Motor device inspection apparatus and motor device inspection method

Abstract

To provide an inspection device and the like capable of inspecting a motor device with high accuracy while reducing a workload.SOLUTION: An inspection device for inspecting a motor device comprises: a data acquisition unit that acquires time-series data including sound pressure or vibration waveforms based on sound or vibration generated during the operation of the motor device; a data processing unit that executes frequency analysis on the time-series data acquired by the data acquisition unit; and a determination unit that determines evaluation classification of the motor device on the basis of a result of frequency analysis obtained by the data processing unit. The data processing unit executes the frequency analysis on the time-series data by limiting to a section of a temporal axis set according to a type of abnormal sound or a frequency band set according to the type of abnormal sound. The determination unit performs determination by an evaluation classification determined on the basis of audibility of an operation sound of the motor device and an algorithm on the basis of machine learning for the result of frequency analysis.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an inspection device for a motor device and an inspection method for a motor device.

Background Art

[0002] Patent Document 1 discloses a quality inspection device that compares a pre-created threshold value with a psychoacoustic parameter to determine the presence or absence of abnormal noise for the purpose of realizing a quality inspection similar to a human auditory sensation inspection.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, when trying to determine the presence or absence of abnormal noise in a motor device with high accuracy, the number of necessary threshold parameters becomes extremely large, and the work burden for setting the threshold and making the determination increases.

[0005] The present invention has been made to solve the above problems, and an object thereof is to provide an inspection device or the like that can perform inspection of a motor device with high accuracy while reducing the work burden.

Means for Solving the Problems

[0006] To solve the above problems, one aspect of the present invention is an inspection device for inspecting a motor device, comprising The aforementioned a data acquisition unit that acquires time-series data of sound pressure or vibration waveform based on the sound or vibration generated during the operation of the motor device, a data processing unit that performs frequency analysis on the time-series data acquired by the data acquisition unit, A determination unit determines the evaluation category of the motor device based on the results of frequency analysis obtained by the data processing unit, Equipped with, The data processing unit performs the following actions on the time-series data: Determined by the number of slots or poles of the motor in the aforementioned motor device. Frequency analysis is performed within a time interval or frequency band set according to the type of abnormal noise. The partial overall value is calculated as a feature quantity combined with the target interval on the time axis, and the motor current value is output to the determination unit as time series data aligned with the time series data on the time axis. The determination unit makes a determination based on the evaluation category determined based on the auditory perception of the operating sound of the motor device, and on the results of the frequency analysis using an algorithm based on machine learning. The evaluation category of the motor device is determined by using the information relating to the current value of the motor as a feature quantity, in conjunction with the results of the frequency analysis. , to provide inspection equipment. [Effects of the Invention]

[0007] According to this invention, it is possible to perform high-precision inspections of motor devices while reducing the workload. [Brief explanation of the drawing]

[0008] [Figure 1] This diagram shows the configuration of the inspection apparatus in this embodiment. [Figure 2] This figure shows the fluctuations in the load on the motor during inspection of the motor device. [Figure 3] This figure shows the state of the decision-making unit during machine learning. [Figure 4] This figure shows an example of using the motor current value to determine the evaluation category of a motor device. [Modes for carrying out the invention]

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0010] Figure 1 shows the configuration of the inspection device in this embodiment. The inspection device 1 in this embodiment is a device that makes it possible to replace inspections in which an inspector determined the evaluation category based on the operating sound of a motor device based on their auditory perception with an inspection using AI (Artificial Intelligence).

[0011] As shown in Figure 1, the inspection device 1 includes a data acquisition unit 11 that acquires time-series data containing sound pressure or vibration waveforms based on the sound or vibration generated when the motor device 20 is in operation, a data processing unit 12 that performs frequency analysis on the time-series data acquired by the data acquisition unit 11, a determination unit 13 that determines the evaluation category of the motor device 20 using artificial intelligence (AI) such as a random forest based on the feature quantities which are the results of the frequency analysis obtained by the data processing unit 12, and a control unit 15 that controls the operation of the inspection device 1 and the motor device 20.

[0012] The motor device 20 is, for example, a device having a motor and a mechanism driven by the motor, and the motor device 20 may be the motor itself. The evaluation categories for the motor device 20 are, for example, pass or fail (pass, fail), but three or more evaluation categories may be provided.

[0013] As shown in Figure 1, the motor unit 20 to be inspected is placed in an inspection booth 40 installed on or near the manufacturing line and is controlled by a control unit 15. Sensors 32a and 32b, such as acceleration sensors, are attached to the motor unit 20 to detect vibrations of the motor unit 20. In addition, a sound pressure sensor 31 (microphone) is provided inside the inspection booth 40 to capture the operating sound of the motor unit 20.

[0014] In the example shown in Figure 1, two sensors 32a and 32b may be provided to acquire data on vibrations in different directions (e.g., accelerations in different directions), or to acquire data on vibrations at different locations within the motor device 20. However, the number of sensors used to detect vibrations is arbitrary. Multiple vibration data points in different directions, and multiple vibration data points at different locations within the motor device 20, can each be treated as separate time-series data. In this case, since the direction of vibration and the location where vibrations occur will differ depending on the type of abnormal noise, it becomes possible to appropriately detect abnormal noises by adding the direction of vibration and the location where vibrations occur as features.

[0015] Alternatively, instead of the sound pressure sensor 31, a plurality of sound pressure sensors (microphones) with different installation locations and directivities may be set. The sound pressure data obtained by the plurality of sound pressure sensors can be handled as individual time-series data respectively.

[0016] The motor device 20 operates according to the control by the control unit 15. During operation, the vibration is converted into an electrical signal by the sensors 32a and 32b, and the operating sound is converted into an electrical signal by the sound pressure sensor 31, and then acquired by the data acquisition unit 11. The vibration waveform obtained by the sensors 32a and 32b and the sound pressure obtained by the sound pressure sensor 31 are handled as time-series data with a common time axis.

[0017] The control method of the motor device 20 to be inspected is arbitrary. For example, the control unit 15 may control the rotation speed and load of the motor in the motor device 20, or the drive voltage waveform applied to the motor. Since the motor device 20 performs a periodic operation, the period of one inspection cycle is set to include at least one operation cycle of the motor device 20. The period of one inspection cycle may be set to include a plurality of operation cycles of the motor device 20. In any case, the time axis of the time-series data needs to be grasped in relation to the operation cycle of the motor device 20.

[0018] FIG. 2 is a diagram showing the variation of the load on the motor during the inspection of the motor device.

[0019] [[ID=第十八]] Generally, the motor device 20 exhibits different sound pressure and vibration frequency bands depending on the type of abnormal noise, based on factors such as the motor's rotational state, the meshing of the drive mechanism, and the magnetic circuit. Therefore, in this embodiment, each frequency band corresponding to a different type of abnormal noise is set as the target range for frequency analysis. For example, if the frequency of an abnormal noise is determined by the number of slots or poles, the frequency band corresponding to the motor's rotational speed at that time can be set as the target range for frequency analysis. The data processing unit 12 calculates the partial overall value in the set frequency band as a feature and outputs it to the determination unit 13.

[0020] Furthermore, since the types of abnormal noises that can occur generally differ depending on the load and rotational speed of the motor, in this embodiment, the load and rotational speed are continuously changed within a single inspection cycle. Figure 2 shows an example of changing the load within an inspection cycle. In the example in Figure 2, the time axis of the inspection cycle is divided into eight intervals t1 to t7 in accordance with the load fluctuations. Intervals t1 to t7 can be set considering the timing of motor start-up, load fluctuations, and reversal of rotation direction. Then, from intervals t1 to t7, the interval corresponding to the type of abnormal noise is set as the target interval for frequency analysis.

[0021] For example, for types of abnormal noises that occur in low-load sections (e.g., section t1) or high-load sections (e.g., section t3), the target section for frequency analysis is set to include the corresponding sections.

[0022] Furthermore, for example, if the drive mechanism is a reduction mechanism that converts the rotation of the motor into the reciprocating motion of the wiper, there is a possibility that abnormal noise may occur in the section where the direction of the wiper's motion changes due to a change in the meshing state of the reduction mechanism. In such cases, the target section for frequency analysis may be set to include the section in the operating cycle of the motor device 20 so that this type of abnormal noise can be detected.

[0023] In this way, the data processing unit 12 sets a target interval in the time axis and a frequency band corresponding to the type of abnormal noise, and performs frequency analysis limited to the range combining the frequency band and time period to calculate the partial overall value. Furthermore, the calculated partial overall value is used as a feature quantity combined with the time period (target interval in the time axis) for the determination unit 13 to make a determination. As described above, the direction of vibration and the location where the vibration is occurring can also be combined with the partial overall value of the vibration waveform in the time period and a specific frequency band and used as feature quantities.

[0024] Therefore, the load on the frequency analysis in the data processing unit 12 and the determination in the determination unit 13 can be reduced, and the determination accuracy in the determination unit 13 can be improved.

[0025] Furthermore, the scope of frequency analysis may be limited to either the frequency band or the time period. For example, for abnormal noises that may occur regardless of the time period, only the frequency band may be limited, and if it is difficult to identify the frequency band that may be included in the abnormal noise, only the time period may be limited. Also, the scope of frequency analysis may be limited to either sound pressure or vibration waveform.

[0026] Figure 3 shows the state of the judgment unit during machine learning.

[0027] In this embodiment, the determination unit 13 is provided with learning data consisting of partial overall values ​​in a specific frequency band associated with a specific time period, which are the results of frequency analysis performed by the data processing unit 12 on a predetermined number of motor devices 20, and auditory judgment results for the same motor devices 20. Data from multiple devices is randomly extracted from the predetermined number without removing duplicates, and learning is performed to select multiple features with high importance from among them. This creates a random forest that determines the evaluation category of the motor devices 20 by majority voting with multiple different decision trees. During learning, as with determining the evaluation category of the motor devices 20, a target interval in the frequency band and time axis corresponding to the type of abnormal sound is set, and the data processing unit 12 performs frequency analysis limited to a range combining the frequency band and time period. Thus, the determination unit 13 is provided with the partial overall value calculated by the data processing unit 12 as data combined with the time period, just as with determining the evaluation category of the motor devices 20.

[0028] Furthermore, in this embodiment, data containing disturbances may be used as time-series data during learning. Specifically, the motor device 20 can be placed in the same environment as when the evaluation category of the motor device 20 is determined, and learning can be performed using the time-series data of sound pressure and vibration waveform obtained at that time. As shown in Figure 1, when the evaluation category of the motor device 20 is determined, the motor device 20 is placed in the inspection booth 40. However, it is difficult to provide the inspection booth 40 with high sound insulation performance, and noise generated in the manufacturing line is captured as disturbance by the sound pressure sensor 31. In addition, the vibrations captured by sensors 32a and 32b may also be affected by disturbances. Therefore, even during learning, by using the frequency analysis results of time-series data containing such disturbances, robustness to disturbances can be increased, and the accuracy of the determination unit 13 can be improved.

[0029] In this embodiment, the determination unit 13 determines the evaluation category of the motor device 20 not only by using the results of frequency analysis of the sound pressure generated during the operation of the motor device 20, but also by using the results of frequency analysis of the vibration waveform generated during the operation of the motor device 20. By determining the evaluation category of the motor device 20 using vibration waveforms as well as sound pressure, the influence of external disturbances can be suppressed, and the accuracy of the determination can be greatly improved. In other words, in a manufacturing environment, noise from surrounding manufacturing machinery and conveying machinery mixes with the operating sound of the motor device 20, and it is difficult to eliminate the influence of noise on the time-series data of sound pressure. On the other hand, it is relatively easy to significantly suppress the influence of vibrations transmitted from the outside on the vibration of the motor device 20. Also, vibration has a relatively high correlation with the generation of abnormal noise. For this reason, in this embodiment, determining the evaluation category of the motor device 20 based on vibration waveforms as well as sound pressure is a very effective means of suppressing the influence of external disturbances on sound pressure. If the evaluation category of the motor device 20 is determined using only sound, a complex algorithm to eliminate external noise and soundproofing equipment will be required, leading to increased manufacturing costs and increased calculation time required for determination. In contrast, as in this embodiment, by using vibration in combination, it is possible to improve robustness against disturbances while suppressing cost increases and computational burdens.

[0030] Figure 4 shows an example of using the motor current value to determine the evaluation category of a motor device.

[0031] In determining the evaluation category of the motor device 20, time-series data of the current value (motor current value) supplied to the motor included in the motor device 20 can be used in addition to time-series data of sound pressure or vibration waveform.

[0032] The motor current value can be obtained from the motor device 20 or the control unit 15, and for example, as shown in Figure 4, it is obtained in the data processing unit 12. The data processing unit 12 outputs the obtained motor current value to the determination unit 13 as time-series data aligned with the time-series data of sound pressure or vibration waveform. The determination unit 13 uses the information related to the motor current value as a feature quantity, along with the results of frequency analysis of sound pressure or vibration waveform, to determine the evaluation category of the motor device.

[0033] Depending on the type of abnormal noise, the waveform of the motor's current value may show a high correlation with the occurrence of the abnormal noise. For example, when an abnormal noise occurs, there may be an increase or decrease in the motor's current value, or the waveform of the motor's current value may show a specific shape. In addition, when an abnormal noise occurs, the current value may show oscillations in the same frequency band as the abnormal noise, or in a correlated frequency band. On the other hand, the motor's current value is basically not affected by external disturbances to the motor device 20.

[0034] Therefore, by taking the motor's current value into account when determining the evaluation category of a motor device, the accuracy of the evaluation category determination can be improved. In other words, by adding the motor's current value to the features, robustness to disturbances can be further improved without increasing manufacturing costs or the computation time required for determination.

[0035] The parameter form of the motor current value used for the determination is arbitrary. For example, the direction (increase or decrease) or waveform of the motor current value may be used as a feature, or the result of frequency analysis of the motor current value may be used as a feature.

[0036] As described above, in the above embodiment, the determination unit 13 performs determination using an algorithm based on machine learning. Therefore, it is not necessary to prepare thresholds for a huge number of parameters to determine the presence or absence of abnormal noise in the motor device, and the workload for setting thresholds and making determinations can be reduced. In addition, the motor device can be inspected with high accuracy.

[0037] Furthermore, in this embodiment, the determination unit 13 uses not only sound pressure but also the results of frequency analysis of the vibration waveform generated during the operation of the motor device 20 as a feature to determine the evaluation category of the motor device 20. Therefore, the influence of disturbances can be suppressed, and the accuracy of determining the evaluation category of the motor device 20 can be improved. Also, in the example shown in Figure 4, the determination unit 13 uses not only sound pressure or vibration waveform but also parameters related to the motor current value during the operation of the motor device 20 as a feature to determine the evaluation category of the motor device 20. Therefore, the influence of disturbances can be suppressed, and the accuracy of determining the evaluation category of the motor device 20 can be improved.

[0038] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and includes designs and the like that do not depart from the spirit of this invention.

[0039] Furthermore, the following additional information is disclosed regarding the embodiments of the present invention described above.

[0040] [Note 1] An inspection device (1) for inspecting a motor device (20), A data acquisition unit (11) acquires time-series data that includes sound pressure and vibration waveforms based on the sound and vibration generated when the motor device is in operation, A data processing unit (12) performs frequency analysis on the time-series data acquired by the data acquisition unit, A determination unit (13) determines the evaluation category of the motor device based on the results of the frequency analysis obtained by the data processing unit, Equipped with, The data processing unit performs frequency analysis on the time-series data, limited to a time axis interval set according to the type of abnormal sound, or to a frequency band set according to the type of abnormal sound. The determination unit is an inspection device that determines the evaluation category based on the auditory perception of the operating sound of the motor device and makes a determination based on the results of the frequency analysis using an algorithm based on machine learning.

[0041] According to the configuration described in Appendix 1, the determination unit performs determination using an algorithm based on machine learning, thereby reducing the workload while obtaining highly accurate determination results. Furthermore, the data processing unit performs frequency analysis on time-series data, limited to time intervals set according to the type of abnormal sound, or to frequency bands set according to the type of abnormal sound. This allows for effective detection of abnormal sounds while reducing the processing burden associated with frequency analysis and determination.

[0042] [Note 2] The inspection apparatus as described in Appendix 1, wherein data containing disturbances is used as the time-series data during the learning process for the determination unit.

[0043] According to the configuration described in Appendix 2, when training the judgment unit, data containing disturbances is used as time-series data, thereby increasing robustness to disturbances and improving the judgment accuracy of the judgment unit.

[0044] [Note 3] The determination unit is an inspection device according to Appendix 1 or Appendix 2, which performs determination using a random forest.

[0045] According to the configuration described in Appendix 3, a random forest can be used to obtain highly accurate judgment results while reducing the workload.

[0046] [Note 4] A method for inspecting a motor device, A data acquisition step to acquire time-series data of sound pressure or vibration waveform based on sound or vibration generated during the operation of a motor device, A data processing step in which frequency analysis is performed on the time series data acquired in the data acquisition step, A determination step to determine the evaluation category of the motor device based on the results of the frequency analysis obtained in the above data processing step, The computer executes this, In the data processing step, frequency analysis is performed only within a time interval set according to the type of abnormal sound, or within a frequency band set according to the type of abnormal sound. The inspection method, in the determination step, involves determining the evaluation category based on the auditory perception of the operating sound of the motor device and making a determination based on the results of the frequency analysis using an algorithm based on machine learning.

[0047] According to the configuration described in Appendix 4, the judgment step uses a machine learning-based algorithm to make a judgment, thereby reducing the workload while obtaining highly accurate judgment results. Furthermore, the data processing unit performs frequency analysis on time-series data, limited to time intervals set according to the type of abnormal sound, or to frequency bands set according to the type of abnormal sound. This reduces the processing burden for frequency analysis and judgment, while effectively detecting abnormal sounds. [Explanation of Symbols]

[0048] 1. Inspection device 11 Data Acquisition Unit 12 Data Processing Unit 13 Judgment section

Claims

1. An inspection device for inspecting motor devices, A data acquisition unit that acquires time-series data of sound pressure or vibration waveform based on sound or vibration generated during the operation of the motor device, A data processing unit performs frequency analysis on the time-series data acquired by the data acquisition unit, A determination unit determines the evaluation category of the motor device based on the results of frequency analysis obtained by the data processing unit, Equipped with, The data processing unit performs frequency analysis on the time series data, limiting it to time intervals set according to the type of abnormal noise determined by the number of slots or poles of the motor in the motor device, or to frequency bands set according to the type of abnormal noise. It calculates the partial overall value as a feature quantity combined with the target interval on the time axis, and outputs the motor current value to the determination unit as time series data aligned with the time series data on the time axis. The determination unit determines the evaluation category based on the auditory perception of the operating sound of the motor device and makes a determination based on the results of the frequency analysis using an algorithm based on machine learning. The inspection device uses the results of the frequency analysis along with information relating to the current value of the motor as a feature to determine the evaluation category of the motor device.

2. The inspection apparatus according to claim 1, wherein data containing disturbances is used as the time-series data during the learning process for the determination unit.

3. The inspection apparatus according to claim 1 or 2, wherein the determination unit performs determination using a random forest.

4. A method for inspecting a motor device, A data acquisition step of acquiring time-series data of sound pressure or vibration waveform based on sound or vibration generated during the operation of the motor device, A data processing step in which frequency analysis is performed on the time series data acquired in the data acquisition step, A determination step to determine the evaluation category of the motor device based on the results of the frequency analysis obtained in the above data processing step, The computer executes this, In the data processing step, frequency analysis is performed on the time series data, limited to time intervals set according to the type of abnormal noise determined by the number of slots or poles of the motor in the motor device, or to frequency bands set according to the type of abnormal noise. The partial overall value is calculated as a feature quantity combined with the target interval on the time axis, and the motor current value is output as time series data aligned with the time series data on the time axis. The determination step is a step of making a determination based on the evaluation category determined based on the auditory perception of the operating sound of the motor device and the results of the frequency analysis using an algorithm based on machine learning, and the determination of the evaluation category of the motor device is made by using information relating to the current value of the motor as a feature quantity in conjunction with the results of the frequency analysis.

Images