WO2019187138A1

WO2019187138A1 - Remaining lifespan prediction device and machine tool

Info

Publication number: WO2019187138A1
Application number: PCT/JP2018/013941
Authority: WO
Inventors: 友英那須; 左京松下
Original assignee: 株式会社牧野フライス製作所
Priority date: 2018-03-30
Filing date: 2018-03-30
Publication date: 2019-10-03

Abstract

This remaining lifespan prediction device (50) comprises: a first input unit (51) for receiving inspection data and inspection times for machines (12, 13); a storage unit (53) for storing the inspection data and inspection times; a second input unit (52) for inputting the lifespans of machines (12, 13) when the machines (12, 13) have become unusable; a machine learning unit (54b) configured so as to carry out, for a plurality of machines (12, 13) that have become unusable, machine learning having the inspection data as input and having, as output, remaining lifespans calculated for each past inspection time on the basis of the lifespans and inspection times; a remaining lifespan prediction unit (54b) configured so as to calculate predicted remaining lifespans for machines (12, 13) using inspection data and an algorithm obtained from the machine learning; and an output unit (55) for outputting the predicted remaining lifespans.

Description

Remaining life prediction device and machine tool

The present application relates to a remaining life prediction apparatus and a machine tool including a remaining life prediction unit.

Various devices for predicting equipment failure have been proposed. For example, Patent Document 1 discloses a rolling bearing failure diagnosis device that outputs a failure state of a rolling bearing. This apparatus includes a neural network that receives a Fourier spectrum based on the detected vibration data of the rolling bearing as an input, and outputs whether or not the rolling bearing has failed and the state of the failure.

JP-A-6-186136

In equipment maintenance, predictive maintenance may be used to reduce downtime. In order to efficiently perform predictive maintenance, it is important to know the remaining life of the equipment. Therefore, an object of the present invention is to provide a remaining life prediction apparatus and a machine tool that can indicate the remaining life of an apparatus.

One aspect of the present disclosure inputs inspection data obtained from a predetermined inspection operation in an apparatus and an inspection time that is a use time of the apparatus at the time when the inspection operation is performed in the apparatus for predicting the remaining life of the apparatus. A first input unit for storing, a storage unit for storing the inspection data and the inspection time input in the first input unit, and a lifetime for inputting the lifetime of the device when the device becomes unusable Machine learning that inputs inspection data and outputs the remaining life calculated for each past inspection time based on the life and inspection time for the second input unit and a plurality of devices that have become unusable The apparatus is configured to calculate a predicted remaining life of the device using the machine learning unit configured to perform the inspection, the inspection data input in the first input unit, and the algorithm obtained from the machine learning. The remaining life prediction unit and the remaining life Comprising an output unit for outputting the predicted residual life obtained by the life predicting portion, a residual life predicting device of the apparatus.

In the device remaining life prediction apparatus according to an aspect of the present disclosure, machine learning is performed in which inspection data obtained from a predetermined inspection operation is input and the remaining life at each inspection time is output. And the estimated remaining lifetime of an apparatus is calculated using the algorithm obtained from machine learning, and the newly input test | inspection data. Therefore, the remaining life prediction apparatus can indicate the remaining life of the device.

Machine learning includes (i) a convolutional neural network that uses the result of wavelet transform from inspection data, (ii) a neural network based on long- and short-term memory that uses feature values preprocessed from inspection data, and (iii) inspection data May be included in at least one of neural networks based on long-term and short-term memory.

Also, the machine learning may include a convolutional neural network and a neural network based on long- and short-term memory using preprocessed feature values. In this case, highly accurate prediction based on machine learning including two methods can be realized.

In addition, the preprocessed feature amount includes the peak amplitude obtained by fast Fourier transform of the envelope-processed inspection data, the peak amplitude obtained by direct fast Fourier transform of the inspection data, and the inspection data The mean square value and kurtosis obtained from the inspection data may be included. In this case, highly accurate prediction based on machine learning including the plurality of feature amounts can be realized.

The inspection operation may be included in the warm-up operation. In this case, it is not necessary to operate the device only for the inspection operation.

The usage time is the length of time from the time when the use of the device is started to the time when the use time is confirmed, or the time when the use of the device is started and the use time is confirmed. Two pieces of time data may be used.

Further, another aspect of the present disclosure is a sensor for obtaining inspection data related to equipment included in a machine tool in a machine tool, wherein the inspection data is obtained from a predetermined inspection operation in the equipment, and the sensor A remaining life prediction unit configured to calculate a predicted remaining life of a device using inspection data measured at a predetermined frequency and sampled at a predetermined frequency and an algorithm obtained from machine learning, and a remaining life prediction unit And a display device for displaying the predicted remaining life obtained in step 4, and the algorithm is determined by performing machine learning with inspection data as input and remaining life as output for multiple devices that have become unusable The remaining lifetime is based on the lifetime that is the usage time of the device when the device becomes unusable and the inspection time that is the usage time of the device when the inspection operation is performed. Te is calculated for each of the past inspection time, a machine tool. Such a machine tool can also indicate the remaining life of the device in the same manner as the remaining life prediction apparatus.

According to the aspect of the present disclosure, it is possible to provide a remaining life prediction apparatus and a machine tool that can indicate the remaining life of a device.

It is a schematic block diagram which shows the system which comprises the remaining life prediction apparatus which concerns on embodiment. An example of inspection data is shown. The other example of test | inspection data is shown. The other example of test | inspection data is shown. The other example of test | inspection data is shown. The other example of test | inspection data is shown. It is the schematic which shows the example of the machine learning which concerns on embodiment. An example of the image wavelet-transformed from vibration data is shown. It is the schematic which shows the other example of the machine learning which concerns on embodiment. It is a graph which shows an actual remaining life and an estimated remaining life. It is a schematic block diagram which shows the system which comprises the machine tool which concerns on embodiment.

Hereinafter, the remaining life prediction apparatus and the machine tool according to the embodiment will be described with reference to the accompanying drawings. Similar or corresponding elements are denoted by the same reference numerals, and redundant description is omitted. In order to facilitate understanding, the scale of the figures may be changed.

FIG. 1 is a schematic block diagram showing a system including a remaining life prediction apparatus according to an embodiment. In the system 100, the remaining lifetime of the equipment constituting the machine tool 10 is predicted. The system 100 includes a machine tool 10 and a remaining life prediction device 50.

The machine tool 10 can be, for example, various NC (Numerical Control) machine tools such as a machining center. In the machine tool 10, for example, only one or a plurality of predetermined processes can be mainly performed, and in this case, only a certain load mainly acts on devices such as the spindle 11 and the feed mechanism. The machine tool 10 includes, for example, a main shaft 11, a front bearing 12 and a rear bearing 13, a bearing case 14, and a sensor 15. The machine tool 10 may include other components.

The spindle 11 grips the tool T or the workpiece according to the type or application of the machine tool. The

bearings

12 and 13 rotatably support the main shaft 11. The

bearings

12 and 13 can be various bearings such as a rolling bearing such as a ball bearing or a roller bearing. For example, the inner rings of the

bearings

12 and 13 can be fixed to the main shaft 11, and the outer rings of the

bearings

12 and 13 can be fixed to the bearing case 14. The bearing case 14 accommodates the

bearings

12 and 13.

The sensor 15 is used to obtain inspection data relating to a device (in this embodiment, the main spindle 11) that is a target of calculation of the predicted remaining life. Such equipment may be, for example, a component of a machine tool used to move or transport a tool T or a workpiece (not shown). From another point of view, the equipment may be a moving element of a machine tool that includes two or more parts that are in rolling contact or sliding contact. For example, the sensor 15 can be used to obtain inspection data regarding the

bearings

12, 13 that support the main shaft 11.

The inspection data obtained from the measurement value detected by the sensor 15 is used to calculate the predicted remaining life. Such inspection data can be, for example, data representing a relationship between time and a physical quantity that varies according to the operation of the device. The inspection data can change over time depending on various factors, such as equipment damage and / or equipment usage time. For example, the sensor 15 can be an acceleration sensor, and in this case, the inspection data is vibration data representing a relationship between time and acceleration. Further, the sensor 15 may be a displacement sensor, and in this case, the inspection data is vibration data representing a relationship between time and displacement. The sensor 15 can be attached, for example, in the vicinity of the main shaft 11. In the present embodiment, the sensor 15 is attached to the outer surface of the bearing case 14 (for example, screwed). The sensor 15 may be attached to another component in the vicinity of the main shaft 11.

The inspection data may be data obtained from a servo motor (not shown) for rotating the spindle 11 instead of the vibration data as described above. In this case, the sensor 15 may be incorporated in a servo motor, for example.

The sensor 15 may be used to obtain inspection data relating to a feed mechanism (not shown) for moving a workpiece or a tool, instead of the spindle 11. In this case, the device for which the expected remaining life is calculated is, for example, a ball screw or a linear motion rolling guide included in a linear motion feed mechanism (for example, a table or a saddle), or a bearing included in a rotary table. Also good. Further, the sensor 15 may be used to obtain inspection data relating to an automatic tool changer (ATC) for moving a tool or an automatic pallet changer (APC) for moving a workpiece via a pallet. Good. In this case, a device such as a bearing included in ATC or APC can be a target for calculating the predicted remaining life. The sensor 15 can be mounted, for example, in the vicinity of the feed mechanism, ATC or APC, or can be incorporated in a servo motor for driving the feed mechanism, ATC or APC.

FIG. 2 shows an example of inspection data, and shows an example of vibration data obtained by the sensor 15 when the sensor 15 is an acceleration sensor. As shown in FIG. 2, the vibration data represents the relationship between time and acceleration.

FIGS. 3 to 6 show other examples of inspection data, showing examples of inspection data obtained by a servo motor. For example, the inspection data may be data representing a relationship between time and the position of the device, as shown in FIG. Further, for example, as shown in FIG. 4, the inspection data may be data representing a relationship between time and a position error representing a deviation from a desired position. Further, for example, the inspection data may be data representing a relationship between time and a torque command value of the servo motor, as shown in FIG. Further, for example, as shown in FIG. 6, the inspection data may be data representing a relationship between time and a reading value of an encoder of a servo motor.

Inspection data is obtained from a predetermined inspection operation on the equipment. The inspection operation can be performed, for example, once every several hours, once every day to every few days, once every week to every few weeks, or once every month to every few months. In the inspection operation, for example, the spindle 11, the feed mechanism, the ATC, or the APC or the like to which the predetermined tool T or the workpiece is attached or not attached, is processed at a predetermined rotational speed without machining the workpiece. It can be operated at a predetermined speed and / or a predetermined distance. The inspection operation may be performed for a relatively short time, for example, may be performed for 1 second to several tens of seconds or 1 minute to several minutes, or may be performed for a time shorter or longer than these times. The inspection operation can be, for example, a warm-up operation of the device or can be part of a warm-up operation.

Referring to FIG. 1, the machine tool 10 further includes a control device 20 and a data collection device 30.

The control device 20 may be, for example, an NC device for controlling the spindle 11 and the feed mechanism of the machine tool 10 and / or other equipment (for example, ATC and / or APC) of the machine tool 10. It may be a machine control device for controlling. The control device 20 includes a display unit 21 such as a touch panel or a liquid crystal display. In addition to the display unit 21, the control device 20 includes a processor such as a CPU (Central Processing Unit), a memory such as a hard disk drive, a ROM (read only memory), a RAM (random access memory), and / or a mouse and / or Other components such as an input device such as a keyboard can be included.

The data collection device 30 is connected to the sensor 15 in a wired or wireless manner, and receives a measurement value detected by the sensor 15. Further, the data collection device 30 is connected to the control device 20 so as to be communicable by wire or wirelessly. The data collection device 30 has a storage unit 31. The storage unit 31 can be, for example, a hard disk drive. The storage unit 31 stores a plurality of sets of inspection data based on the measurement value received from the sensor 15 and inspection time when the corresponding inspection operation is performed. The measurement value detected by the sensor 15 is sampled at a predetermined frequency (for example, 18 kHz to 120 kHz) and stored in the storage unit 31 as inspection data. In the present disclosure, “inspection time” may mean the usage time of the device at the time when the inspection operation is performed. Further, in the present disclosure, the “use time” may mean the total time that the device actually operates from the time when the device is installed or from a predetermined time after the device is installed. . Alternatively, for example, the “use time” may be the length of time from the time when use of the device is started to the time when the use time is confirmed. Furthermore, two time data of the time when the use of the device is started and the time when the use time is confirmed are stored as the use time, and in the following machine learning and calculation of the predicted remaining life, etc. The remaining time may be calculated as the usage time in the remaining life prediction apparatus 50. The data collection device 30 and / or the control device 20 may include a counter for counting the usage time of the device. The storage unit 31 may further store the “life” of the device (details will be described later). The “life” may be input to the data collection device 30 by the user of the machine tool 10 when the device becomes unusable, for example. Further, the data collection device 30 may calculate the “remaining life” (details will be described later) of the device at each past inspection time based on the “inspection time” and the “lifetime”. The recorded “remaining life” may be stored.

The data collection device 30 may be provided separately from the control device 20 or may be incorporated in the control device 20. When the data collection device 30 is incorporated in the control device 20, each component of the data collection device 30 may be realized by a corresponding component in the control device 20, or what is a component in the control device 20? It may be provided separately. In addition to the storage unit 31, the data collection device 30 may include other components such as a processor such as a CPU, a ROM, a RAM, and / or an input / output device such as a touch panel, a mouse, and / or a keyboard. .

The remaining life prediction device 50 can communicate with the control device 20 and the data collection device 30 via a network. The remaining life prediction apparatus 50 can be provided on the manufacturer side of the machine tool 10, for example, and can be, for example, a computer, a server, and / or a mobile terminal such as a smartphone and / or a tablet. The network is, for example, a communication system based on a communication standard defined by 3GPP (Third Generation Partnership Project) such as the Internet, WiFi (registered trademark), Bluetooth (registered trademark), 3G and 4G, or other communication methods, or Any combination of these can be used, wired or wireless, or a combination thereof. In FIG. 1, one machine tool 10 is connected to the remaining life prediction apparatus 50, but a plurality of machine tools may be connected to the remaining life prediction apparatus 50.

The remaining life prediction apparatus 50 can include a first input unit 51, a second input unit 52, a storage unit 53, a processor 54, and an output unit 55. These elements are connected via a bus or the like. Are connected to each other. The remaining life prediction apparatus 50 may further include other components.

The first input unit 51 receives input of inspection data and inspection time. The first input unit 51 can be, for example, an I / O port. For example, the first input unit 51 can receive the inspection data and the inspection time transmitted from the data collection device 30. The first input unit 51 may be configured to read inspection data and inspection time stored in a storage medium. The inspection data and the inspection time input from the first input unit 51 are transmitted to the storage unit 53. For example, when data of a plurality of devices of the machine tool 10 is stored, the storage device 53 is used to indicate which device of the machine tool 10 the inspection data and the inspection time input from the first input unit 51 are related to. Identification information indicating which region of the inspection data and inspection time should be stored may be input to the first input unit 51. Similarly, for example, in a case where data of a plurality of machine tools 10 are stored, in order to indicate which machine tool the inspection data and inspection time input from the first input unit 51 are associated with, Identification information indicating whether inspection data and inspection time should be stored in the area may be input to the first input unit.

The second input unit 52 receives an input of the life of the device when the device becomes unusable (for example, when it fails or is damaged). In the present disclosure, the “lifetime” may mean the usage time of the device when the device becomes unusable. Moreover, the 2nd input part 52 can receive the input of failure mode, when an apparatus becomes unusable. In the present disclosure, the “failure mode” may mean a failure type indicating how the device has become unusable. For example, if the device is a

bearing

12 or 13 of the spindle 11, the failure modes can include outer ring breakage, inner ring breakage, and cage breakage. Failure modes may also include, for example, bearing indentation, poor lubrication and / or preload loss, or ball screw wear, indentation, misalignment and / or poor lubrication.

The second input unit 52 can be, for example, an input device such as a mouse, a keyboard, and / or a touch panel. In this case, when the operator on the manufacturer side receives information from the user of the machine tool 10 that the device has become unusable, the lifetime and failure mode can be input by the second input unit 52. The second input unit 52 can be, for example, an I / O port or the like. In this case, the lifetime and failure mode of the device transmitted from the data collection device 30 can be received by the second input unit 52. In this case, the second input unit 52 may be realized by an I / O port for the first input unit 51, or may be provided separately from the I / O port for the first input unit 51. The lifetime and failure mode input to the second input unit 52 are transmitted to the storage unit 53. As described above, identification information indicating in which area of the storage device 53 the life should be stored may be input to the second input unit 52.

The storage unit 53 stores the inspection data and the inspection time transmitted from the first input unit 51, and the life and failure mode transmitted from the second input unit 52. The storage unit 53 can be, for example, a hard disk drive. The storage unit 53 may store other programs and / or data such as a program executed by the processor 54 described below.

The processor 54 can be, for example, one or more CPUs. The processor 54 includes a machine learning unit 54 a and a remaining life prediction unit 54 b, and these can be realized by a program stored in the storage unit 53, for example.

The machine learning unit 54a performs machine learning to obtain an algorithm for predicting the remaining life of the device. In the machine learning unit 54a, for example, a neural network can be used, and an algorithm in which the weight in each neuron is determined is obtained by machine learning.

The machine learning unit 54a obtains “remaining life” in each past inspection time based on the life and inspection time stored in the storage unit 53 for each of the plurality of devices that have become unusable. Specifically, the “remaining life” at each inspection time can be calculated by subtracting each inspection time from the life. Note that, as described above, when the remaining life is calculated by the data collection device 30 on the user side of the machine tool 10, the machine learning unit 54a does not need to calculate the remaining life. I want.

Subsequently, the machine learning unit 54a performs machine learning with the inspection data as an input and the remaining life as an output, using data of a plurality of devices that have become unusable.

The machine learning unit 54a can further perform machine learning to obtain an algorithm for predicting the failure mode of the device. In this case, the machine learning unit 54a performs machine learning using test data as input and failure mode as output using data of a plurality of devices that have become unusable.

FIG. 7 is a schematic diagram illustrating an example of machine learning according to the embodiment. As shown in FIG. 7, the machine learning unit 54a can perform machine learning based on various methods. For example, the machine learning performed by the machine learning unit 54a may be a convolutional neural network (CNN (Convolutional Neural Network)) that uses preprocessed inspection data. For example, wavelet transformation can be used as the preprocessing, and a wavelet transformed image is obtained by the preprocessing.

FIG. 8 shows an example of an image obtained as a result of wavelet transform from inspection data. The image may include information indicating a comparison between the inspection data at the first inspection time and the inspection data at a certain inspection time after the second inspection. For example, the image can include a red portion (R), a green portion (G), and a blue portion (B). A green part shows the area | region which has not changed from the first test data. The red part indicates a region where there is a change from the first inspection data. The blue part indicates an area where there is no data. Referring to FIG. 7, the preprocessed image as described above can be used as an input to the CNN, and the remaining life (RUL (Remaining Useful Life)) and the failure mode are output.

Further, for example, the machine learning may be a neural network based on long-short-term memory (LSTM (Long Short-Term Memory)) using feature values preprocessed from inspection data. The feature amount may include, for example, the peak amplitude obtained by performing fast Fourier transform on the inspection data subjected to envelope processing. In this case, the inspection data is subjected to, for example, a band pass filter, MED (Minimum Entropy Deconvolution), and envelope processing (for example, Hilbert transform), and then to a fast Fourier transform (FFT (Fast Fourier Transform)). May be. The feature amount may include, for example, the peak amplitude obtained by directly performing fast Fourier transform on the inspection data, the root mean square value of the inspection data, and the kurtosis obtained from the inspection data. The kurtosis can be obtained, for example, from inspection data passed through a bandpass filter and MED. These preprocessed feature quantities can be used as inputs to the LSTM, with RUL and failure modes being the outputs. It should be noted that the feature amount may include a larger number of preprocessed numerical values.

Further, for example, the machine learning may be a neural network based on LSTM that uses inspection data as it is. In such machine learning, unprocessed inspection data can be used as input to the LSTM, and RUL and failure mode are output.

FIG. 9 is a schematic diagram illustrating another example of machine learning according to the embodiment. As shown in FIG. 9, the machine learning performed by the machine learning unit 54 a includes CNN that uses the wavelet-transformed image in FIG. 7, a neural network that is based on LSTM that uses feature values preprocessed from inspection data, and A combination of these may be used. Machine learning may further include other neural networks. In the machine learning of FIG. 9, the inspection data is passed through a filter that uses the results obtained by the CNN. Specifically, in a wavelet-transformed image, if the red portion occupies a predetermined ratio or more of the entire image, the corresponding inspection data is determined as “abnormal”; otherwise, the corresponding inspection data is “normal”. It is determined. Only the inspection data determined to be “abnormal” can pass through the filter. As shown in FIG. 9, only the inspection data that has passed through the filter is preprocessed, and the feature value obtained from the preprocess is used in the neural network based on LSTM, and the RUL and failure mode are output.

Referring to FIG. 1, the remaining life prediction unit 54 b uses the new inspection data input by the first input unit 51 and the algorithm obtained from the above machine learning, and the predicted remaining life of the device and Calculate the failure mode. In the present disclosure, the “predicted remaining life” is an operation that is performed when it is assumed that a device that has been used for a certain period of use will continue to perform machining that is subject to a load similar to that before the machine tool 10 performs the prediction. It can mean the remaining time expected to be able to. When new data of an unusable device is added to the storage unit 53 and the algorithm is updated in the machine learning unit 54a, the remaining life prediction unit 54b can be updated.

The inspection data used in the machine learning unit 54a and the remaining life prediction unit 54b may be obtained by, for example, an acceleration test. In this case, operations such as algorithm acquisition and system verification can be performed earlier. For example, the acceleration test can be performed by providing an indentation on a device such as a

bearing

12, 13 or a ball screw.

The output unit 55 outputs the predicted remaining life and the failure mode obtained by the remaining life prediction unit 54b to the control device 20. The output unit 55 can be, for example, an I / O port.

Next, the operation of the system 100 will be described.

In the system 100, an inspection operation is executed on the spindle 11 every time the engine is warmed up, and a measured value is obtained by the sensor 15. The measured value is sampled at a predetermined frequency and stored in the storage unit 31 as inspection data. The storage unit 31 also stores the inspection time.

Each time an inspection operation is performed, the data stored in the storage unit 31 is transmitted to the remaining life prediction apparatus 50. Alternatively, when inspection data exceeding a predetermined capacity is stored in the storage unit 31 or when the bearing 12 or 13 becomes unusable, the data stored in the storage unit 31 is the remaining life prediction device. 50 may be transmitted. Data input through the first input unit 51 is stored in the storage unit 53.

When the bearing 12 or 13 becomes unusable, the operator of the manufacturer inputs the life of the corresponding bearing through the second input unit. The data input at the second input unit is stored in the storage unit 53.

When the data of the

unusable bearings

12 or 13 exceeding the predetermined number is stored in the storage unit 53, the machine learning unit 54a performs machine learning using the inspection data as an input and the remaining life as an output. Further, the machine learning unit 54a performs machine learning using the inspection data as an input and the failure mode as an output. Then, the remaining life prediction unit 54 b calculates the predicted remaining life and failure mode of the device using the algorithm obtained from the machine learning and the new inspection data input by the first input unit 51. The calculated predicted remaining life and failure mode can be output from the output unit 55 to the control device 20 and displayed on the display unit 21 of the control device 20.

FIG. 10 is a graph showing the actual remaining life and the predicted remaining life. The horizontal axis of FIG. 10 shows the usage time (or inspection time) of the bearing, and the vertical axis shows the ratio between the remaining life and the life of the bearing at each usage time. The “actual remaining life” (Actual RUL) in FIG. 10 is calculated by subtracting each inspection time from the life of a bearing that has actually become unusable. The “predicted remaining life” (Predicted１０RUL) in FIG. 10 is calculated using the same inspection data of the bearing that has actually become unusable and the algorithm obtained from machine learning. As shown in FIG. 10, the predicted remaining life is in good agreement with the actual remaining life.

The algorithm used to calculate the “predicted remaining life” in FIG. 10 uses the inspection data and remaining life of a plurality of other bearings that have actually become unusable, and the two neural networks shown in FIG. It is derived from machine learning based on combinations. In the inspection operation, the main shaft 11 was rotated at 1800 rpm (30 Hz). The inspection data is vibration data representing the relationship between time and acceleration. In LSTM, 41 feature quantities preprocessed from vibration data were used. Specifically, the 41 feature values are the inner ring, the outer ring, the rolling element, and the holding for each of the front bearing 12 and the rear bearing 13 (× 2) obtained by direct Fourier transform of vibration data. Front bearing 12 obtained by fast Fourier transform of amplitude (16 feature amount = 2 × 4 × 2) and envelope-processed vibration data at the first and second (× 2) of the device (× 4) And the peak amplitude at the first and second (× 2) for the inner ring, outer ring, rolling element and cage (× 4) for each of the rear bearings 13 (× 2) (16 feature amounts = 2 × 4 × 2) In addition, the rotational frequency (30 Hz) of the spindle 11 obtained by fast Fourier transform of the vibration data subjected to the envelope processing and a frequency (60 Hz) (×) twice that of the spindle 11 are obtained. Peak amplitude (2 feature quantities), vibration data power (energy) in the low frequency range (5 Hz to less than 1300 Hz), intermediate frequency range (1300 Hz to less than 5000 Hz) and high frequency range (5000 Hz to less than 9000 Hz) (3 features) Quantity), and the mean square value, entropy, skewness, and kurtosis (four feature quantities) of vibration data.

As described above, in the remaining life prediction apparatus 50 according to the embodiment, machine learning is performed in which the inspection data obtained from a predetermined inspection operation is input and the remaining life at each inspection time is output. Then, the predicted remaining life of the

bearing

12 or 13 is calculated using an algorithm obtained from machine learning and newly input inspection data. Therefore, the remaining life prediction device 50 can indicate the remaining life of the

bearing

12 or 13.

In the remaining life prediction apparatus 50, machine learning is performed by (i) a convolutional neural network using an image wavelet transformed from inspection data, and (ii) a neural network based on long- and short-term memory using feature values preprocessed from the inspection data. It may include at least one of a network and (iii) a neural network based on long and short-term memory using the inspection data as it is. In the remaining life prediction apparatus 50, the machine learning can also include a convolutional neural network and a neural network based on long- and short-term memory using preprocessed feature values. In this case, highly accurate prediction based on machine learning including two methods can be realized.

Further, in the remaining life prediction apparatus 50, the preprocessed feature amount is obtained by performing fast Fourier transform directly on the amplitude of the peak obtained by performing fast Fourier transform on the inspection data subjected to envelope processing. The peak amplitude, the root mean square value of the inspection data, and the kurtosis obtained from the inspection data are included. Therefore, highly accurate prediction based on machine learning including the plurality of feature amounts can be realized.

In the remaining life prediction apparatus 50, the inspection operation is included in the warm-up operation. Therefore, it is not necessary to operate the machine tool 10 only for the inspection operation.

Next, a system according to another embodiment will be described.

FIG. 11 is a schematic block diagram illustrating a system including the machine tool according to the embodiment. In the system 200, the remaining life prediction unit 32a is provided in the data collection device 30 on the user side of the machine tool 40, and the device 60 on the manufacturer side does not have the remaining life prediction unit 54b. Different from the system 100 of FIG. Specifically, in the system 200, the processor 32 of the data collection device 30 has a remaining life prediction unit 32a. For other components, the system 200 may be similar to the system 100.

The processor 32 can be, for example, one or more CPUs. In the system 200, an algorithm for predicting the remaining life and failure mode of the device is obtained in the machine learning unit 54a of the device 60 on the manufacturer side in the same manner as the system 100 described above. The obtained algorithm is copied to the data collection device 30 as the remaining life prediction unit 32a. The remaining life prediction unit 32a can be realized by a program stored in the storage unit 31, for example. When new data of an unusable device is added to the storage unit 53 and the algorithm is updated in the machine learning unit 54a, the remaining life prediction unit 32a can be updated through the network. The remaining life prediction unit 32a uses the inspection data measured by the sensor 15 and sampled at a predetermined frequency, and the algorithm obtained from the machine learning in the machine learning unit 54a, and the predicted remaining life and failure mode of the device. Is calculated. The calculated predicted remaining life and failure mode can be output to the control device 20 and displayed on the display unit 21.

The machine tool 40 as described above can indicate the remaining life of the device, similarly to the remaining life prediction apparatus 50 described above.

Although the embodiment of the remaining life prediction apparatus and the machine tool has been described, the present invention is not limited to the above embodiment. Those skilled in the art will appreciate that various modifications of the above-described embodiments are possible. In addition, a person skilled in the art can incorporate a feature included in one embodiment into another embodiment or replace a feature included in the other embodiment as long as no contradiction arises. Will understand.

10 Machine tools 11 Spindles 12 Front bearings (equipment)
13 Rear bearing (equipment)
DESCRIPTION OF SYMBOLS 15 Sensor 21 Display part 32a Remaining life prediction part 40 Machine tool 50 Remaining life prediction apparatus 51 1st input part 52 2nd input part 53 Memory | storage part 54a Machine learning part 54b Remaining life prediction part 55 Output part T tool

Claims

In the equipment remaining life prediction device,
A first input unit for inputting inspection data obtained from a predetermined inspection operation in the device and an inspection time which is a usage time of the device at the time when the inspection operation is performed;
A storage unit for storing the inspection data and the inspection time input by the first input unit;
A second input unit for inputting a lifetime which is a usage time of the device when the device becomes unusable;
Machine learning is performed on a plurality of devices that have become unusable, with the inspection data being input and the remaining life calculated for each past inspection time based on the life and the inspection time being output. A machine learning part configured in
A remaining life prediction unit configured to calculate a predicted remaining life of the device using the inspection data input in the first input unit and an algorithm obtained from the machine learning;
An output unit for outputting the predicted remaining life obtained by the remaining life prediction unit;
An apparatus for predicting the remaining life of equipment.
The machine learning is
A convolutional neural network using a result of wavelet transform from the inspection data,
A neural network based on long- and short-term memory using feature quantities pre-processed from the inspection data; and
A neural network based on long-term memory using the inspection data as it is,
The apparatus remaining life prediction apparatus of Claim 1 containing at least 1 of these.
3. The remaining life prediction apparatus according to claim 2, wherein the machine learning includes the convolutional neural network and a neural network based on long- and short-term memory using the preprocessed feature amount.
The preprocessed feature amount is a peak amplitude obtained by fast Fourier transform of the envelope-processed inspection data, a peak amplitude obtained by direct fast Fourier transform of the inspection data, the inspection The apparatus remaining life prediction apparatus according to claim 2, comprising a root mean square value of data and a kurtosis obtained from the inspection data.
The remaining life prediction apparatus according to claim 1, wherein the inspection operation is included in a warm-up operation.
The usage time is the length of time from the time when the use of the device is started to the time when the use time is confirmed, or the time when the use of the device is started and the confirmation of the use time. The apparatus remaining life prediction apparatus according to claim 1, wherein the apparatus is two pieces of time data of performed times.
In machine tools,
A sensor for obtaining inspection data relating to equipment included in the machine tool, wherein the inspection data is obtained from a predetermined inspection operation in the equipment;
A remaining life prediction unit configured to calculate a predicted remaining life of the device using the inspection data measured by the sensor and sampled at a predetermined frequency, and an algorithm obtained from machine learning;
A display device for displaying the predicted remaining life obtained by the remaining life prediction unit;
With
The algorithm is determined by performing machine learning with the inspection data as an input and the remaining life as an output for a plurality of devices that have become unusable, and the remaining life becomes unusable for the device. Calculated for each past inspection time based on the lifetime that is the usage time of the device at the time of the inspection and the inspection time that is the usage time of the device when the inspection operation is performed. A machine tool characterized by