Failure prediction system

Abstract

A motor control device (40) comprises: an inverter (60); a current sensor (70) that detects an inverter current; a rotation angle sensor (78) that detects a rotation angle of a motor; and a microcomputer (50) that acquires sensor values of the current sensor and the rotation angle sensor and controls the inverter (60). In the motor control device (40), the failure prediction system determines a deterioration state of and predicts a failure of inverter elements (61-66) or inverter peripheral elements (43, 44, 46, 67-69) for switching the energization of the inverter. A learning unit (23, 53) performs machine learning by associating learning data with the deterioration state of the inverter elements or the inverter peripheral elements and creates a learned model. A failure prediction unit (34, 54) checks data obtained by frequency-analyzing inverter current against the learned model.

Description

Failure prediction system CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application is based on Japanese Application No. 2024-100780 filed on June 21, 2024, the contents of which are incorporated herein by reference.

[0002] The present disclosure relates to a failure prediction system.

[0003] Conventionally, failure prediction devices and systems for predicting failures of motors and the like have been known. For example, a failure prediction device disclosed in Patent Document 1 uses an autoencoder to detect changes in current values ​​and the like over time and predicts failures of motors whose degradation state changes over time. Patent Document 2 discloses a semiconductor manufacturing device equipped with a failure prediction circuit for a motor that drives the mechanism of an electrolytic plating device. This device predicts motor failures based on time changes in feature quantities (such as frequency spectrum) based on the motor's load factor, vibration, etc.

[0004] Patent Literature 3 discloses a failure prediction system equipped with a machine learning unit that learns the failure conditions of a robot from various sensor values. The learning unit of the machine learning device learns the failure conditions by supervised learning or unsupervised learning using a neural network according to a training data set of the various sensor values.

[0005] JP 2023-002202 A JP 2021-102817 A JP 2021-002398 A

[0006] This specification focuses on failure prediction for three-phase brushless motors and other motors that receive power from an inverter. Typical failure modes in this type of motor include overheating of the inverter elements that switch the inverter's power supply and wear of the bearing parts that support the motor shaft. Similar to inverter element failure, failures in peripheral inverter elements such as relays and capacitors that are installed in the power path from the power source through the inverter to the motor are also anticipated.

[0007] Determining the deterioration state that indicates a failure before it actually occurs is particularly important for on-board motor control devices, which require high reliability.Patent Documents 1 to 3 only provide broad and general disclosures regarding failure prediction for industrial motors and the like, and do not mention specific configurations or methods for determining the deterioration state of inverter elements or bearing parts.

[0008] An object of the present disclosure is to provide a failure prediction system that appropriately determines the deterioration state of inverter elements and bearing parts and predicts failures in a motor control device that supplies power from an inverter to a motor.

[0009] The failure prediction system disclosed herein is a system that determines a deterioration state that indicates a failure in a motor control device and predicts failure. This motor control device includes an inverter that supplies power to the motor, a current sensor that detects the inverter current, which is a current of one or more phases flowing through the inverter, a rotation angle sensor that detects the rotation angle of the motor, and a microcomputer that acquires sensor values ​​from the current sensor and the rotation angle sensor and controls the inverter. This failure prediction system includes a learning unit that performs machine learning and creates a trained model, and a failure prediction unit that determines the deterioration state and predicts failure.

[0010] A failure prediction system according to a first aspect of the present disclosure determines the deterioration state of inverter elements that switch the power supply to the inverter, or inverter peripheral elements that are provided in the power supply path from the power source through the inverter to the motor, and predicts failure.

[0011] The learning unit performs machine learning to associate learning data obtained by frequency analysis of the inverter current during trial energization of the motor control device with the deterioration state of the inverter elements or inverter peripheral elements, creating a trained model. The failure prediction unit compares data obtained by frequency analysis of the inverter current acquired by the microcomputer from the current sensor while the motor is energized with the trained model to determine the deterioration state of the inverter elements or inverter peripheral elements. By predicting failures in this way, the user can take appropriate measures early.

[0012] A failure prediction system according to a second aspect of the present disclosure determines the deterioration state of a bearing component that supports a motor shaft and predicts failure.

[0013] The learning unit performs machine learning to associate learning data obtained by frequency analysis of changes in the motor's rotation angle during trial motor rotation by the motor control device with the deterioration state of the bearing components, creating a trained model. The failure prediction unit compares the data obtained by frequency analysis of changes in the motor's rotation angle obtained by the microcomputer from the rotation angle sensor while the motor is rotating with the trained model to determine the deterioration state of the bearing components. By predicting failures in this way, users can take appropriate measures early.

[0014] The above and other objects, features, and advantages of the present disclosure will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, in which Fig. 1 is a configuration diagram of a motor control device, Fig. 2 is a mechanical cross-sectional view of an electromechanical integrated motor, Fig. 3 is a time chart illustrating determination of a deterioration state of a power transistor (inverter element) by frequency analysis of an inverter current, Fig. 4 is a time chart illustrating determination of a deterioration state of a bearing (bearing component) by frequency analysis of a change in a motor rotation angle, Fig. 5 is a block diagram of a learning system of a failure prediction system according to first and second embodiments, Fig. 6 is a block diagram of a monitoring system of the failure prediction system according to the first embodiment, Fig. 7 is a block diagram of a monitoring system of the failure prediction system according to the second embodiment, Fig. 8 is a block diagram of a learning and monitoring system of the failure prediction system according to the third embodiment, Fig. 9 is a flowchart of learning by the failure prediction system, Fig. 10 is a sub-flowchart of creating frequency spectrum data, Fig. 11 is a flowchart of failure prediction during product operation, and Fig. 12 is a diagram illustrating inverter peripheral elements in a motor control device.

[0015] A failure prediction system according to several embodiments will be described with reference to the drawings. This failure prediction system includes a "learning unit" and a "failure prediction unit" in a motor control device that supplies power from an inverter to a motor, and determines the deterioration state of inverter elements and bearing components to predict failure. The following first to third embodiments differ in the functional arrangement of the learning unit and the failure prediction unit. The first to third embodiments are collectively referred to as "the present embodiment." Note that failure prediction of inverter peripheral elements such as relays and capacitors will be supplemented as "other embodiments."

[0016] The configuration of the motor control device 40 will be described with reference to Figures 1 and 2. The motor control device of this embodiment is, for example, a device that drives a steering assist motor in an electric power steering device of a vehicle. This steering assist is not limited to manual steering by a driver, but also includes assistance in response to steering commands from an automatic driving control device. Additionally, the failure prediction system of this embodiment may be applied to the motor control device 40 of an AGV (automated guided vehicle) or the like.

[0017] The motor 80 is configured, for example, as a three-phase brushless motor. As shown in Figure 2, the motor 80, which is the mechanical part, and the drive control unit corresponding to the motor 80 are integrated into a mechanically and electrically integrated system. Possible failure modes include both an electrical failure of the drive control unit and a mechanical failure of the motor 80. For this reason, in this embodiment, the motor 80 is included as part of the motor control device 40.

[0018] Figure 1 shows the electrical circuit configuration of the motor control device 40. The motor control device 40 includes a microcomputer 50, an inverter 60, a motor 80, a current sensor 70, and a rotation angle sensor 78. The microcomputer 50 acquires sensor values ​​from the current sensor 70 and the rotation angle sensor 78 and controls the supply of current from the inverter 60 to the motor 80. Although Figure 1 shows an example of a single-system configuration, multiple systems may be provided for redundancy. For example, in a two-system motor control device 40, the two inverters are each controlled by a corresponding microcomputer.

[0019] The inverter 60 has power transistors 61-66 such as MOSFETs connected in a bridge configuration as "inverter elements" that switch the energization of the inverter 60, i.e., three-phase upper and lower arm switching elements. The power transistors 61, 62, and 63 are upper arm elements for the U, V, and W phases, respectively, and the power transistors 64, 65, and 66 are lower arm elements for the U, V, and W phases, respectively. The connection points between the arms of each phase are connected to the windings of the motor 80. In the inverter 60, the power transistors 61-66 perform switching operations in response to drive signals from the driver 56 of the microcomputer 50, converting the DC power of the battery BT into three-phase AC power, which is then supplied to the motor 80.

[0020] The phase current flowing through inverter 60 is called the "inverter current," and inverter currents of one or more phases are collectively represented by the symbol Im. Current sensor 70 is composed of, for example, shunt resistors 71, 72, and 73 provided on the ground side of the lower arm elements of each phase, and detects inverter current Im by converting the divided voltage of shunt resistors 71, 72, and 73. The output of current sensor 70 is input to receiver 57 of microcomputer 50 as an analog signal of, for example, 0 to 5 V, and is then AD converted.

[0021] In other embodiments, a shunt resistor may be provided on the power supply side of the upper arm element or in the power supply path to the motor 80. Also, a magnetic field detection type current sensor such as a core current sensor, a Hall element, or a magnetic impedance sensor may be used.

[0022] The rotation angle sensor 78 is configured with, for example, a Hall element (see FIG. 2), an MR element, a resolver, etc., and detects the motor rotation angle θm. The output of the rotation angle sensor 78 is input to the receiver 58 of the microcomputer 50 as an analog signal of, for example, −5 to 5 V, and is then AD converted. Alternatively, if the rotation angle sensor 78 is configured with an encoder, the digital signal may be serially communicated by the receiver 58.

[0023] In this way, the microcomputer 50 performs feedback control of the inverter 60 using information on the inverter current Im and the motor rotation angle θm output by the current sensor 70 and the rotation angle sensor 78. Furthermore, in this embodiment, the sensor values ​​of the current sensor 70 and the rotation angle sensor 78 are used not only as information for feedback control but also as information for failure prediction.

[0024] The data acquisition unit 51 of the microcomputer 50 acquires the inverter current Im and the motor rotation angle θm from the receiving units 57 and 58 as information for failure prediction. In this case, the data acquisition unit 51 may acquire the inverter current Im for three phases, as indicated by the dashed lines, or may acquire the current for only one phase. In the following explanation of failure prediction, the "current sensor 70" does not refer to current sensors for all phases, but rather to the current sensor for the phase from which the data acquisition unit 51 acquires the inverter current Im.

[0025] An example of the mechanical configuration of an electromechanical integrated motor is shown in Figure 2. The output shaft side of the motor 80 shown in the lower part of Figure 2 is referred to as the "front side," and the cover 88 side shown in the upper part of Figure 2 is referred to as the "rear side." A drive control unit 500 including a microcomputer 50 and an inverter 60 is mounted on a substrate 87 and covered by a resin cover 88. The motor 80 includes a case 81, a stator 82, a rotor 83, a shaft 84, a rear frame 85, bearings 861 and 862, etc. These components are arranged coaxially with respect to the rotation axis Z.

[0026] The case 81 is made of metal such as iron, has a cylindrical portion 811 and a bottom portion 812, and is open on the rear side opposite the bottom portion 812. The bottom portion 812 side of the case 81 houses a stator 82 and a rotor 83, and a rear frame 85 is fitted to the opening side.

[0027] The stator 82 has a three-phase winding 822 wound around a stator core 821 fixed to the inner wall of the case 81. The three-phase winding 822 is connected to a substrate 87 via lead wires 823. The drive control unit 500 controls the supply of current to the three-phase winding 822, thereby forming a rotating magnetic field in the stator 82.

[0028] The rotor 83 is provided radially inside the stator 82, and a shaft 84 is fixed to the center of the rotor core 831. The shaft 84 is rotatably supported by a front bearing 861 held by the bottom 812 of the case 81 and a rear bearing 862 held by the rear frame 85. The front bearing 861 and the rear bearing 862 correspond to "bearing components" that support the shaft 84 of the motor 80.

[0029] A plurality of permanent magnets 832 are provided on the outer edge of rotor core 831. For example, in an IPM motor, permanent magnets 832 are embedded in rotor core 831. Rotor 83 rotates about shaft 84 due to a rotating magnetic field formed in stator 82 when current is applied to three-phase windings 822. A sensor magnet 847 for detecting the angle of rotation is provided at the rear end of shaft 84. Rotation angle sensor 78, formed for example by a Hall element, is disposed opposite sensor magnet 847 on the surface of substrate 87 facing rear frame 85, and detects the angle of rotation of shaft 84 from changes in the magnetic field of sensor magnet 847.

[0030] The rear frame 85 is made of an aluminum alloy or the like, and supports the shaft 84 via a rear bearing 862. The rear frame 85 also supports the board 87 and functions as a heat sink that receives heat dissipated from elements such as the power transistors 61-66 mounted on the board 87.

[0031] In the motor control device 40 configured as described above, the failure prediction system targets the power transistors 61-66, which are "inverter elements," and the bearings 861, 862, which are "bearing components," and determines the deterioration state that is a sign of failure and predicts failure.

[0032] Power transistors 61-66 may overheat and deteriorate due to heat generated by current flow, which may lead to failure. Bearings 861, 862 may wear out due to rotation of shaft 84, which may lead to failure. Detecting a deterioration state that is a sign of failure before a failure causes loss of function is particularly important for an in-vehicle motor control device 40 that requires high reliability.

[0033] Therefore, the failure prediction system of this embodiment uses frequency spectrum data obtained by frequency analysis of the inverter current Im as feature quantity data related to the deterioration state of the power transistors 61 to 66. The failure prediction system of this embodiment also uses frequency spectrum data obtained by frequency analysis of the change in the motor rotation angle θm, i.e., the rotation speed, as feature quantity data related to the deterioration state of the bearings 861 and 862.

[0034] In this failure prediction system, a learning unit performs machine learning on frequency spectrum data, which is a feature quantity, during trials of energizing and rotating the motor control device 40 in a laboratory or factory, to create a trained model. Furthermore, a frequency spectrum created from the inverter current Im and motor rotation angle θm during operation of the motor control device 40 mounted on a vehicle or transport vehicle is compared with the frequency spectrum of the trained model. In this way, the failure prediction unit determines the deterioration state of the power transistors 61-66 or bearings 861, 862 and predicts failure. By predicting failure, the user can take appropriate measures early on.

[0035] Next, an example of signal processing and learning for failure prediction will be described with reference to Figures 3 and 4. As shown in Figure 3, when the power transistors 61-66 are normal, the inverter current Im is sinusoidal. When the power transistors 61-66 deteriorate, the peak value of the sinusoidal wave decreases due to waveform distortion, but the degree of change is slight, making it difficult to determine the deterioration state from the signal waveform as is. Therefore, in the frequency spectrum obtained by frequency analysis of the inverter current Im, the intensity ratio of the Nth harmonic frequency components (i.e., frequency components that are integer multiples of the rotation frequency fr) 2fr, 3fr, 4fr, ... to the rotation frequency fr is evaluated.

[0036] When the power transistors 61-66 are normal, the frequency component of the Nth harmonic is almost absent, but when the power transistors 61-66 deteriorate, the intensity ratio of the zeroth harmonic component and the rotation frequency component fr decreases, and the intensity ratios of the second, third, and fourth harmonic frequency components 2fr, 3fr, and 4fr increase.The frequency components and intensity ratios are compared with the trained model to determine the deterioration state of the power transistors 61-66.

[0037] As shown in Figure 4, when bearings 861, 862 are normal, the load (i.e., weight) increases at a constant gradient before returning to its initial value, forming a sawtooth waveform. When bearings 861, 862 deteriorate, a protruding waveform appears as the load increases at a constant gradient due to resistance factors that hinder smooth rotation, such as increased surface roughness or deformation. Therefore, frequency analysis is performed on the change in motor rotation angle θm, i.e., the rotation speed. When bearings 861, 862 are normal, the rotation speed is constant, and only the peak of the zeroth-order component appears in the frequency spectrum data.

[0038] When bearings 861 and 862 deteriorate due to wear or deformation, peaks of specific frequency components appear in one or more locations depending on the position and magnitude of the wear deformation and the rotation frequency. For example, the first, second, and third frequency components other than the zeroth-order component in descending order of peak value are designated as fp1, fp2, fp3, and so on. For example, the peak value of the zeroth-order component is used as a reference, and the intensity ratios of the frequency components fp1, fp2, and fp3 at which peaks appear are evaluated. The frequency components and intensity ratios are compared with a trained model, and the deterioration state of bearings 861 and 862 is determined.

[0039] Next, the configuration of the failure prediction system of the first to third embodiments will be described with respect to the functional arrangement of the learning unit that creates a trained model through machine learning and the failure prediction unit that monitors sensor values ​​during operation of the motor control device and performs failure prediction. Substantially identical components in multiple embodiments will be assigned the same reference numerals and descriptions thereof will be omitted.

[0040] Here, the reference numeral "40L" is used to designate the motor control device used to create learning data. The reference numerals "401," "402," and "403" are used to designate the motor control devices of the first to third embodiments whose sensor values ​​are monitored during operation. The reference numerals for the microcomputers of each motor control device are "50L," "501," "502," and "503," corresponding to the reference numerals of the motor control devices.

[0041] (First embodiment) Figures 5 and 6 show a failure prediction system of the first embodiment. In the first embodiment, a learning system (Figure 5) that is installed in a laboratory or factory and creates a learned model LM is separated from a monitoring system (Figure 6) that is installed in a vehicle or transport vehicle and monitors the sensor values ​​of the motor control device 401 during operation. Note that Figure 5 is also used in the second embodiment.

[0042] In the learning system, a plurality of motor control devices 40L for creating learning data are subjected to trials of energization and rotation of the motor 80. A data acquisition unit 51 provided in the microcomputer 50L of each motor control device 40L acquires data on the inverter current Im from the trial energization and data on the motor rotation angle θm from the trial rotation of the motor 80.

[0043] 5, the learning system includes a learning device 200 dedicated to machine learning. The learning device 200 includes an FFT processing unit 22 and a learning unit 23 provided in a PC 20. The FFT processing unit 22 performs FFT processing on the inverter current Im and the motor rotation angle θm acquired from the data acquisition unit 51 of the motor control device 40L, and creates learning data of the frequency spectrum.

[0044] In order to improve the learning accuracy, it is preferable to create a large amount of learning data. In the first embodiment, the learning unit 23 that processes a large amount of data is provided outside the motor control device 401, so the processing load of the microcomputer 501 of the motor control device 401 can be reduced.

[0045] The learning unit 23 performs machine learning by associating learning data of the frequency spectrum, which is a feature, with the deterioration state of the failure prediction target to create a learned model LM. The learning unit 23 may perform learning according to a neural network model.

[0046] The learning data of the frequency spectrum regarding the deterioration state of the power transistors 61-66 is created by frequency analyzing the inverter current Im in a trial of energizing a plurality of motor control devices 40L. The learning unit 23 performs machine learning by associating the learning data of the frequency spectrum with the deterioration state of the power transistors 61-66, and creates a learned model LM.

[0047] Furthermore, the learning data of the frequency spectrum regarding the deterioration state of the bearings 861, 862 is created by frequency analyzing the change in the motor rotation angle θm during trial rotation of the motor 80 by a plurality of motor control devices 40L. The learning unit 23 associates the learning data of the frequency spectrum with the deterioration state of the bearings 861, 862 and performs machine learning to create a learned model LM.

[0048] The learned model LM created by the learning unit 23 of the learning device 200 is stored in the microcomputer 501 of the motor control device 401 in the monitoring system shown in Figure 6. The microcomputer 501 of the motor control device 401 has, in addition to the data acquisition unit 51, an FFT processing unit 52 and a failure prediction unit 54. In other words, the failure prediction unit 54 is provided inside the microcomputer 501 of the motor control device 401.

[0049] The FFT processing unit 52 performs frequency analysis on the sensor values ​​acquired by the data acquisition unit 51 from the current sensor 70 and the rotation angle sensor 78 while the motor 80 is energized and rotating, and creates frequency spectrum data.

[0050] The failure prediction unit 54 compares the frequency spectrum data obtained by the FFT processing unit 52 performing frequency analysis on the inverter current Im with the learned model LM to determine the deterioration state of the power transistors 61-66. The failure prediction unit 54 also compares the frequency spectrum data obtained by the FFT processing unit 52 performing frequency analysis on changes in the motor rotation angle θm with the learned model LM to determine the deterioration state of the bearings 861, 862. When the failure prediction unit 54 predicts a failure of the power transistors 61-66 or the bearings 861, 862, it notifies the user via communication within the vehicle or within the device.

[0051] Second Embodiment Figures 5 and 7 show an example of the configuration of a failure prediction system according to a second embodiment. As in the first embodiment, the second embodiment is separated into a learning system (Figure 5) that is installed in a laboratory or factory and creates a trained model LM, and a monitoring system (Figure 7) that monitors the sensor values ​​of the motor control device 402 during operation. The explanation of the learning system in Figure 5 is the same as that of the first embodiment.

[0052] As shown in Figure 7, in the monitoring system of the second embodiment, a failure prediction device 300 provided in the data management unit and a microcomputer 502 of a motor control device 402 mounted on a vehicle or a transport vehicle communicate signals via communication devices 39, 49. The data management unit is a communication base station that controls the vehicle and manages the transport vehicle, and is provided with the failure prediction device 300 and the communication device 39. The vehicle or transport vehicle is equipped with a motor control device 402 and a communication device 49.

[0053] The microcomputer 502 of the second embodiment only needs to have at least the data acquisition unit 51 of the configuration related to failure prediction that the microcomputer 501 (FIG. 6) of the first embodiment has. That is, the microcomputer 502 of the second embodiment can have the same configuration as the microcomputer 50L (FIG. 5) of the motor control device 40L used to create learning data. The data on the inverter current Im and the motor rotation angle θm acquired by the data acquisition unit 51 is communicated from the communication device 49 of the vehicle or transport vehicle to the failure prediction device 300 via the communication device 39 of the data management unit.

[0054] The failure prediction device 300 includes an FFT processing unit 32 and a failure prediction unit 34. The learned model LM created by the learning unit 23 of the learning device 200 is stored in the failure prediction device 300. The FFT processing unit 32 and the failure prediction unit 34 of the failure prediction device 300 function in the same way as the FFT processing unit 52 and the failure prediction unit 54 of the microcomputer 501 in the first embodiment, and determine the deterioration state of the power transistors 61-66 or the bearings 861, 862 and predict failures.

[0055] When the failure prediction unit 54 predicts a failure in the power transistors 61-66 or the bearings 861, 862, it transmits a failure prediction signal to the microcomputer 502 of the motor control device 402 and also notifies a user who may be near the data management unit. Upon receiving the failure prediction signal, the microcomputer 502 notifies the user via communication within the vehicle or device.

[0056] As described above, in the second embodiment, the failure prediction unit 34 is provided outside the motor control device 402 and communicates signals with the motor control device 402. This can further reduce the amount of processing by the microcomputer 502 of the motor control device 402. Furthermore, when applied to a transport vehicle in a factory, for example, the data management unit can collectively manage the deterioration states of the motor control devices of the transport vehicles that are the targets of failure prediction, making it easy to respond to cases where the learned model LM is updated.

[0057] (Third embodiment) Figure 8 shows a failure prediction system of the third embodiment. In the third embodiment, a learning unit 53 is provided in addition to an FFT processing unit 52 and a failure prediction unit 54 inside a microcomputer 503 of a motor control device 403. The learned model LM created by the learning unit 53 is stored in the microcomputer 503. In short, the microcomputer 503 of the third embodiment has an all-in-one configuration that includes all the components related to failure prediction. It is suitable for a microcomputer 503 equipped with a CPU with a large processing capacity.

[0058] The microcomputer 503 of the third embodiment can be used as a learning system installed in a laboratory or factory, or as a monitoring system mounted on a vehicle or transport vehicle. However, when used as a learning system, the microcomputer 503 only needs to have the functionality to create the learned model LM, and does not need the failure prediction unit 54. Therefore, as a modification of the third embodiment, a motor control device equipped with a microcomputer that does not include the failure prediction unit 54 may be used as a motor control device for creating learning data.

[0059] When used as a learning system, the learning unit 53 of the motor control device 403 may perform learning as a motor control device for creating learning data. In this case, for example, a single motor control device 403 may perform multiple trials of energizing and rotating the motor 80.

[0060] Regarding the deterioration state of the power transistors 61-66, learning data of the frequency spectrum is created by frequency analyzing the inverter current Im during multiple current-flow trials by the motor control device 403. Similarly, regarding the deterioration state of the bearings 861, 862, learning data of the frequency spectrum is created by frequency analyzing changes in the motor rotation angle θm during multiple trials by the motor control device 403 to rotate the motor 80.

[0061] In this learning method, a learned model LM is created that corresponds to the characteristics of each individual motor control device 403, but the learning process takes time. Furthermore, there is a possibility that the motor control device 403 may actually deteriorate during the learning process. Therefore, as in the first embodiment, the learned model LM created by the learning device 200 may be stored in the microcomputer 503 of the motor control device 403 in the monitoring system.

[0062] However, the microcomputer 503 can not only use the initially saved learned model LM as is, but also have the learning unit 53 perform additional learning and update the learned model LM based on data acquired during operation of the motor control device 403. In the monitoring system, when the failure prediction unit 54 of the microcomputer 503 predicts a failure of the power transistors 61-66 or the bearings 861, 862, it notifies the user via communication within the vehicle or within the device.

[0063] [Algorithm of the Failure Prediction System] Next, the algorithm executed by the failure prediction system will be explained with reference to the flowcharts in Figures 9 to 11. Figure 9 shows the learning algorithm used by the learning system. Figure 11 shows the algorithm used by the monitoring system to predict failures during product operation. The sub-flowchart in Figure 10 shows a detailed algorithm for S3 "Creating frequency spectrum data" in Figures 9 and 11. In the explanation of each flowchart, the symbol S means a step.

[0064] The "product" in the flowchart refers to a mechanically and electrically integrated motor control device. Here, the motor control device 401 and each part are designated by symbols in accordance with the configuration of the failure prediction system of the first embodiment. The basic algorithm is the same in the failure prediction systems of the second and third embodiments, and they can be interpreted in the same way with appropriate interchangeable symbols.

[0065] As shown in Figure 9, in S1 of "learning," a normal product and multiple products in stages of deterioration are prepared. In S3, frequency spectrum data is created as feature quantities when electricity is applied to the product and the motor is rotated. This data is learning data. In S4, the learning unit 23 performs machine learning by associating the frequency spectrum data with the deterioration state of the power transistors 61-66 or the bearings 861, 862, and creates a learned model.

[0066] As shown in FIG. 10 , S3 of "creating frequency spectrum data" includes S31 to S34. In S31, the data acquisition unit 51 acquires sensor values ​​from the current sensor 70 and the rotation angle sensor 78 at predetermined times. "Acquiring at predetermined times" may mean, for example, that the sensor values ​​are acquired at constant intervals from start-up to stop of the motor control device 40. Alternatively, the sensor values ​​may be acquired at relatively short intervals during a predetermined period immediately after startup or immediately after a change in drive conditions, and at relatively long intervals during other periods. In S32, the FFT processing unit 52 performs FFT processing, i.e., frequency analysis, based on the acquired sensor values.

[0067] In S33, spectrum data of the rotation frequency and Nth-order harmonic frequency components is created from the FFT results of the inverter current Im. In S34, frequency spectrum data is created for each frequency component at which a peak appears, including the frequency component with the largest peak value other than the zeroth-order component, from the FFT results of the change in the motor rotation angle θm. In this case, data may be created for frequency components with the largest peak value in a predetermined order. Alternatively, the number of frequency components may not be determined, and data may be created for frequency components with peak values ​​equal to or greater than a predetermined percentage of the maximum peak value, for example.

[0068] As shown in Figure 11, in S2 of "Failure prediction during product operation," the motor control device 401 is operating normally as a product. In S3, frequency spectrum data is created as a feature quantity during current flow and motor rotation during normal product operation. In S5, the failure prediction unit 54 compares the frequency spectrum data with the learned model LM and estimates the state.

[0069] As a result, the determination is classified as "not deteriorated" or "deteriorated." Alternatively, "deteriorated" may be further classified into multiple stages such as "deterioration level 1," "deterioration level 2," etc. In S6, it is determined whether the determination result is "deteriorated" (or deterioration level 1 or higher), and if NO, the process returns to before S2 and repeats. If YES in S6, it is determined in S7 whether the "deteriorated" determination is the first time or has reached a predetermined number of times. Note that being the first time corresponds to reaching the predetermined number of times when the predetermined number is set to 1. If NO in S7, the process returns to before S2 and repeats.

[0070] If the answer is YES in S7, the failure prediction unit 54 notifies the user, a vehicle management unit, or the like of the predicted failure in S8. This allows the user to grasp the symptoms of the failure before an actual failure occurs and the motor control device 40 loses its functionality, and to take appropriate measures such as part replacement or maintenance. This improves the durability and reliability of the product.

[0071] (Other Embodiments) (a) The method of determining the degradation state of the inverter elements (power transistors 61-66) based on the inverter current Im and predicting failure according to the above embodiment can also be extended to predict failure of "elements around the inverter provided on the current path from the power supply to the motor 80 via the inverter 60." A specific example of the "elements around the inverter" will be described with reference to FIG. 12. Each of the semiconductor relays 43, 44, 67-69 is configured, for example, with a MOSFET.

[0072] 12, DC power is input from a battery BT as a "power supply" to an inverter 60. The high-potential side of the inverter 60 is connected to the positive electrode of the battery BT via a power supply line Lp, and the low-potential side of the inverter 60 is connected to the negative electrode of the battery BT via a ground line Lg. A power supply relay 43 and a reverse connection protection relay 44 are connected in series to the power supply line Lp.

[0073] The power supply relay 43 has a parasitic diode connected in parallel that conducts current from the inverter 60 side to the battery BT side, and when it is turned off, it cuts off the current from the battery BT side to the inverter 60 side. The reverse connection protection relay 44 has a parasitic diode connected in parallel that conducts current from the battery BT side to the inverter 60 side, and when it is turned off, it cuts off the current from the inverter 60 side to the battery BT side. This prevents current from flowing from the low potential side to the high potential side of the inverter 60 when the positive and negative electrodes of the battery BT are connected in reverse.

[0074] Motor relays 67, 68, and 69 are connected to current paths between the inter-arm connection points Nu, Nv, and Nw of each phase of the inverter 60 and the windings of the motor 80. Parasitic diodes of the motor relays 67, 68, and 69 conduct current from the inter-arm connection points Nu, Nv, and Nw to the windings of the motor 80. When turned off, the motor relays 67, 68, and 69 cut off current from the motor 80 side to the inverter 60 side. Drive signals from the microcomputer 50 to the power transistors 61-66 and the relays 43, 44, and 67-69 are not shown.

[0075] A capacitor 46 for smoothing the input voltage is connected between the power supply line Lp and the ground line Lg on the input side of the inverter 60 .

[0076] These relays 43, 44, 67-69 and capacitor 46 correspond to examples of "inverter peripheral elements provided in the current path from the power supply to the motor 80 via the inverter 60." As with the inverter element, when the inverter peripheral elements are in a deteriorated state that indicates a failure, distortion occurs in the waveform of the inverter current Im, and Nth-order harmonic frequency components tend to appear in the frequency spectrum. Therefore, the description of "failure prediction of the inverter element" in the above embodiment can be expanded to "failure prediction of the inverter element or the inverter peripheral elements."

[0077] (b) The "bearing parts supporting the shaft 84 of the motor 80" are not limited to ball bearings such as the front bearing 861 and rear bearing 862 illustrated in FIG. 2, but may be any parts that have bearing functions, such as lubricating resin rings.

[0078] (c) As a method for evaluating frequency analysis data, in the example shown in Fig. 3 of the above embodiment, the intensity ratio of the frequency component of the Nth harmonic to the rotation frequency is evaluated in the frequency spectrum of the inverter current Im. In the example shown in Fig. 4, in the frequency spectrum of the change in the motor rotation angle θm, the intensity ratio of each frequency component at which a peak appears is evaluated, for example, with the peak value of the zeroth order component as the reference.

[0079] However, the evaluation method of frequency analysis data is not limited to evaluating the intensity ratio of multiple frequency components. For example, it is also possible to focus only on the third harmonic of the inverter current Im and evaluate the ratio between the peak value of the third harmonic and the inverter current Im. Alternatively, it is also possible to focus only on the frequency component other than the zeroth-order component that has the largest peak value in the trained model regarding changes in the motor rotation angle θm and evaluate the ratio between the peak value and the motor rotation speed.

[0080] (d) In the above embodiment, the inverter element failure prediction based on the inverter current Im and the bearing component failure prediction based on the motor rotational angle θm are performed independently using separate trained models. However, this is not limiting, and the failure prediction unit may predict failures using a trained model that is trained by combining information on the inverter current Im and the motor rotational angle θm.

[0081] Furthermore, the motor control device 40 may include sensors for detecting other physical quantities in addition to the current sensor 70 and the rotational angle sensor 78, and the failure prediction unit may predict a failure using a trained model trained by combining the inverter current Im or the motor rotational angle θm with other sensor values. For example, for inverter elements or inverter peripheral elements, a trained model trained by combining the inverter current Im and a substrate temperature detected by a temperature sensor may be used. For bearing components, a trained model trained by combining the motor rotational angle θm and shaft vibration detected by a vibration sensor may be used.

[0082] (e) The motor control device that is the target of failure prediction is not limited to the electromechanical integrated type shown in Figure 2, but may be an electromechanical separate type in which the corresponding mechanical part and the drive control part are separated. Also, it is not limited to a three-phase motor, but may be a multi-phase motor with four or more phases.

[0083] As described above, the present disclosure is not limited to the above-described embodiments, and can be embodied in various forms without departing from the spirit of the present disclosure.

[0084] Each control unit (learning unit, failure prediction unit, etc.) and its method described in this disclosure may be implemented by a special-purpose computer provided by configuring a processor and memory programmed to execute one or more functions embodied in a computer program. Alternatively, each control unit and its method described in this disclosure may be implemented by a special-purpose computer provided by configuring a processor with one or more dedicated hardware logic circuits. Alternatively, each control unit and its method described in this disclosure may be implemented by one or more special-purpose computers configured by combining a processor and memory programmed to execute one or more functions with a processor configured with one or more hardware logic circuits. Furthermore, the computer program may be stored as instructions executed by a computer on a computer-readable non-transitory tangible recording medium.

[0085] The present disclosure has been described based on the embodiments. However, the present disclosure is not limited to the embodiments and structures. The present disclosure also encompasses various modifications and variations within the scope of equivalents. Furthermore, various combinations and forms, including only one element, more than one element, or less than one element, are also within the scope and spirit of the present disclosure.

Claims

1. A motor control device (40) including an inverter (60) that supplies power to a motor (80), a current sensor (70) that detects an inverter current, which is a current of one or more phases flowing through the inverter, a rotation angle sensor (78) that detects a rotation angle of the motor, and a microcomputer (50) that acquires sensor values ​​from the current sensor and the rotation angle sensor and controls the inverter, the system determining a deterioration state that is a sign of a failure and predicting a failure for inverter elements (61-66) that switch the power supply to the inverter, or inverter peripheral elements (43, 44, 46, 67-69) that are provided on a power supply path from a power source (BT) to the motor via the inverter, comprising: a learning unit (23, 53) that performs machine learning by associating learning data obtained by frequency analysis of the inverter current during a trial of power supply to the motor control device with the deterioration state of the inverter elements or the inverter peripheral elements, and creating a trained model; a failure prediction unit (34, 54) that compares data obtained by frequency analysis of the inverter current acquired by the microcomputer from the current sensor while the motor is energized with the learned model, and determines the deterioration state of the inverter element or the inverter peripheral element.

2. The failure prediction system of claim 1, wherein the failure prediction unit compares the intensity ratio of the frequency component of the Nth harmonic to the rotation frequency in the frequency spectrum of the inverter current with the learned model to determine the deterioration state of the inverter element or the inverter peripheral elements.

3. A system for determining a deterioration state that is a sign of failure and predicting failure of bearing components (861, 862) that support a shaft (84) of the motor in a motor control device (40) including an inverter (60) that supplies power to a motor (80), a current sensor (70) that detects an inverter current, which is a current of one or more phases flowing through the inverter, a rotation angle sensor (78) that detects the rotation angle of the motor, and a microcomputer (50) that acquires sensor values ​​from the current sensor and the rotation angle sensor and controls the inverter, the system comprising: a learning unit (23, 53) that performs machine learning to associate learning data obtained by frequency analysis of changes in the rotation angle of the motor during trial rotation of the motor by the motor control device with the deterioration state of the bearing components and creates a trained model; and a failure prediction unit (34, 54) that compares data obtained by frequency analysis of changes in the rotation angle of the motor obtained by the microcomputer from the rotation angle sensor while the motor is rotating with the trained model and determines the deterioration state of the bearing components.

4. A failure prediction system as described in claim 3, wherein the failure prediction unit compares the intensity ratio of each frequency component at which a peak appears in the frequency spectrum of the change in the rotation angle of the motor with the trained model to determine the deterioration state of the bearing parts.

5. A failure prediction system according to any one of claims 1 to 4, wherein the failure prediction unit (54) is provided inside the microcomputer (501, 503) of the motor control device (401, 403).

6. A failure prediction system according to claim 5, wherein the learning unit (53) is further provided inside the microcomputer (503) of the motor control device (403).

7. A failure prediction system according to any one of claims 1 to 4, wherein the failure prediction unit (34) is provided outside the motor control device (402) and communicates signals with the motor control device.

Images