing of step SP16, and the display device 31 displays it. After that, the process proceeds to the processing of step SP17.

[0067] (Step SP17) The inference unit 16 determines whether or not a condition has been changed by user operation. A change in condition includes the threshold being reset by the threshold setting unit 17 based on user input, or learning data being added from the data insertion unit 15A to the learning memory unit 15. If the determination is positive, the process proceeds to step SP18; if the determination is negative, the process proceeds to step SP12.

[0068] (Step SP18) The learning unit 14 retrains based on the conditions changed in step SP17. Afterward, the process moves on to step SP19.

[0069] (Step SP19) The inference unit 16 re-infers (updates the inference result) the service life and lifespan of the vacuum pump 10B based on the results relearned in step SP18. After that, the process proceeds to step SP20.

[0070] (Step SP20) The display command unit 19 outputs a display command to the display device 31 that indicates the degree of aging deterioration or lifespan of the vacuum pump 10B, which was re-inferred during the processing of step SP19, and the display device 31 re-displays (updates) this command. After that, the process proceeds to step SP12. The above is a description of the processing of the vacuum pump life prediction device 1 according to the first embodiment.

[0071] Effects and Benefits In the first embodiment, the system includes a data acquisition unit 11 (acquisition unit) that acquires data detected by vibration acceleration sensors 21, 22, and 23 that detect vibrations during operation of the vacuum pump 10B; a shift amount calculation unit 13 (calculation unit) that calculates the amount of shift (change) of the natural vibration frequency of the vacuum pump 10B from a reference value based on the data acquired by the data acquisition unit 11; and an inference unit 16 that infers the lifespan of the vacuum pump 10B based on the change calculated by the shift amount calculation unit 13.

[0072] With this configuration, the lifespan of the vacuum pump 10B can be inferred based on the change in its natural vibration frequency from a reference value. Since the natural vibration frequency of the vacuum pump 10B is constant under various process conditions, there is no need to set multiple reference values ​​for each process condition as in conventional technology, and the lifespan of the vacuum pump 10B can be easily predicted.

[0073] Furthermore, in the first embodiment, the system includes a learning unit 14 that performs learning to predict the lifespan based on the amount of shift (change) of the natural vibration frequency of the learning vacuum pump 10A from a reference value and the state information of the learning vacuum pump 10A. The inference unit 16 infers the lifespan based on the learning results from the learning unit 14 and the amount of shift calculated by the shift amount calculation unit 13.

[0074] With this configuration, the lifespan of the vacuum pump 10B can be predicted with high accuracy because it is inferred from the learning results based on the shift amount and usage conditions of the learning vacuum pump 10A.

[0075] Furthermore, in the first embodiment, a threshold setting unit 17 (reception unit) is provided to receive state information input from the user. The learning unit 14 performs learning each time it receives state information input from the threshold setting unit 17, and the inference unit 16 infers the lifespan each time learning is performed by the learning unit 14, based on the learning results from the learning unit 14 and the shift amount calculated by the shift amount calculation unit 13.

[0076] With this configuration, the system can relearn each time it receives status information (such as years of use) from the user, and each time it relearns, it can infer the lifespan of the vacuum pump 10B based on the relearned results. Therefore, it can dynamically learn and infer in response to user input, thereby improving user convenience.

[0077] Furthermore, the first embodiment further includes a display command unit 19 (output unit) that outputs the lifespan inferred by the inference unit 16.

[0078] With this configuration, the inferred lifespan can be visualized by outputting it to an external display device 31, or output as audio, thereby allowing the user to recognize the lifespan.

[0079] Furthermore, the program according to the first embodiment causes the vacuum pump life prediction device 1, which is a computer capable of communicating with vibration acceleration sensors 21, 22, and 23 that detect vibrations during operation of the vacuum pump 10B, to function as a data acquisition unit 11 (acquisition unit) that acquires data detected by the vibration acceleration sensors 21, 22, and 23 that detect vibrations during operation of the vacuum pump 10B, a shift amount calculation unit 13 (calculation unit) that calculates the amount of shift (change) of the natural vibration frequency of the vacuum pump 10B from a reference value based on the data acquired by the data acquisition unit 11, and an inference unit 16 that infers the life of the vacuum pump 10B based on the change calculated by the shift amount calculation unit 13.

[0080] With this configuration, the lifespan of the vacuum pump 10B can be inferred based on the change in its natural vibration frequency from a reference value. Since the natural vibration frequency of the vacuum pump 10B is constant under various process conditions, there is no need to set multiple reference values ​​for each process condition as in conventional technology, and the lifespan of the vacuum pump 10B can be easily predicted.

[0081] ===Second Embodiment=== Next, a vacuum pump life prediction device according to the second embodiment will be described. The second embodiment differs from the first embodiment in that the learning unit 14 uses statistical regression in addition to the machine learning regression described above. In the following, descriptions that overlap with the first embodiment will be omitted as much as possible, and the differences will be the focus of the description. The configuration and functions of the vacuum pump life prediction device according to the second embodiment, which will not be described below, are the same as those of the vacuum pump life prediction device 1 according to the first embodiment.

[0082] In the second embodiment, a statistical storage unit is further provided for storing a statistical regression model acquired in advance for a statistical vacuum pump. This statistical regression model is, for example, data showing the correspondence between the amount of shift (change) of the natural vibration frequency of the statistical vacuum pump from a reference value and the degree of deterioration of the vacuum pump over time. This statistical regression model is calculated by regression analysis based on measured values ​​of the amount of shift of the natural vibration frequency of the statistical vacuum pump from a reference value and the number of years of use (condition information) of the statistical vacuum pump.

[0083] Figure 7 is a graph showing an example of a statistical regression model stored in the vacuum pump life prediction device according to the second embodiment. The horizontal axis of the graph in Figure 7 represents the shift amount [Hz] of the natural vibration frequency from the reference value, and the vertical axis represents the degree of deterioration over time [years]. In Figure 7, the parabolic graph shown by the dashed line represents the statistical regression model and is used for inference by the inference unit 16. The solid circle in Figure 7 indicates measured values.

[0084] When the inference unit 16 infers the lifespan using a statistical regression model, it identifies the degree of aging degradation corresponding to the shift amount calculated by the shift amount calculation unit 13 for the vacuum pump 10B in a statistical regression model, for example, as shown in Figure 7. For example, the inference unit 16 can determine the difference between the identified degree of aging degradation and a preset limit lifespan as the current lifespan of the vacuum pump 10B.

[0085] As described above, in the vacuum pump life prediction device 1 according to the second embodiment, the lifespan of the vacuum pump 10B can be easily predicted based on the amount of change in the natural vibration frequency of the vacuum pump 10B from a reference value, similar to the first embodiment.

[0086] ===Literal translation=== It should be noted that the present invention is not limited to the embodiments described above. That is, any design modifications made to the above specific examples by those skilled in the art are also included within the scope of the present invention, as long as they retain the features of the present invention. Furthermore, the elements of the embodiments described above and the modifications described later can be combined to the extent that it is technically possible, and any combination thereof is also included within the scope of the present invention, as long as it retains the features of the present invention.

[0087] For example, in the first and second embodiments, vibration acceleration data in the three axes is detected using vibration acceleration sensors 21, 22, and 23, respectively. However, vibration acceleration data in one direction may be detected using a single vibration acceleration sensor. In that case, the calculation of the three-dimensional geometric distance is omitted in the shift amount calculation unit 13.

[0088] Furthermore, in the first and second embodiments, the shift amount of the natural vibration frequency of the vacuum pump 10 from the reference value (the value in the newly installed state) is calculated based on the spectral values ​​obtained by the frequency analysis unit 12. However, the relative value of the peak frequency normalized by the absolute value of the natural vibration frequency in the newly installed state may be calculated and used instead of the shift amount for learning by the learning unit 14 and inference by the inference unit 16.

[0089] Furthermore, in the first and second embodiments, the display command unit 19 (output unit) outputs commands to the display device 31 and the sound output device, but it may also be done by writing to a specific register or device of the PLC.

[0090] Furthermore, while the first embodiment described an example in which the learning unit 14 performs machine learning, this machine learning is not limited to supervised learning as described in the first embodiment, but may also be unsupervised learning.

[0091] Furthermore, while the first embodiment described a method using a machine learning regression model and the second embodiment described a method using a statistical regression model as methods for inferring lifetime by the inference unit 16, the present invention is not limited thereto. For example, the lifetime may be estimated by appropriately combining a machine learning regression model and a statistical regression model. Alternatively, the inference unit 16 may set a predetermined threshold for the shift amount and estimate the lifetime by comparing the shift amount calculated by the shift amount calculation unit 13 with the threshold. [Explanation of symbols]

[0092] 1... Vacuum pump life prediction device (computer), 10... Vacuum pump, 11... Data acquisition unit (acquisition unit), 13... Shift amount calculation unit (calculation unit), 14... Learning unit, 16... Inference unit, 17... Threshold setting unit (reception unit), 19... Display command unit (output unit)

Claims

1. An acquisition unit that acquires data detected by a vibration acceleration sensor that detects vibrations during the operation of a vacuum pump, A calculation unit calculates the amount of change from a reference value of the natural vibration frequency of the vacuum pump based on the data acquired by the acquisition unit, An inference unit that infers the lifespan of the vacuum pump based on the amount of change calculated by the calculation unit, A vacuum pump life prediction device equipped with the following features.

2. The system further includes a learning unit that performs learning to predict the lifespan based on the change in the natural vibration frequency of the learning vacuum pump from a reference value and the state information of the learning vacuum pump, The inference unit infers the lifetime based on the learning results from the learning unit and the change amount calculated by the calculation unit. The vacuum pump life prediction device according to claim 1.

3. The system further includes a reception unit that receives the aforementioned status information from the user, The learning unit performs the learning each time it receives the state information input from the reception unit. The inference unit, each time the learning unit performs the learning, infers the lifetime based on the learning results from the learning unit and the change amount calculated by the calculation unit. The vacuum pump life prediction device according to claim 2.

4. The system further includes an output unit that outputs the lifespan inferred by the inference unit. A vacuum pump life prediction device according to any one of claims 1 to 3.

5. A vibration acceleration sensor that detects vibrations during the operation of a vacuum pump and a computer capable of communicating with it, An acquisition unit that acquires data detected by the vibration acceleration sensor, A calculation unit calculates the amount of change from the reference value of the natural vibration frequency of the vacuum pump based on the data acquired by the acquisition unit. An inference unit that infers the lifespan of the vacuum pump based on the amount of change calculated by the calculation unit, A program designed to function as such.

Images