Learning processing apparatus, state monitoring device, learning processing method, state monitoring method, and program

Abstract

To enable state monitoring of a rolling bearing without using rotation speed information.SOLUTION: A learning processing apparatus acquires the rotation speed of a rotating rolling bearing, derives the timing to perform sampling from acquired vibration data or sound data according to the rotation speed so as to adjust the amount of sampling per revolution of the rolling bearing to a predetermined value, performs sampling from the acquired vibration data or sound data on the basis of the derived timing to generate sampling data, associates the acquired vibration data or sound data with the sampling data from the vibration data or sound data to generate learning data, and performs learning processing, by using a predetermined learning algorithm, with the vibration data or sound data in the generated learning data as input, so as to bring output closer to the sampling data corresponding to the vibration data or sound data, thereby generating a learned model.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to a learning processing device, a state monitoring device, a learning processing method, a state monitoring method, and a program. [Background technology]

[0002] Conventionally, in mechanical devices such as wind turbines, the condition of the rolling bearings installed therein has been monitored and control has been performed according to the condition in order to prevent malfunctions of the mechanical devices and ensure more appropriate operation. Information used for monitoring the condition of rolling bearings includes vibration, sound, rotation speed, and the like.

[0003] Patent Document 1 discloses a configuration in which sampling is performed from measured vibration information while taking into account changes in the rotation speed of a rolling bearing, and the sampled data is used to monitor the condition. Patent Document 1 derives the sampling timing by considering instantaneous changes in the rotation speed during one rotation of a shaft connected to the rolling bearing. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2023-9580 Summary of the Invention [Problem to be solved by the invention]

[0005] In the configuration of Patent Document 1, a rotation speed sensor is installed to measure the rotation speed of the rolling bearing, and the condition is monitored using the rotation speed detected by the rotation speed sensor. Therefore, the rotation speed sensor is essential for condition monitoring. However, there may be cases where the rotation speed sensor is difficult to install due to certain restrictions on its installation, or the rotation speed sensor itself is not operating normally, and so on, and therefore it may not be possible to obtain and use rotation speed information during desired condition monitoring. In such cases, the method of Patent Document 1 cannot be used as is.

[0006] In view of the above problems, an object of the present invention is to enable monitoring of the condition of a rolling bearing without using rotational speed information, regardless of whether or not there is a change in the rotational speed of the rolling bearing. [Means for solving the problem]

[0007] In order to solve the above problems, the present invention has the following configuration. That is, a learning processing device includes: A first acquisition means for acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition means for acquiring a rotational speed of the rolling bearing during rotation; a derivation means for deriving a sampling timing from the vibration data or the sound data acquired by the first acquisition means in accordance with the rotation speed acquired by the second acquisition means so that the number of samples per rotation of the rolling bearing becomes a predetermined value; a sampling processing means for generating sampling data by sampling the vibration data or the sound data acquired by the first acquisition means based on the timing derived by the derivation means; A generating means for generating learning data by associating the vibration data or sound data acquired by the first acquiring means with sampling data from the vibration data or sound data; A learning means for generating a trained model by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated by the generating means as input, so that an output approaches sampling data corresponding to the vibration data or sound data; has.

[0008] Another aspect of the present invention has the following configuration. That is, the status monitoring device includes: An acquisition means for acquiring vibration data or sound data of the rolling bearing during rotation; A generation means for generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired by the acquisition means into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring means for monitoring a state of the rolling bearing by using the sampling data generated by the generating means; has.

[0009] Another aspect of the present invention has the following configuration. That is, a learning processing method includes: A first acquisition step of acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition step of acquiring a rotational speed of the rolling bearing during rotation; a derivation step of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotational speed acquired in the second acquisition step so that the number of samples per rotation of the rolling bearing becomes a predetermined value; a sampling process step of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition step based on the timing derived in the derivation step; a generating step of generating learning data by associating the vibration data or sound data acquired in the first acquiring step with sampling data from the vibration data or sound data; A learning process in which a trained model is generated by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that an output approaches sampling data corresponding to the vibration data or sound data; has.

[0010] Another aspect of the present invention has the following configuration. That is, a status monitoring method includes: an acquisition step of acquiring vibration data or sound data of the rolling bearing during rotation; A generation process of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring step of monitoring a state of the rolling bearing by using the sampling data generated in the generating step; has.

[0011] Another aspect of the present invention has the following configuration. That is, a program comprising: On the computer, A first acquisition step of acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition step of acquiring a rotational speed of the rolling bearing during rotation; a derivation step of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotational speed acquired in the second acquisition step so that the number of samples per rotation of the rolling bearing is a predetermined value; a sampling process step of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition step based on the timing derived in the derivation step; a generating step of generating learning data by associating the vibration data or sound data acquired in the first acquiring step with sampling data from the vibration data or sound data; A learning process in which a trained model is generated by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that an output approaches sampling data corresponding to the vibration data or sound data; Execute the command.

[0012] Another aspect of the present invention has the following configuration. That is, a program comprising: On the computer, an acquisition step of acquiring vibration data or sound data of the rolling bearing during rotation; A generation process of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring step of monitoring a state of the rolling bearing by using the sampling data generated in the generating step; Execute the command. Effect of the Invention

[0013] According to the present invention, it is possible to monitor the condition of a rolling bearing without using rotational speed information, regardless of whether or not there is a change in the rotational speed of the rolling bearing. [Brief description of the drawings]

[0014] [Figure 1] FIG. 2 is a schematic diagram for explaining sampling according to the present invention. [Diagram 2] FIG. 2 is a schematic diagram showing an example of an apparatus configuration in a learning phase according to an embodiment of the present invention. [Diagram 3] FIG. 2 is a schematic diagram showing an example of an apparatus configuration in a state monitoring phase according to an embodiment of the present invention. [Figure 4] FIG. 4 is a conceptual diagram for explaining a process flow of a learning phase according to an embodiment of the present invention. [Diagram 5]4 is a flowchart of a learning process according to an embodiment of the present invention. [Figure 6] 4 is a flowchart of a state monitoring process according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment for carrying out the present invention will be described with reference to the drawings. Note that the embodiment described below is one embodiment for explaining the present invention, and is not intended to be interpreted as limiting the present invention, and all configurations described in each embodiment are not necessarily essential configurations for solving the problems of the present invention. In addition, in each drawing, the same components are given the same reference numbers to indicate their correspondence.

[0016] In the following description, "learning" or "machine learning" refers to generating a "trained model" by repeatedly performing a learning process using training data and an arbitrary learning algorithm. The trained model is updated from time to time as learning progresses using multiple training data, and its output changes even for the same input. Therefore, the trained model is not limited to a state at any point in time. Here, a model used in learning is referred to as a "trained model," and a learning model that has undergone a certain degree of learning is referred to as a "trained model."

[0017] Furthermore, specific examples of "learning data" will be described later, but the configuration may be adjusted or changed depending on the learning algorithm used and its application. For example, in addition to vibration data described later, sound data, ultrasonic data, numerical data, and other data may be used as the learning data. Furthermore, the configuration of the learning data and preprocessing of the learning data may change depending on the type of learning algorithm and the contents of the input and output of the learning model.

[0018] The learning data may include teacher data used for learning itself, verification data used for verifying a trained model, and test data used for testing a trained model. In the following description, when data related to learning is collectively indicated, it is written as "learning data". Note that it is not intended to clearly classify the teacher data, verification data, and test data contained in the learning data, and for example, depending on the methods of learning, verification, and testing, all of the learning data may also be teacher data.

[0019] <First embodiment> A first embodiment of the present invention will be described below.

[0020] [Sampling process] The rotational speed of rolling bearings installed in mechanical devices such as wind power generation equipment can fluctuate intermittently due to external factors (for example, wind). When the rotational speed fluctuates, the vibration frequency of each part of the rolling bearing also fluctuates, and the frequency that serves as the judgment standard for use in condition monitoring cannot be determined. This affects the accuracy of monitoring and diagnosis, for example, when the condition of a rolling bearing is monitored using vibration information. In addition, since the rotational speed of wind power generation equipment is relatively slow, there is a high possibility that the rotational speed will fluctuate even during one rotation.

[0021] A sampling process assuming the above-mentioned change in rotation speed will be described. Fig. 1 is a schematic diagram for explaining data sampling according to this embodiment. In Fig. 1, the vertical axis represents vibration value and the horizontal axis represents time. The vibration value corresponds to the value of an electrical signal detected by a vibration sensor and subjected to A / D conversion.

[0022] FIG. 1 shows two sections as an example. The first section shows an example in which the rotation speed of the rolling bearing is relatively slower than the second section. That is, the rotation speed of the rolling bearing fluctuates. As shown in FIG. 1, the sampling timing is different between the first section and the second section. This is because the sampling period is changed according to the rotation speed so that the number of samples (number of data) per rotation is the same. That is, the sampling timing is adjusted so that the number of samples from the vibration information is the same in one rotation period, taking into account the fluctuation in the rotation speed of the rolling bearing. The circle shown in the graph of FIG. 1 indicates the position of the data to be sampled. In this way, the sampling data is generated by sampling data based on the rotation speed from the original vibration information. The sampling timing is derived so that the number of samples per rotation is constant, for example, using a sampling clock synchronized with the rotation speed. The derivation method may be derived using a predetermined calculation formula, or may be derived using a table in which the rotation speed and the sampling timing (time width, etc.) are associated with each other.

[0023] In the example of FIG. 1, the rotation speed changes after one rotation in the first section, and one rotation in the second section is performed. In reality, the rotation speed may vary during one rotation, so the timing of sampling may vary during one rotation. Although FIG. 1 shows an example in which the rotation changes from a slow rotation state to a fast rotation state, the reverse may also occur. Furthermore, depending on the situation, there may be cases in which the rotation speed does not change within a predetermined time or within a predetermined period. In other words, the rotation speed of the rolling bearing may not change or may change irregularly depending on the operating environment of the device in which the rolling bearing is provided.

[0024] [Device configuration] An embodiment of a device to which the method according to the present embodiment can be applied will be described below. The method according to the present embodiment can be broadly divided into a learning phase and a condition monitoring phase. The learning phase is a phase in which learning data is generated using data detected from the rolling bearing, and a trained model is generated by performing a learning process using the learning data. The condition monitoring phase is a phase in which the condition of the rolling bearing is monitored using the trained model generated in the learning phase. Since the device configurations applicable to each phase are different, each will be described separately.

[0025] The method according to the present embodiment can be applied to a wind power generating device including a rolling bearing, but is not limited thereto. The method according to the present embodiment can be applied to any device in which the rotation speed may fluctuate during rotation as described above. For example, this includes a device that includes a rolling bearing and has a relatively slow rotation speed, or a device that has a large fluctuation in the number of rotations. Here, the following description focuses on the rolling bearing and its surroundings, which are the object of monitoring, within such a device.

[0026] Fig. 2 shows an example of the device configuration in the learning phase, and Fig. 3 shows an example of the device configuration in the state monitoring phase. Note that in both cases, the rolling bearing 110 to be monitored has the same configuration.

[0027] (Device configuration in the learning phase) 2 is a schematic diagram showing an example of an apparatus configuration in the learning phase according to this embodiment, which shows a bearing apparatus 100 including a rolling bearing 110 to be monitored, and a learning processing apparatus 200 that performs learning processing.

[0028] The learning processing device 200 may be configured integrally with the bearing device 100 including the rolling bearing 110, or may be provided as a separate device. The bearing device 100 may be, for example, a wind power generation device or a compressor as described above.

[0029] The rolling bearing 110 includes an inner ring 111 which is a rotating ring fitted on the outside of the main shaft 120, an outer ring 113 which is a fixed ring fitted on the inside of a housing (not shown), a plurality of balls (rollers) which are a plurality of rolling elements 112 arranged between the inner ring 111 and the outer ring 113, and a cage 114 which holds the rolling elements 112 so that they can roll. In addition, in the rolling bearing 110, friction between the inner ring 111 and the rolling elements 112 and between the outer ring 113 and the rolling elements 112 is reduced by a predetermined lubrication method. The lubrication method is not particularly limited, but for example, grease lubrication or oil lubrication is used. In addition, the type of lubricant is not particularly limited. In addition, the rolling bearing 110 can be applied to, for example, a tapered roller bearing, a cylindrical roller bearing, etc., but is not particularly limited. The rolling bearing 110 rotatably supports the main shaft 120.

[0030] A vibration sensor 130 is provided to detect vibrations generated from the rolling bearing 110 while the main shaft 120 is rotating. The vibration sensor 130 is fixed near the outer ring of the housing by bolting, gluing, bolting and gluing, or embedding with a molding material. When the vibration sensor 130 is fixed by bolts, it may be provided with a rotation prevention function. The vibration sensor 130 is not limited to a configuration in which it is fixed and installed at a detection position, but may be installed at a position for detecting vibrations caused by the rolling bearing 110 when acquiring measurement data. Therefore, the vibration sensor 130 may be configured to be detachable or movable.

[0031] The vibration sensor 130 may be any sensor capable of detecting vibration, such as an acceleration sensor, an AE (Acoustic Emission) sensor, an ultrasonic sensor, or a shock pulse sensor, as long as it can convert detected acceleration, speed, strain, stress, displacement, or other vibrations into an electrical signal. When mounting the sensor on a device located in a noisy environment (e.g., a wind power generation device), it is more preferable to use an insulated type sensor, since it is less susceptible to noise. Furthermore, when the vibration sensor 130 uses a vibration detection element such as a piezoelectric element, the element may be molded in plastic or the like.

[0032] In addition, in the learning phase, the rolling bearing 110 is provided with a rotational speed sensor 150 that detects the rotational speed of the inner ring 111 fitted on the main shaft 120 or the main shaft 120. In this embodiment, the rotational speed and number of revolutions of the inner ring 111, which is a rotating ring, and the main shaft 120 are the same. The rotational speed of the main shaft 120 may vary depending on the load and torque received by the bearing device 100. For example, in the case of a wind power generation device, the rotational speed of the main shaft 120 may vary depending on the wind direction, wind volume, and wind pressure. Furthermore, the rotational speed may be adjusted by a brake device (not shown). The rotational speed sensor 150 may detect the rotational speed by detecting, for example, an encoder (not shown) provided on the inner ring 111 of the rolling bearing 110. For example, a wind power generation device or the like rotates at a relatively low rotational speed. Therefore, the rotational speed sensor 150 is configured to be able to detect the change in the rotational speed during one rotation of the rolling bearing 110. The rotational speed sensor 150 detects the interval between timings at which the signal strength of the pulse signal accompanying the rotation of the rolling bearing 110 changes, and detects the rotational speed based on the timing of the changes. For example, the number of changes in the pulse signal detected during one rotation of the rolling bearing 110 is specified in advance, and the rotational speed is derived from the number of changes in the pulse signal and the time at which the pulse signal is detected. Note that the acquisition of the rotational speed is not limited to the rotational speed sensor. For example, the rotational speed sensor 150 may be configured to acquire the rotational speed from the vibration frequency generated by the meshing of gears.

[0033] The amplifier 140 amplifies the electrical signal detected by the vibration sensor 130 and inputs it to the learning processing device 200. The degree of amplification here is not particularly limited, but is defined in advance. The detection timings of the vibration sensor 130 and the rotation speed sensor 150 correspond to each other, and their measurement information is stored and processed in association with each other.

[0034] The learning processing device 200 may be realized by an information processing device including, for example, a control device, a storage device, and an input / output device (not shown). The control device may be composed of a central processing unit (CPU), a graphical processing unit (GPU), a general-purpose computing on graphics processing units (GPGPU), a micro processing unit (MPU), a digital signal processor (DSP), a field programmable gate array (FPGA), or a dedicated circuit. The storage device is composed of volatile and non-volatile storage media such as a hard disk drive (HDD), a read only memory (ROM), and a random access memory (RAM), and is capable of inputting and outputting various information in response to an instruction from the control device. The input / output device transmits and receives data to and from an external device and notifies an operator in response to an instruction from the control device. The input / output method by the input / output device is not particularly limited. For example, the output method may be an auditory notification by voice, or a visual notification by screen output. The input / output device may be a network interface equipped with a communication function, and may perform various input / output operations by transmitting and receiving data to and from an external device (not shown) via a network (not shown).

[0035] The learning processing device 200 includes a data acquisition unit 201, a sampling processing unit 202, a learning data generation unit 203, a learning processing unit 204, and a trained model management unit 205. These units may be realized by a control device (not shown) included in the learning processing device reading out corresponding programs from a storage device (not shown) and executing them.

[0036] The data acquisition unit 201 acquires measurement information measured by the vibration sensor 130 and the rotation speed sensor 150, and stores the information in association with the measurement time. The data acquisition unit 201 acquires electrical signals measured by the various sensors via the amplifier 140, and performs A / D (Analog / Digital) conversion according to the contents of the electrical signals.

[0037] The sampling processing unit 202 samples data to be used as learning data (described later) from the vibration data based on the rotation speed among various data acquired by the data acquisition unit 201. The sampling method here is adjusted so that the number of samples is constant in a predetermined period based on the rotation speed of the rolling bearing 110, as described in Fig. 1.

[0038] The learning data generating unit 203 generates learning data by associating the vibration data acquired by the data acquiring unit 201 (hereinafter also referred to as "first data") with the sampling data processed by the sampling processing unit 202 (hereinafter also referred to as "second data"). The learning data here may be generated for each predetermined time interval.

[0039] The learning processing unit 204 performs a learning process using the learning data generated by the learning data generating unit 203. In this embodiment, when the vibration data acquired by the data acquiring unit 201 is input, the learning process is performed so that the output approaches the sampling data associated with the input vibration data. The learning algorithm used by the learning processing unit 204 is not particularly limited. A learning algorithm using so-called supervised learning may be used, and for example, a method such as feature extraction using deep learning may be used. For example, a learned model can be generated by teacher data in which a vibration waveform before order ratio analysis is input and a vibration waveform obtained by performing order ratio analysis on the vibration waveform is output. In this case, when an arbitrary vibration waveform is input to the data acquiring unit 301 in the state monitoring phase described later, the state estimating unit 303 can obtain a vibration waveform obtained by performing order ratio analysis on the arbitrary vibration waveform based on the acquired arbitrary vibration waveform and the learned model. That is, order ratio analysis can be performed without using a rotation sensor. The learning process by the learning processing unit 204 is repeated to generate a learned model.

[0040] The trained model management unit 205 manages the trained models generated by the learning processing unit 204. Here, the trained models may be managed according to the type or configuration of the rolling bearing 110 to be monitored, or versions may be managed according to the level of learning.

[0041] Since the learning process generally involves a large processing load, it is preferable that the learning phase and the state monitoring phase, which will be described later, are executed at separate times. It is also preferable that the learning processing device 200 in the learning phase and the state monitoring device 300 in the state monitoring phase, which will be described later, are implemented in separate devices. In this case, in the state monitoring phase, the state monitoring device 300 is configured to be able to use a trained model that is generated and managed by the learning processing device 200. The data and processing flow of the learning processing device 200 in the learning phase will be described later with reference to FIG. 4 and the like.

[0042] 2 shows a configuration in which data is measured for one rolling bearing 110 and one learning processing device 200 performs learning processing using this measurement data for the sake of simplicity of explanation. However, the present invention is not limited to this configuration, and the learning processing may be performed by acquiring data from multiple rolling bearings 110. Furthermore, separate learning processes may be performed in accordance with the type and configuration of the rolling bearing 110.

[0043] (Device configuration in the status monitoring phase) Fig. 3 is a schematic diagram showing an example of an apparatus configuration in the condition monitoring phase according to this embodiment. Shown are a bearing device 100 equipped with a rolling bearing 110 to be monitored, a learning processing device 200 that provides a trained model, and a condition monitoring device 300 that performs condition monitoring processing. Therefore, the overall configuration shown in Fig. 3 functions as a monitoring system for the bearing device 100.

[0044] In the condition monitoring phase according to this embodiment, the rotation speed sensor 150 is not used. This may be the case, for example, when the rotation speed sensor 150 cannot be installed, or when rotation speed data cannot be obtained from the rotation speed sensor 150 due to an obstacle or the like. The types and configurations of the rolling bearing 110 and the vibration sensor 130 are the same as those in the learning phase.

[0045] The state monitoring device 300 may be realized by, for example, an information processing device including a control device, a storage device, and an input / output device (not shown). The hardware configuration of the state monitoring device 300 may be the same as or different from the hardware configuration of the learning processing device 200. In this embodiment, the state monitoring device 300 and the learning processing device 200 are connected to each other via a network so as to be able to communicate with each other.

[0046] The state monitoring device 300 includes a data acquisition unit 301, a trained model acquisition unit 302, a state estimation unit 303, and a state output unit 304. These units may be realized by a control device (not shown) included in the state monitoring device 300 reading out and executing corresponding programs from a storage device (not shown).

[0047] The data acquisition unit 301 acquires measurement information measured by the vibration sensor 130 and stores the information in association with the measurement time. The data acquisition unit 301 acquires electrical signals measured by various sensors via the amplifier 140 and performs A / D (Analog / Digital) conversion according to the contents of the electrical signals.

[0048] The trained model acquisition unit 302 acquires a trained model generated corresponding to the rolling bearing 110 to be monitored from the learning processing device 200. Since it is expected that the trained model may be updated by appropriately executing the learning process, the condition monitoring device 300 may be configured to acquire the trained model in the latest state.

[0049] The state estimation unit 303 estimates the state of the rolling bearing 110 using the vibration data acquired by the data acquisition unit 301 and the trained model acquired by the trained model acquisition unit 302. First, the state estimation unit 303 acquires data for estimation by applying the vibration data acquired by the data acquisition unit 301 to the trained model acquired by the trained model acquisition unit 302. This data for estimation corresponds to the sampling data used in the learning process, and corresponds to a state similar to that to which the sampling process has been applied.

[0050] Furthermore, the state estimation unit 303 performs signal analysis processing using data output from the trained model. In the signal analysis processing, FFT (Fast Fourier Transform) analysis is performed, followed by order ratio analysis. The signal analysis processing may be performed by envelope processing, or filtering processing using a low-pass filter, a band-pass filter, or the like. The state estimation unit 303 estimates the state of the rolling bearing 110 using the data that has been subjected to the signal analysis processing. The state estimation unit 303 may process all data to which the trained model has been applied, or may extract a part of the data and estimate the state using the extracted data. The items of state monitoring performed by the state estimation unit 303 are not particularly limited, and may be any items such as abnormal contact, poor lubrication, and damage or deterioration of parts of each part that constitutes the rolling bearing 110. It should be noted that a publicly known method can be used to determine the presence or absence and the degree of abnormality. For example, the normality or abnormality of the bearing can be determined based on the signal strength at a specific frequency in the above-mentioned signal diffraction processing, but is not limited thereto.

[0051] The state output unit 304 outputs the state estimated by the state estimation unit 303. The output by the state output unit 304 may be notified to the outside via a network (not shown), or may be used to control the operation of the rolling bearing 110. In this case, the operation of the rolling bearing 110 may be controlled, for example, by a brake mechanism (not shown) so that the rotation speed of the spindle 120 becomes a predetermined speed or so that the spindle 120 stops.

[0052] 3 shows a configuration in which data is measured for one rolling bearing 110 and one condition monitoring device 300 performs condition monitoring processing using this measurement data for the sake of simplicity of explanation. However, the present invention is not limited to this configuration, and a configuration may also be adopted in which data is acquired from a plurality of rolling bearings 110 and the conditions of each rolling bearing 110 are monitored by a single condition monitoring device 300. In this case, the condition monitoring device 300 acquires a trained model in which learning has been performed corresponding to each of the plurality of rolling bearings 110 and performs processing.

[0053] 2 and 3 are merely examples, and one block may be divided into multiple blocks, or multiple blocks may be integrated into one. Also, each function does not have to be realized by one device, and may be realized by multiple devices working together.

[0054] In the learning phase, the learning processing device 200 may be realized as a cloud service established on a network, for example, the Internet. In this case, the state monitoring device 300 may be configured to obtain and use the trained model from the cloud service on the network.

[0055] [Learning process] 4 is a conceptual diagram for explaining the flow of data and processing in the learning phase according to this embodiment. In this embodiment, the configuration shown in FIG.

[0056] The rotation speed data 401 is data obtained by the rotation speed sensor 150, and is configured by associating rotation speed with time. The vibration data 402 is data obtained by the vibration sensor 130, and is configured by associating acceleration with time. It is assumed that the time ranges of the rotation speed data 401 and the vibration data 402 are the same.

[0057] First, as described with reference to Fig. 1, sampling process S410 is executed using rotation speed data 401 and vibration data 402 to generate sampling data 403 that takes into account changes in rotation speed. The sampling data 403 is a part of the vibration data 402, and acceleration and time are associated with each other. Then, the vibration data 402 before the sampling process and the sampling data 403 obtained by the sampling process are associated with each other to generate learning data 404.

[0058] In the learning process S420, the learning process is repeated using the generated learning data 404 and a predetermined learning algorithm to generate a trained model 405. As described above, the number of trained models 405 is not limited to one, and multiple trained models may be generated and managed depending on the type of the rolling bearing 110 to be monitored and the level of learning.

[0059] [Processing flow] (Learning Phase) Fig. 5 is a flowchart of a learning process in the learning phase according to this embodiment. This process is executed by the learning processing device 200 in the configuration example shown in Fig. 2. For example, the learning processing device 200 may be realized by a control device (not shown) included in the learning processing device 200 reading out a program for implementing each part shown in Fig. 2 from a storage device (not shown) and executing the program.

[0060] In S501, the learning processing device 200 acquires vibration data of the rolling bearing 110 detected by the vibration sensor .

[0061] In S502, the learning processing device 200 stores the vibration data acquired in 2 in a storage unit (not shown). Then, the process of the learning processing device 200 proceeds to S505.

[0062] In S503, the learning processing device 200 acquires rotation speed data of the rolling bearing 110 detected by the rotation speed sensor 150. The vibration data acquired in S501 and the rotation speed data acquired in S503 correspond in detection timing.

[0063] In S504, the learning processing device 200 derives a sampling timing for the vibration data acquired in S501 based on the rotation speed acquired in S503. Note that the acquisition of the vibration data in S501 and the acquisition of the rotation speed data in S503 do not necessarily need to be performed in real time during the learning process. In other words, the processes of S501 and S503 may be configured to acquire the vibration data and the rotation speed data from pre-stored data.

[0064] In S505, the learning processing device 200 samples data from the vibration data based on the timing derived in S504, and generates sampling data.

[0065] In S506, the learning processing device 200 generates learning data by associating the vibration data acquired in S501 with the sampling data generated in S505.

[0066] In S507, the learning processing device 200 determines whether a predetermined amount of learning data not used in the learning process has been accumulated. The predetermined amount here may be predefined and set so as to increase the efficiency of the learning process. If the predetermined amount of learning data has been accumulated (YES in S507), the process of the learning processing device 200 proceeds to S508. If the predetermined amount of learning data has not been accumulated (NO in S507), the process of the learning processing device 200 returns to data acquisition (S501, S503) and repeats the process.

[0067] In S508, the learning processing device 200 performs a learning process using the generated learning data and a predetermined learning algorithm. In the learning process here, vibration data is input, and learning is performed so that the output approaches sampling data corresponding to the vibration data, thereby generating a learned model.

[0068] In S509, the learning processing device 200 stores the learned model generated in S508 in a storage device (not shown), and then ends this processing flow.

[0069] 4, the process flow of the learning phase shows an example in which the generation of learning data (S501 to S506) and learning using the learning data (S507 to S509) are performed in a single sequence, but the present invention is not limited to this. The above processes may be performed at separate times.

[0070] (Condition monitoring phase) Fig. 6 is a flowchart of the state monitoring process in the state monitoring phase according to this embodiment. This process is executed by the state monitoring device 300 in the configuration example shown in Fig. 3. For example, this may be realized by a control device (not shown) included in the state monitoring device 300 reading out from a storage device (not shown) and executing a program for implementing each part shown in Fig. 3. This process flow may be executed continuously while the rolling bearing 110 to be monitored continues to operate, or may be executed at a predetermined timing based on an instruction from a user.

[0071] In S601, the condition monitoring device 300 acquires a trained model corresponding to the state monitoring target from the learning processing device 200. The trained model may be acquired by the condition monitoring device 300 making a request to the learning processing device 200 using information on the rolling bearing 110 that is the state monitoring target, or may be acquired by the learning processing device 200 referring to a trained model stored in a predetermined storage area.

[0072] In S602, the condition monitoring device 300 acquires vibration data of the rolling bearing 110 detected by the vibration sensor 130. The vibration data is acquired here in the same manner as in the step of S501 in Fig. 5 in the learning phase.

[0073] In S603, the state monitoring device 300 applies the vibration data acquired in S602 to the trained model acquired in S601. As a result, data corresponding to sampling data obtained by extracting a portion of the vibration data is acquired.

[0074] In S604, the state monitoring device 300 performs signal analysis processing using the sampling data acquired in S603. As the signal analysis processing, a process of deriving a vibration value corresponding to each vibration frequency is performed using the sampling data. At this time, the signal analysis processing may be performed after performing envelope processing, or LPF or BPF filter processing.

[0075] In S605, the condition monitoring device 300 determines a vibration frequency to be focused on from among the theoretical frequency of the rolling bearing 110 and its higher-order vibration frequencies based on the result of the analysis process in S604. The correspondence relationship between the vibration frequency to be focused on and the rotation speed is specified in advance, and the vibration frequency to be focused on is determined based on this correspondence. At this time, one or more vibration frequencies may be focused on.

[0076] In S606, the status monitoring device 300 notifies the monitoring result of S605. The notification method is not particularly limited, and the notification may be visually or audibly to the manager of the monitored object. Alternatively, instead of notification, the configuration may be such that the operation of the monitored object is controlled based on the monitoring result. Then, this processing flow ends.

[0077] As described above, this embodiment makes it possible to monitor the condition of a rolling bearing without using rotational speed information, regardless of whether or not there is a change in the rotational speed of the rolling bearing. In particular, even in a situation where a rotational speed sensor cannot be used, it is possible to generate data that assumes fluctuations in rotational speed and use the data to perform condition monitoring.

[0078] <Other embodiments> In the configuration example of FIG. 3 of the above embodiment, a configuration that does not include the rotation speed sensor 150 during state monitoring is shown. However, the present invention is not limited to this configuration, and the rotation speed sensor 150 may be included. When the rotation speed sensor 150 is operating normally, sampling data is generated based on the rotation speed detected by the rotation speed sensor 150. On the other hand, when the rotation speed sensor 150 is not operating normally (due to a malfunction or communication failure, etc.), processing may be switched to generation of sampling data using the above-mentioned trained model. In this case, it is determined whether the rotation speed can be acquired normally, and the state monitoring method is switched depending on the result.

[0079] The learning processing device 200 may be configured to collect measurement data from various sensors as appropriate, generate learning data, and update the learned model. The updated learned model may then be provided to one or more condition monitoring devices 300 that monitor the same type of rolling bearings 110 to be monitored.

[0080] In the above embodiment, the vibration sensor 130 is used to detect the vibration of the rolling bearing 110, but the present invention is not limited to this. Sound information may be detected using a sound sensor including a microphone instead of the vibration sensor 130. In this case, sampling data is extracted from the sound information, learning data is generated, and then a learned model for the state monitoring process is generated.

[0081] In addition, in the present invention, a program or application for realizing the functions of one or more of the above-mentioned embodiments can be supplied to a system or device via a network or a storage medium, etc., and one or more processors in a computer of the system or device can read and execute the program, thereby realizing the present invention.

[0082] Also, a circuit that realizes one or more functions (e.g., ASIC (Application Specific Integrated Circuit) pecific Integrated Circuit) and FPGA (Field P This may be realized by a programmable gate array.

[0083] In addition, when the terms "first" and "second" are used in the description of this specification, they are merely used for convenience to distinguish from other elements, and are not intended to be interpreted as being limited to specific elements. Therefore, these expressions should be interpreted appropriately depending on the combination and number of components.

[0084] As such, the present invention is not limited to the above-described embodiments, and the present invention also contemplates mutual combinations of the various components of the embodiments, as well as modifications and applications by those skilled in the art based on the descriptions in the specification and well-known technologies, and these are included in the scope of protection sought.

[0085] As described above, the present specification discloses the following: (1) a first acquisition means (e.g., 130, 140, 201) for acquiring vibration data or sound data (e.g., 402) of a rotating rolling bearing (e.g., 110); A second acquisition means (e.g., 150, 201) for acquiring a rotational speed (e.g., 401) of the rolling bearing during rotation; a derivation means (e.g., 202) that derives sampling timing from the vibration data or sound data acquired by the first acquisition means in accordance with the rotation speed acquired by the second acquisition means so that the number of samples per rotation of the rolling bearing becomes a predetermined value; A sampling processing means (e.g., 202) that generates sampling data (e.g., 403) by sampling the vibration data or the sound data acquired by the first acquisition means based on the timing derived by the derivation means; A generating means (e.g., 203) that generates learning data (e.g., 404) by associating the vibration data or sound data acquired by the first acquiring means with sampling data from the vibration data or sound data; A learning means (e.g., 204) that uses a predetermined learning algorithm, inputs vibration data or sound data (e.g., 402) from the learning data generated by the generating means, and performs a learning process so that the output approaches sampling data (e.g., 403) corresponding to the vibration data or sound data, thereby generating a trained model (e.g., 405); A learning processing device (e.g., 200) having the following: According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, it is possible to generate a trained model for generating monitoring data that assumes fluctuations in rotational speed even in a situation where a rotational speed sensor cannot be used.

[0086] (2) an acquisition means (e.g., 130, 140, 301) for acquiring vibration data or sound data of a rotating rolling bearing (e.g., 110); A generation means (e.g., 303) for generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired by the acquisition means into a trained model (e.g., 405) that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; A monitoring means (e.g., 303) for monitoring the state of the rolling bearing using the sampling data generated by the generating means; A condition monitoring device having the following: According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, even in a situation where a rotational speed sensor cannot be used, it is possible to generate data that assumes fluctuations in rotational speed and perform rolling bearing condition monitoring.

[0087] (3) a second acquisition means for acquiring a rotational speed of the rolling bearing during rotation; a determination means for determining whether or not the rotation speed can be acquired by the second acquisition means; a derivation means for deriving a sampling timing from the vibration data or the sound data acquired by the acquisition means in accordance with the rotational speed acquired by the second acquisition means, when the determination means determines that the rotational speed can be acquired, so that the number of samples per rotation of the rolling bearing becomes a predetermined value; a sampling processing means for generating sampling data by sampling the vibration data or the sound data acquired by the acquisition means based on the timing derived by the derivation means; Further comprising: The condition monitoring device described in (2), wherein the monitoring means monitors the condition of the rolling bearing using sampling data generated by the sampling processing means when the determination means determines that the rotational speed can be acquired. According to this configuration, it is possible to switch the generation method for state monitoring depending on whether the rotation speed can be acquired or not, thereby enabling state monitoring according to the situation.

[0088] (4) A monitoring system for a rolling bearing, comprising the learning processing device according to (1) and the condition monitoring device according to (2) or (3). According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, even in a situation where a rotational speed sensor cannot be used, it is possible to generate data that assumes fluctuations in rotational speed and perform rolling bearing condition monitoring.

[0089] (5) a first acquisition step (e.g., S501) of acquiring vibration data or sound data of a rotating rolling bearing (e.g., 110); A second acquisition step (e.g., S503) of acquiring a rotational speed of the rolling bearing during rotation; a derivation step (e.g., S504) of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotation speed acquired in the second acquisition step so that the number of samples per rotation of the rolling bearing is a predetermined value; A sampling process (e.g., S505) of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition process based on the timing derived in the derivation process; A generation step (e.g., S506) of generating learning data by associating the vibration data or sound data acquired in the first acquisition step with sampling data from the vibration data or sound data; A learning process (e.g., S508) for generating a trained model by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that the output approaches sampling data corresponding to the vibration data or sound data; A learning processing method having the following steps. According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, it is possible to generate a trained model for generating monitoring data that assumes fluctuations in rotational speed even in a situation where a rotational speed sensor cannot be used.

[0090] (6) an acquisition step (e.g., S602) of acquiring vibration data or sound data of the rolling bearing during rotation; A generation process (e.g., S603) of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model (e.g., 405) that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; A monitoring step (e.g., S604, S605) of monitoring a state of the rolling bearing using the sampling data generated in the generating step; A condition monitoring method comprising: According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, even in a situation where a rotational speed sensor cannot be used, it is possible to generate data that assumes fluctuations in rotational speed and perform rolling bearing condition monitoring.

[0091] (7) A computer (e.g., 200), A first acquisition step (e.g., S501) of acquiring vibration data or sound data of a rotating rolling bearing (e.g., 110); A second acquisition step (e.g., S503) of acquiring a rotational speed of the rolling bearing during rotation; a derivation step (e.g., S504) of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotation speed acquired in the second acquisition step so that the number of samples per rotation of the rolling bearing is a predetermined value; A sampling process (e.g., S505) of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition process based on the timing derived in the derivation process; A generation step (e.g., S506) of generating learning data by associating the vibration data or sound data acquired in the first acquisition step with sampling data from the vibration data or sound data; A learning process (e.g., S508) for generating a trained model by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that the output approaches sampling data corresponding to the vibration data or sound data; A program for executing the above. According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, it is possible to generate a trained model for generating monitoring data that assumes fluctuations in rotational speed even in a situation where a rotational speed sensor cannot be used.

[0092] (8) a computer (e.g., 300), An acquisition step (S602) of acquiring vibration data or sound data of a rotating rolling bearing (e.g., 110); A generation process (e.g., S603) of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model (e.g., 405) that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; A monitoring step (e.g., S604, S605) of monitoring a state of the rolling bearing using the sampling data generated in the generating step; A program for executing the above. According to this configuration, it is possible to monitor the condition of the rolling bearing without using rotational speed information, regardless of whether the rotational speed of the rolling bearing changes. In particular, even in a situation where a rotational speed sensor cannot be used, it is possible to generate data that assumes fluctuations in rotational speed and perform rolling bearing condition monitoring. [Explanation of symbols]

[0093] 100...Bearing device 110...Roller bearing 111…Inner circle 112...Rolling element 113…Outer ring 114...Cage 120...Spindle 130...Vibration sensor 140…Amplifier 150...Rotational speed sensor 200...Learning processing device 201…Data acquisition section 202...Sampling processing unit 203...Learning data generation unit 204...Learning processing unit 205…Trained model management unit 205 300...Condition monitoring device 301…Data Acquisition Section 302…Trained model acquisition unit 303... State estimation unit 304...Status output unit

Claims

1. A first acquisition means for acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition means for acquiring a rotational speed of the rolling bearing during rotation; a derivation means for deriving a sampling timing from the vibration data or the sound data acquired by the first acquisition means in accordance with the rotational speed acquired by the second acquisition means so that the number of samples per one rotation of the rolling bearing becomes a predetermined value; a sampling processing means for generating sampling data by sampling the vibration data or the sound data acquired by the first acquisition means based on the timing derived by the derivation means; A generating means for generating learning data by associating the vibration data or sound data acquired by the first acquiring means with sampling data from the vibration data or sound data; A learning means for generating a trained model by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated by the generating means as input, so that an output approaches sampling data corresponding to the vibration data or sound data; A learning processing device having the above configuration.

2. An acquisition means for acquiring vibration data or sound data of the rolling bearing during rotation; A generation means for generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired by the acquisition means into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring means for monitoring a state of the rolling bearing by using the sampling data generated by the generating means; A condition monitoring device having the following:

3. A second acquisition means for acquiring a rotational speed of the rolling bearing during rotation; a determination means for determining whether or not the rotation speed can be acquired by the second acquisition means; a derivation means for deriving a sampling timing from the vibration data or the sound data acquired by the acquisition means in accordance with the rotational speed acquired by the second acquisition means, when the determination means determines that the rotational speed can be acquired, so that the number of samples per rotation of the rolling bearing becomes a predetermined value; a sampling processing means for generating sampling data by sampling the vibration data or the sound data acquired by the acquisition means based on the timing derived by the derivation means; Further comprising:

3. The condition monitoring device according to claim 2, wherein when the determination means determines that the rotational speed can be acquired, the monitoring means monitors the condition of the rolling bearing using the sampling data generated by the sampling processing means.

4. The learning processing device according to claim 1 ; A status monitoring device according to claim 2; A monitoring system for a rolling bearing comprising:

5. A first acquisition step of acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition step of acquiring a rotational speed of the rolling bearing during rotation; a derivation step of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotational speed acquired in the second acquisition step so that the number of samples per one rotation of the rolling bearing becomes a predetermined value; a sampling process step of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition step based on the timing derived in the derivation step; a generating step of generating learning data by associating the vibration data or sound data acquired in the first acquiring step with sampling data from the vibration data or sound data; A learning process in which a trained model is generated by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that an output approaches sampling data corresponding to the vibration data or sound data; A learning processing method having the following steps.

6. an acquisition step of acquiring vibration data or sound data of the rolling bearing during rotation; A generation process of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring step of monitoring a state of the rolling bearing by using the sampling data generated in the generating step; A condition monitoring method comprising:

7. On the computer, A first acquisition step of acquiring vibration data or sound data of a rotating rolling bearing; A second acquisition step of acquiring a rotational speed of the rolling bearing during rotation; a derivation step of deriving sampling timing from the vibration data or sound data acquired in the first acquisition step in accordance with the rotational speed acquired in the second acquisition step so that the number of samples per one rotation of the rolling bearing becomes a predetermined value; a sampling process step of generating sampling data by sampling the vibration data or the sound data acquired in the first acquisition step based on the timing derived in the derivation step; a generating step of generating learning data by associating the vibration data or sound data acquired in the first acquiring step with sampling data from the vibration data or sound data; A learning process in which a trained model is generated by performing a learning process using a predetermined learning algorithm, with vibration data or sound data among the training data generated in the generation process as input, so that an output approaches sampling data corresponding to the vibration data or sound data; A program for executing the above.

8. On the computer, an acquisition step of acquiring vibration data or sound data of the rolling bearing during rotation; A generation process of generating sampling data corresponding to the input vibration data or sound data by inputting the vibration data or sound data acquired in the acquisition process into a trained model that outputs sampling data corresponding to the vibration data or sound data when the vibration data or sound data is input; a monitoring step of monitoring a state of the rolling bearing by using the sampling data generated in the generating step; A program for executing the above.

Images