Machine learning device, insulation state diagnosis device, machine learning method, machine learning program, insulation state diagnosis method, and insulation state diagnosis program

Abstract

The present invention provides a machine learning device capable of diagnosing an insulation state of a winding without stopping a canned motor pump. A machine learning device 4 generates a training model 6 to be used by an insulation state diagnosis device 5 for diagnosing an insulation state of a winding 31 of a stator 23 used in a canned motor pump 2. The machine learning device 4 trains the training model 6 to learn a correlation between input data and information on diagnosis of the insulation state of the winding by inputting training data including at least the input data into the training model, the input data being unprocessed or at least partially preprocessed data on a voltage value and current value supplied to a motor part 20, a temperature of the winding 31 of the stator 23 or a correlated temperature, temperature difference, or pressure, and an inflow temperature and flow rate of a working fluid in a pump part 10.

Description

Machine learning device, insulation condition diagnosis device, machine learning method, machine learning program, insulation condition diagnosis method, insulation condition diagnosis program

[0001] The present invention relates to a machine learning device, an insulation condition diagnosis device, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program that are suitable for diagnosing the insulation condition of the windings of the motor section of a canned motor pump that has a motor section and a pump section.

[0002] In a canned motor pump, the rotor of the motor that drives the pump is housed in a can filled with the pump fluid. The stator of the motor is housed in a motor housing that surrounds the can and functions as a housing. In such canned motor pumps, the motor output is set based on the allowable temperature of the stator coil. That is, the rotor of a canned motor pump is quickly cooled because it is surrounded by the pump fluid. Therefore, heat generation in canned motor pumps occurs primarily on the stator side, with the coil end being the hottest. Therefore, conventionally, the motor output and fluid temperature have been specified so that the coil end temperature does not exceed the allowable temperature. In addition, canned motor pumps typically guide a portion of the pump fluid into the can where the rotor is located to lubricate and cool the bearings and absorb heat from the motor (rotor, etc.) before returning it to the pump.

[0003] However, canned motor pumps are often used in plants handling high-temperature liquids. When high-temperature liquid is introduced into the can, the motor windings become hotter than the liquid and are easily exposed to temperatures higher than the heat-resistant insulation can tolerate. Therefore, the motor of a canned motor pump is susceptible to insulation deterioration due to long-term use in high-temperature environments. As insulation deterioration progresses, the windings must be replaced. However, because the stator is sealed between the can and the motor housing, replacing the stator windings requires disassembly of the sealed motor. Because such disassembly requires significant costs and labor, it is necessary to properly assess the lifespan of the windings, i.e., the condition of their insulation.

[0004] Therefore, in the past, in order to diagnose deterioration of the windings, the canned motor pump was stopped and the insulation resistance and inter-terminal resistance of the windings were measured periodically, and these values ​​were compared with normal values ​​and past measured values ​​to manage the trend of the insulation condition (see Patent Document 1).

[0005] JP 2010-256348 A

[0006] However, in order to monitor the insulation condition of the windings, it is necessary to stop the canned motor pump and measure the resistance values ​​(insulation resistance, terminal-to-terminal resistance), and it is difficult to stop the canned motor pump frequently while the plant is operating.

[0007] The present invention has been made in consideration of the above circumstances, and its main objective is to provide a machine learning device, an insulation condition diagnosis device, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program that enable the diagnosis of the insulation condition of a winding without stopping a canned motor pump.

[0008] a canned motor pump that circulates a portion of a handled liquid in the pump section to the rotor chamber; a learning data acquisition unit that acquires a plurality of sets of learning data including, as input data, unprocessed data or data that has been at least partially preprocessed, the input data being data of voltage and current values ​​supplied to the motor section, data of the temperature of the stator winding or a temperature, temperature difference, or pressure correlated thereto, and data of the inflow temperature and flow rate of the handled liquid; a machine learning unit that inputs the learning data to the learning model, thereby causing the learning model to learn a correlation between the input data and diagnostic information on the insulation state of the winding; and a trained model storage unit that stores the trained model trained by the machine learning unit.

[0009] An insulation condition diagnosis device according to one embodiment of the present invention is an insulation condition diagnosis device for diagnosing the insulation condition of a stator winding used in a canned motor pump using a learning model generated by the above-mentioned machine learning device, and includes: an input data acquisition unit that acquires input data including unprocessed data or data that has been at least partially preprocessed, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; and an inference unit that inputs the input data acquired by the input data acquisition unit into the learning model and infers diagnostic information for the insulation condition of the stator winding.

[0010] Furthermore, one embodiment of the machine learning method according to the present invention is a machine learning method for learning a learning model to be used in an insulation condition diagnosis device for diagnosing the insulation condition of a winding of a stator used in a canned motor pump that circulates a portion of a handled liquid in the pump section to the rotor chamber, the machine learning method comprising: a motor section; and a pump section driven by the motor section, the motor section comprising: a rotor rotatable about a shaft; a stator facing the rotor with a gap therebetween; and a cylindrical can that defines a rotor chamber on the inside for accommodating the rotor and, together with a motor housing, defines a stator chamber on the outside for accommodating the stator; the machine learning method includes a learning data acquisition step of acquiring a plurality of sets of learning data comprising at least the input data, the input data being data on voltage and current values ​​supplied to the motor section, data on the temperature of the stator winding or data on a temperature, temperature difference, or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid, the data being unprocessed data or data that has been at least partially preprocessed; The method includes a machine learning process of inputting the learning data into the learning model to cause the learning model to learn the correlation between the input data and diagnostic information on the insulation state of the winding, and a learned model storage process of storing the learning model learned by the machine learning process in a learned model storage unit.

[0011] Furthermore, a machine learning program according to an embodiment of the present invention causes a computer to execute each step of the machine learning method.

[0012] An insulation condition diagnosis method according to one embodiment of the present invention is an insulation condition diagnosis method for diagnosing an insulation condition of a stator winding used in the canned motor pump using a learning model generated by the machine learning device, the insulation condition diagnosis method comprising: an input data acquisition step of acquiring input data including unprocessed data or data having at least a portion thereof subjected to preprocessing, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; and an inference step of inputting the input data acquired by the input data acquisition step into the learning model and inferring diagnostic information on the insulation condition of the stator winding.

[0013] An insulation state diagnosis program according to an embodiment of the present invention causes a computer to execute each step of the insulation state diagnosis method.

[0014] According to the present invention, it is possible to provide a machine learning device, an insulation condition diagnosis device, a machine learning method, a machine learning program, an insulation condition diagnosis method, and an insulation condition diagnosis program that enable the diagnosis of the insulation condition of a winding without stopping a canned motor pump.

[0015] 1 is a schematic configuration diagram showing an insulation state diagnosis device applied to a canned motor pump. FIG. 1 is a cross-sectional view showing an example of a canned motor pump. FIG. 2 is a block diagram showing an example of a machine learning device. FIG. 3 is a flowchart showing an example of a machine learning method by the machine learning device according to the first embodiment. FIG. 4 is a block diagram showing an example of an insulation state diagnosis device. FIG. 5 is a flowchart showing an example of an insulation state diagnosis method by the insulation state diagnosis device according to the first embodiment. FIG. 6 is a cross-sectional view showing another example of a canned motor pump. FIG. 7 is a block diagram showing an example of a machine learning device applied to the canned motor pump shown in FIG. 7. FIG. 7 is a block diagram showing an example of an insulation state diagnosis device applied to the canned motor pump shown in FIG.

[0016] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] The present invention will be described in detail below with reference to the accompanying drawings. First Embodiment

[0017] Fig. 1 is a schematic diagram showing an insulation condition diagnosis system 1 applied to a canned motor pump. The canned motor pump 2 used here is, as an example, a horizontally-mounted canned motor pump in which the motor axis is generally horizontal. However, the present invention is not limited to horizontally-mounted canned motor pumps, and can also be applied to vertically-mounted canned motor pumps in which the motor axis is generally vertical.

[0017] First, an overview of the canned motor pump 2 will be described using Figure 2. Figure 2 is a cross-sectional view showing an example of the configuration of the canned motor pump 2 according to this embodiment. Note that the X direction shown in the figure is the motor axial direction, with the right side of the figure being the front and the left side being the rear.

[0018] In FIG. 2, the canned motor pump 2 has a pump section 10 having the configuration of a centrifugal pump and a motor section 20 that drives the pump section 10, and the pump section 10 and the motor section 20 are coaxially coupled by a shaft 21 of a motor 24, which will be described later.

[0019] The pump section 10 has an impeller 11 and a pump casing 13 that forms a pump chamber 12 that houses the impeller 11. The impeller 11 is coupled to the front end of a shaft 21 of a motor 24. Note that, in this embodiment, the pump section 10 is described as being configured as a centrifugal pump, but the present invention is not limited to this, and the pump section 10 may also be configured as a turbo pump such as a mixed flow pump or an axial flow pump.

[0020] The motor section 20 has a motor 24 including a rotor 22 fixed around the shaft 21 and rotating integrally with the shaft, an annular stator 23 arranged to surround the rotor 22 at a predetermined distance, and a motor housing 25 that accommodates the motor 24.

[0021] The stator 23 has a stator core 30 having teeth arranged circumferentially on its substantially cylindrical inner peripheral surface, and windings (coils) 31 formed by winding conductors around the teeth of the stator core 30. Ends of the conductors of the windings 31 are led into a terminal box 32 extending radially outward from an outer cylinder 41 of the motor housing 25, and connected to terminals (not shown) provided in the terminal box 32.

[0022] A cylindrical can 33 is disposed inside the stator core 30 but outside the rotor 22, i.e., between the rotor 22 and the stator 23. The can 33 is fixed in contact with the inner peripheral surface of the stator core 30, i.e., the tip surfaces of the teeth, and is disposed with a fixed gap between it and the outer peripheral surface of the rotor 22.

[0023] A cylindrical outer cylinder (stator band) 41 that houses the motor is provided along the outer peripheral surface of the stator core 30. The can 33 and the outer cylinder 41 are arranged concentrically and have approximately the same length in the axial direction X. Annular plate-shaped end bells 42, 43 are arranged on both ends of the can 33 and the outer cylinder 41, respectively, and these end bells 42, 43 close the cylindrical space formed between the can 33 and the outer cylinder 41. The end bell 42 on the pump section 10 side is connected to the pump casing 13 with bolts. A bearing holder 44 that closes the space inside the can 33 is connected to the end bell 43 on the opposite side from the pump section 10 with bolts.

[0024] The outer cylinder 41, the end bells 42 and 43 at both ends, and the bearing holder 44 form a motor housing 25 that houses the motor 24. The can 33, the pump section 10, and the bearing holder 44 form a rotor chamber 50 that houses the rotor 22 inside the can 33. The cylindrical space defined by the outer cylinder 41, the can 33, and the end bells 42 and 43 at both ends defines a stator chamber 45 that houses the stator 23. The stator chamber 45 has the stator core 30 located approximately in the axial center, and has annular coil end housing portions 45a and 45b at the front and rear of the stator core 30 that house the coil ends 31a and 31b. Cylindrical backup sleeves 34 (34a and 34b) are attached to the outer peripheral surface of the can 33 facing the coil end housing portions 45a and 45b to prevent deformation of the can 33.

[0025] Shaft 21, on which rotor 22 is integrally fixed, is supported at both ends by bearings 51, 52. Bearing 51 on the pump section 10 side is held by bearing portion 53 extending from bearing holder 44 fixed to end bell 42 behind pump casing 13 of pump section 10. Bearing 52 on the opposite side from pump section 10 is held by bearing portion 54 extending from bearing holder 46 fixed to end bell 43. These bearings 51, 52 are constituted by plain bearings positioned in bearing portions 53, 54.

[0026] The canned motor pump 2 is fixed to a support base 56 by legs 55 fixed to the end bells 42 and 43 .

[0027] The bearing holder 44, fixed to the rear of the pump casing 13 of the pump section 10, is formed with an inlet passage 14 for guiding the treated liquid into the can 33. The shaft 21 also has a central hole 15 formed along its entire length along its axis. A portion of the treated liquid delivered by the impeller 11 becomes internal circulating liquid circulating within the can 33, as indicated by the arrows in FIG. 2 . That is, the liquid flows around behind the impeller 11 and is introduced into the can 33 (rotor chamber 50) through the inlet passage 14. The treated liquid introduced into the can 33 (rotor chamber 50) lubricates and cools the bearings 51 and 52 and the motor section 20 (rotor 22, etc.), before entering the central hole 15 formed in the shaft from the rear end of the shaft 21. The treated liquid is then returned to the suction side of the pump section 10 via the central hole 15 and delivered to the vortex chamber by the impeller 11.

[0028] An insulation condition diagnosis system 1 is provided for the canned motor pump 2 having the above-described configuration, which diagnoses the insulation condition of the windings 31 of the stator 23. As shown in FIG. 1 , this insulation condition diagnosis system 1 includes a measurement device 3, a machine learning device 4, and an insulation condition diagnosis device 5.

[0029] The measuring device 3 is installed in an appropriate location in or around the canned motor pump 2, and measures physical quantities and state quantities of each part. In the learning phase, the measuring device 3 is connected to the machine learning device 4 and used to measure data for learning, and in the inference phase, the measuring device 3 is connected to the insulation state diagnosis device 5 and used to measure data for diagnosis. Here, the measuring device 3 used in the learning phase and the inference phase may be the same or different, and includes a motor voltage / current sensor 101, a winding heat generation state measurement sensor 102, and a handled liquid measurement sensor 103.

[0030] The motor voltage / current sensor 101 measures the motor voltage and current values ​​supplied to the motor unit 20, and is attached, for example, to a power line connecting the motor 24 and the motor control device 26, and measures the voltage and current values ​​supplied from the motor control device 26 to the motor 24. The motor voltage / current sensor 101 may be provided inside the motor unit 20 or inside the motor control device 26. The motor voltage / current sensor 101 may also measure the voltage and current values ​​of each of the three phases.

[0031] The winding heat generation state measuring sensor 102 directly or indirectly measures the temperature of the windings 31 of the motor 24 (stator 23). That is, the winding heat generation state measuring sensor 102 measures the temperature of the windings or data on temperatures, temperature differences, or pressures correlated therewith. When measuring the temperature of the windings 31 directly, a thermocouple is attached to the hottest portion of the windings 31, for example, the coil ends 31a and 31b, to measure the temperature. Alternatively, a thermocouple may be attached to the windings 31 inside the stator core 30 to measure the winding temperature. When measuring the temperature of the windings 31 indirectly, a thermocouple may be attached to the backup sleeves 34a and 34b, the end bells 42 and 43, or the bearing holders 44 and 46, which are affected by radiant heat from the windings 31 and the state (amount, temperature) of the internal circulating fluid, to measure the surface temperatures of those portions. Although the winding heat generation state measuring sensor 102 detects the temperature using a thermocouple in the above example, other contact sensors such as a thermistor or a resistance temperature detector may be used, or a non-contact sensor such as an infrared sensor may be used. Furthermore, the pressure P1 in the stator chamber 45 may be measured to indirectly measure the temperature of the winding 31.

[0032] The handled liquid measurement sensor 103 is composed of, for example, an inflow temperature sensor 103a and a flow rate sensor 103b provided in the inflow pipe 60 connected to the inlet 13a of the pump casing 13. The inflow temperature sensor 103a measures the temperature of the handled liquid flowing into the pump section 10, and measures the fluid temperature in the inflow pipe 60 by, for example, a piping temperature sensor using a resistance temperature detector as the temperature detection element. The flow rate sensor 103b measures the inflow rate of the handled liquid flowing into the pump section 10, and various types of sensors can be used, such as ultrasonic, electromagnetic, Karman vortex, impeller, float, thermal, and differential pressure sensors.

[0033] The measurement device 3 is configured to measure physical quantities and state quantities of each part and output the measurement values ​​to the machine learning device 4 or the insulation state diagnosis device 5. The measurement values ​​measured by this measurement device 3 may be pinpoint measurement values ​​measured at a predetermined timing, a group of discrete time-dependent measurement values ​​measured at predetermined measurement intervals over a predetermined period of time, or continuous measurement values ​​measured over a predetermined period of time.

[0034] The machine learning device 4 operates as the main body of the learning phase, and generates, by machine learning, a learning model 6 that is used when diagnosing the insulation state of the winding 31. Either "supervised learning" or "unsupervised learning" can be used for machine learning, and "supervised learning" is used in the first embodiment, while "unsupervised learning" is used in the second embodiment.

[0035] The insulation condition diagnosis device 5 operates as the main body of the inference phase, and diagnoses (infers) the insulation condition of the winding 31 from the measured values ​​measured by the measuring device 3 using the trained learning model 6 generated by the machine learning device 4.

[0036] (Machine Learning Device) The machine learning device 4 is configured by a general-purpose or dedicated computer. This computer may be a desktop computer or a portable computer, and may also be a client computer, a server computer, or a cloud computer. As shown in FIG. 3 , the machine learning device 4 includes a control unit 111, an operation unit 112, a display unit 113, a communication unit 114, a media input / output unit 115, a RAM 116, a storage unit 117, a bus 118, etc. The various units of the machine learning device 4 are connected via the bus 118.

[0037] The control unit 111 is composed of an arithmetic processing device (a processor such as a CPU, MPU, or DSP), and controls the operation of each unit of the machine learning device 4 by reading and executing programs loaded into the RAM 116 and performing various arithmetic processing. Note that the control unit 111 may have multiple processors, and the various processes of this embodiment may be executed by the multiple processors. In this case, the multiple processors may be involved in a common process, or the multiple processors may independently execute different processes in parallel.

[0038] The operation unit 112 is configured with, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc., and functions as an input unit. The display unit 113 is configured with, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The operation unit 112 and the display unit 113 may be configured integrally, such as a touch panel display.

[0039] The communication unit 114 is connected to the measurement device 3 via the network 7 and functions as a communication interface for sending and receiving various data (measurement data, etc.).

[0040] The media input / output unit 115 is configured with, for example, a DVD drive, a CD drive, a USB port, etc., and reads and writes data from and to media (non-transitory storage media) such as DVDs, CDs, and USBs.

[0041] The RAM 116 stores various data and programs and is configured from a volatile memory (DRAM, SRAM, etc.) that functions as a main memory.

[0042] The storage unit 117 is configured by a storage device such as an HDD or SSD, and stores various data required for the execution of the operating system and programs.

[0043] The program may be stored in the storage unit 117 instead of the RAM 116. The program may be recorded on a medium in an installable file format or an executable file format, and provided to the machine learning device 4 via the media input / output unit 115. The program may also be provided to the machine learning device 4 by being downloaded via the communication unit 114 over the network 7.

[0044] The control unit 111 executes a machine learning program (not shown) to function as a learning data acquisition unit 121 and a machine learning unit 122 .

[0045] The storage unit 117 includes a learning data storage unit 123 that functions as a database for storing multiple sets of learning data acquired by the learning data acquisition unit 121, and a trained model storage unit 124 that functions as a database for storing the trained learning model 6 that has been machine-learned. Note that, although the present embodiment illustrates a case in which the learning data storage unit 123 and the trained model storage unit 124 are provided in a single storage device, they may also be configured as separate storage devices.

[0046] The learning data acquisition unit 121 functions as an interface unit that acquires learning data including at least input data via the communication unit 114 and the network 7 or via the media input / output unit 115. The learning data acquisition unit 121 also acquires learning data by accepting input operations from an operator via the operation unit 112. The input data included in the learning data is acquired by attaching each sensor (motor voltage / current sensor 101, winding heat generation state measurement sensor 102, and handled liquid measurement sensor 103) serving as a measuring device 3 to each part of the canned motor pump 2 serving as a testing machine and connecting this measuring device 3 to the machine learning device 4 via the network 7. The output data included in the learning data can be automatically acquired in association with the input data if the insulation state of the winding at the time of input data acquisition can be identified using image analysis or the like. Alternatively, the output data may be acquired by operating the operation unit 112 in association with each input data.

[0047] In addition, if the learning data is acquired separately and stored in an external device, the learning data acquisition unit 121 may acquire the learning data via the media input / output unit 115 or via the communication unit 114 and the network 7.

[0048] Furthermore, when collecting learning data from the canned motor pump 2 as a test machine, it is preferable to use a pump having the same configuration as the actual machine whose winding insulation condition is to be diagnosed as the canned motor pump 2 as a test machine. However, if it is recognized that the measurement data is not dependent on the model, a similar canned motor pump may be used, or a test device that simulates a canned motor pump may be used.

[0049] The learning data storage unit 123 is a database that stores multiple sets of learning data acquired by the learning data acquisition unit 121. Therefore, the learning data stored in the learning data storage unit 123 is input data actually measured by the measurement device 3, output data input from the operation unit 112, or pre-collected existing learning data that has been loaded.

[0050] The trained model storage unit 124 is a database that stores the trained learning model 6 generated by the machine learning unit 122. The trained model 6 stored in the trained model storage unit 124 is provided to the insulation condition diagnosis device 5 via any communication network or storage medium.

[0051] The machine learning unit 122 performs machine learning using the learning data stored in the learning data storage unit 123. The machine learning unit 122 sequentially inputs multiple sets of learning data to the learning model 6, causing the learning model 6 to learn the correlation between the input data included in the learning data and the insulation information of the windings, thereby generating a trained model. There are various methods for generating a trained model, but a neural network, for example, may be used as a specific method for supervised learning by the machine learning unit 122.

[0052] Here, the learning data includes, as input data, at least data on the voltage and current values ​​supplied to the motor unit 20 (motor 24), data on the temperature of the windings 31 of the stator 23 or data on temperatures, temperature differences, or pressures correlated thereto, and data on the inlet temperature and flow rate of the handled liquid. The input data may also include other data, but in that case, it is preferable to include data that can be collected without stopping the canned motor pump and that is correlated with the insulation state of the windings.

[0053] In addition, here, raw data (unprocessed data) of the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the windings 31 of the stator 23 or data on temperatures, temperature differences, or pressures correlated therewith, and the inlet temperature and flow rate of the treated liquid are used as input data. However, the present invention is not limited to this, and the input data may be data obtained by preprocessing at least a portion of the data (the input data obtained by the learning data acquisition unit 121 and the input data stored in the learning data storage unit 123 may be data obtained by preprocessing at least a portion of the data). Data preprocessing includes, for example, removal of outliers and data standardization. It also includes processes commonly performed in machine learning, such as type conversion, missing value processing (removal, interpolation), and scaling. Preprocessing may also include at least one filtering process to ensure that the input data meets predetermined criteria.

[0054] In the case of "supervised learning," the learning data further includes, as output data (teacher data) associated with the input data, diagnostic information that indicates the insulation state of the windings at the time the input data was acquired, for example, diagnostic information that indicates which of a plurality of insulation states the insulation state is in.

[0055] Here, the diagnostic information is information evaluating the insulation state of the windings 31 (information evaluating a phenomenon for determining insulation degradation), and includes, for example: (1) information determining whether or not the treated liquid is attached to the entire windings 31 (because chemical liquids or conductive liquids are often used as the treated liquid, if the can 33 breaks and the treated liquid flows into the stator chamber 45, the treated liquid will adhere to the windings 31 and induce insulation degradation); (2) information diagnosed from the presence or absence of burnt / discolored state of the windings 31, i.e., information determined from the presence or absence of a phenomenon in which the coil bundle is burned in a circular shape or a phenomenon in which the coil bundle at the entrance (lead port) of the windings 31 is burned (because if the windings 31 are damaged, the coil bundle will heat up and burn in a circular shape, and if a large surge voltage is applied to the supplied voltage, the windings 31 will be damaged at the power supply entrance (lead wire side), causing burnt coil bundle at the winding entrance); and (3) (3) Information determined from the presence or absence of voltage imbalance or the presence or absence of one phase burnout due to voltage imbalance (phenomenon that can occur due to power supply abnormality), (4) Information determined from the degree of discoloration of the entire winding 31 and the presence or absence of signs of heating (phenomenon that can occur due to overheating).

[0056] When diagnosing the insulation state of the windings 31 for the presence or absence of an abnormality, the diagnostic information is configured as information indicating whether the insulation state is normal or abnormal. In this case, the diagnostic information may be defined as "0" when the insulation state of the windings is normal and as "1" when the insulation state of the windings is abnormal. Furthermore, abnormalities related to the insulation of the windings 31 may be classified into multiple levels of insulation deterioration (degree of abnormality). In this case, the diagnostic information may be expressed as multiple values, with "0" when the insulation state is normal, "1" when the degree of insulation deterioration is low, and "2" when the degree of insulation deterioration is high. Furthermore, abnormalities may include not only subsequent abnormalities that are discovered at the time of diagnosis, but also signs of abnormality that are within an acceptable range that is considered normal at the time of diagnosis but that may predict the occurrence of an abnormality in the future.

[0057] Here, we will explain the correlation between the input data included in the learning data (motor voltage and current data, winding temperature or temperature, temperature difference or pressure data correlated therewith, and inflow temperature and flow rate data of the handled liquid) and the diagnostic information on the insulation condition of the windings.

[0058] The voltage and current values ​​supplied to the motor section 20 are applied to the motor 24 from the motor control device 26 when the motor section 20 is driven to rotate, and vary depending on the output of the motor 24 .

[0059] (Correlation between Applied Voltage and Insulation State) The applied voltage value is necessary as basic information for performing various factor analyses. For example, if the voltage imbalance (unbalance of three-phase voltage) increases, the current imbalance increases accordingly, resulting in the phenomenon of only one phase burning. Also, by measuring the voltage waveform, it is possible to check for the presence or absence of harmonic noise or surge voltage, which are factors that can cause burning. Therefore, the applied voltage can be used to monitor or predict insulation deterioration.

[0060] (Correlation between Current Value and Insulation Condition) Motor current is supplied from the motor control device 26 to the windings 31 of the motor unit 20 when the motor is driven at a predetermined rotational speed, and therefore varies depending on the insulation condition of the windings 31. For example, excessive deterioration of the insulation of the windings 31 increases the winding resistance, increasing the motor's heat generation. In other words, when insulation resistance increases due to winding insulation deterioration, the windings overheat, resulting in thermal damage such as a layer short. Furthermore, when there is an imbalance in voltage values ​​between the phases of a three-phase power supply, a large amount of current flows concentrated in one phase, causing that phase to burn and the winding to turn black. While the relationship between voltage unbalance and phase current varies depending on the number of poles and output, it has been confirmed that even a slight voltage unbalance can significantly affect the currents in each phase. Therefore, motor current values ​​can be used to monitor or predict insulation degradation as well as coil burnout.

[0061] (Correlation between Winding Temperature or Correlated Temperature, Temperature Difference, or Pressure Data and Insulation Condition) Winding insulation deterioration is primarily due to thermal degradation. That is, heat generated by the windings causes the insulation to deteriorate, break down, and burn. Furthermore, overheating of the windings can burn and discolor the entire winding. Because heat generated by the motor 24 is primarily generated in the windings 31, directly monitoring the windings is the most preferable way to monitor the insulation condition of the windings. Furthermore, the coil end housing, which is an enclosed space, tends to trap heat generated from the coil ends and makes it difficult to dissipate it, making the coil ends the hottest part of the windings. Therefore, the winding temperature, particularly the coil end temperature (T1: the temperature of the coil end farthest from the pump, or T2: the temperature of the coil end located on the pump side), or the temperature of the windings 31 wound around the teeth within the stator core 30 (T3), can be used to monitor or predict the insulation condition of the windings.

[0062] Furthermore, because the windings 31 are fixed to the stator core 30, the windings and the stator are directly thermally affected by each other. Furthermore, heat generated in the windings 31 is transferred as radiant heat to the backup sleeves 34a and 34b, the end bells 42 and 43, the bearing holders 44 and 46, and other components, causing their surface temperatures to fluctuate according to the temperature of the windings 31. In this embodiment, where an internal circulating fluid exists (as shown in FIG. 2 , when there is a flow of internal circulating fluid within the can 33 indicated by the arrows), these surface temperatures also vary according to the state (amount, temperature) of the internal circulating fluid. Therefore, these temperatures (T4 and T5: temperatures of the backup sleeves 34a and 34b, T6 and T7: temperatures of the end bells 42 and 43, and T8 and T9: temperatures of the bearing holders 44 and 46) that correlate with the winding temperature can also be used to monitor or predict the insulation condition of the windings.

[0063] Furthermore, the pressure P1 in the stator chamber 45 (coil end accommodating portions 45a, 45b) increases as the temperature of the windings 31 increases. Furthermore, as the windings 31 begin to burn, organic matter gasifies, making the pressure in the stator chamber 45 (coil end accommodating portions 45a, 45b) more likely to increase. Therefore, if the correlation between the temperature of the windings 31 and the pressure in the stator chamber 45 is within an acceptable range, the pressure in the stator chamber 45 can also be used to monitor or predict the insulation state of the windings. By also measuring the pressure P2 in the rotor chamber 50 and monitoring the difference (ΔP) between this pressure and the pressure P1 in the stator chamber 45, it is possible to detect a rupture in the can 33 (P1 = P2), and thereby detect motor defects.

[0064] In this way, to monitor or predict the insulation state of the windings, any one of the following may be used: winding temperature; a temperature correlated with the winding temperature (e.g., the temperature of the stator core, the temperature of the stator chamber, the surface temperature of the backup sleeves 34a, 34b, or the surface temperature of the end bells 42, 43); a temperature difference (e.g., the temperature difference of the windings of each phase, the difference (Δt) between the temperature of each phase of the winding and the temperature of the working liquid, or the difference (T5-T4) between the temperatures T4, T5 of the front and rear backup sleeves 34a, 34b); or a pressure correlated with the winding temperature (e.g., the pressure in the stator chamber 45). However, multiple measures may be used as necessary. In this embodiment, a case where any one of these measures is used will be described.

[0065] (Correlation between the Inlet Temperature and Flow Rate of the Handling Liquid and the Insulation Condition) In the canned motor pump 2, the can 33 contacts the inner circumferential surface of the stator core 30, so the temperature of the stator 23 is predominantly affected by the temperature of the handling liquid. That is, heat transfer occurs between the stator 23 and the handling liquid through the can 33, which is thin and has high thermal conductivity. Therefore, the temperature of the stator 23 is significantly affected by the inlet temperature and flow rate of the handling liquid. For example, when the temperature of the handling liquid increases, the temperature of the stator core 30 also increases, and the winding 31 further self-heats, becoming hotter than the handling liquid. For this reason, in canned motor pumps 2 that often handle high-temperature handling liquid, the winding 31 is likely to be exposed to high-temperature atmospheres that are prone to insulation degradation. As such, the flow rate and temperature of the handling liquid are the base temperatures for evaluating the insulation condition of the windings and are therefore important for monitoring or predicting the insulation condition.

[0066] The data on motor voltage and current values, the temperature of the windings 31 of the motor section 20 or data on temperatures, temperature differences or pressures correlated therewith, and the inlet temperature and flow rate of the handled liquid all have an effect on the insulation state as described above, but the insulation state of the windings changes in a complex manner due to the overall influence of these data. For this reason, it is difficult to properly determine the insulation state by simply judging the above data individually, and in order to properly determine the insulation state of the windings, which changes in a complex manner, it is necessary to take the above data into consideration comprehensively.

[0067] 2, even in a canned motor pump 2 that does not have a cooling device in the motor section 20, heat generated in the windings 31 of the motor section 20 is released to the atmosphere via the motor housing 25 (outer cylinder 41, end bells 42 and 43 at both ends, and bearing holder 44), and if the temperature becomes higher than that of the handled liquid, it is also released into the handled liquid via the can 33. However, if the operating conditions of the canned motor pump 2 used in a plant do not change significantly and the heat dissipation conditions are stable, the heat balance of the canned motor pump can be easily determined, and by using data collected at predetermined times under predetermined conditions (data on the voltage and current values ​​supplied to the motor section 20, data on the temperature of the stator winding or temperatures, temperature differences, or pressures correlated thereto, and data on the inlet temperature and flow rate of the handled liquid) as input data for learning, it becomes possible to grasp the tendency of the insulation condition of the windings 31.

[0068] On the other hand, if the data on the voltage and current values ​​supplied to the motor unit 20, the temperature of the windings 31 of the stator 23 or data on a temperature, temperature difference, or pressure correlated therewith, and the data on the inlet temperature and flow rate of the treated liquid fluctuate frequently, and the amount of heat dissipated into the atmosphere also fluctuates considerably due to fluctuations in atmospheric temperature, and these data are thought to have a complex effect on the insulation state of the windings 31, pinpointing input data (input data at a predetermined timing) will result in poor correlation accuracy. In such cases, it is preferable to use data over time over a predetermined period as the input data for the learning data. That is, data on the voltage and current values ​​supplied to the motor unit over a predetermined period, data on the temperature of the stator windings or data on a temperature, temperature difference, or pressure correlated therewith, and data on the inlet temperature and flow rate of the treated liquid over a predetermined period may be used as input data. In addition, in the case of "supervised learning," the output data (teacher data) associated with the input data may also be the insulation state of the windings over time over a predetermined period.

[0069] Here, the input data and output data for a predetermined period are composed of multiple measurement values ​​measured by each sensor at multiple measurement points within the predetermined period. For example, each measurement data is measured at a predetermined sampling period within the predetermined period, associated with the measurement time, and composed as array data arranged in chronological order.

[0070] (Machine Learning Method) When the machine learning device 4 performs "supervised learning" as machine learning, it forms a learning model using, for example, a neural network model. This neural network model is a well-known model, and learns the correlation between the input data and the output data by inputting input data included in the learning data to an input layer and comparing the output data output from the output layer as the inference result with the output data (teacher data) included in the learning data.

[0071] First, as a preliminary preparation for having the machine learning device 4 perform machine learning, the learning data acquisition unit 121 acquires a desired number of sets of learning data and stores the acquired sets of learning data in the learning data storage unit 123. The number of sets of learning data to be prepared here is set appropriately taking into consideration the inference accuracy required for the learning model 6.

[0072] Various methods can be used to acquire the learning data. For example, when an abnormality occurs in the insulation state of the windings 31 or when a sign of an abnormality is recognized, the measurement device 3 acquires measured values ​​from each sensor at that time or for a predetermined period before and after that time. In addition, the diagnostic result (in this case, an abnormality has occurred or a sign of an abnormality has been recognized, so the output data is "1") is input from the operation unit 112 in association with the acquired measured values. In this way, one set of learning data (input data and output data) is prepared, and this process is repeated to prepare multiple sets of learning data.

[0073] Since insulation deterioration can occur over a long period of time or can occur all at once, if the occurrence of an abnormality or signs of an abnormality are difficult to detect, an abnormal insulation state may be intentionally created to acquire learning data. In addition, multiple sets of input data and output data for a normal state in which no abnormalities have occurred in the insulation state are also prepared as learning data.

[0074] Furthermore, to start machine learning, a pre-learning learning model 6 composed of a neural network model is prepared. The input layer of this learning model 6 corresponds to data on motor voltage and current values, data on the temperature of the stator winding 31 or a temperature, temperature difference, or pressure correlated therewith, and data on the inlet temperature and flow rate of the treated liquid, as input data contained in the learning data. The output layer corresponds to diagnostic information (insulation state), as output data contained in the learning data. After the above advance preparations, the machine learning unit 122 performs a series of processes, for example, as shown in the flowchart in FIG. 4 , to generate a trained model. In step S100, the machine learning unit 122 acquires one set of learning data from multiple sets of learning data stored in the learning data storage unit 123. The learning data may be acquired in a predetermined order or randomly.

[0075] Next, in step S110, the machine learning unit 122 performs supervised machine learning. That is, the machine learning unit 122 inputs input data included in the acquired set of learning data into the prepared learning model, causes it to output an inference result, compares this inference result with output data (teacher data) included in the learning data acquired in step S100, and performs machine learning using a known neural network technique. In this way, the machine learning unit 122 causes the learning model 6 to learn the correlation between the input data and the output data (diagnostic information about the insulation state of the winding).

[0076] Thereafter, using this learning model under learning, the above-mentioned process is performed on the remaining prepared learning data, and learning continues in the machine learning unit 122. At this time, the machine learning unit 122 determines whether or not to continue machine learning based on the error between the output data and the teacher data, the number of learning attempts, the remaining amount of unlearned learning data stored in the learning data storage unit 123, etc. (step S120).

[0077] That is, if the machine learning unit 122 determines in step S120 to continue machine learning (YES), it performs steps S100 to S110 on the currently-learned learning model 6 using unlearned learning data. On the other hand, if the machine learning unit 122 determines in step S120 not to continue machine learning (NO), it stores the generated trained learning model 6 in the trained model storage unit 124 in step S130, and ends the machine learning.

[0078] (Effects of the machine learning device and machine learning method according to the first embodiment) According to the machine learning device 4 and machine learning method according to the first embodiment described above, it is possible to provide a learning model 6 that can accurately infer (diagnose) diagnostic information about the insulation state of the windings 31 from data on the voltage and current values ​​supplied to the motor section 20, data on the temperature of the windings 31 of the stator 23 or data on temperatures, temperature differences, or pressures correlated therewith, and data on the inflow temperature and flow rate of the handled liquid.

[0079] (Insulation State Diagnostic Device) Next, an insulation state diagnostic device 5 that infers (diagnoses) the insulation state of the windings 31 using the trained learning model 6 generated using the above-described method will be described. The insulation state diagnostic device 5 is configured by a general-purpose or dedicated computer. This computer may be a stationary computer or a portable computer, and may also be a client computer, a server computer, or a cloud-based computer. Furthermore, the insulation state diagnostic device 5 may be incorporated into the motor control device 26 or into a higher-level management device of the motor control device 26. Furthermore, the computer that constitutes the insulation state diagnostic device 5 may be the same as or different from the computer that constitutes the machine learning device 4.

[0080] 5, the insulation condition diagnosis device 5 includes a control unit 131, an operation unit 132, a display unit 133, a communication unit 134, a media input / output unit 135, a RAM 136, a storage unit 137, and a bus 138. The components of the insulation condition diagnosis device 5 are connected to each other via the bus 138.

[0081] The control unit 131 is composed of an arithmetic processing device (a processor such as a CPU, MPU, or DSP), and reads and executes programs loaded into the RAM 136 to perform various arithmetic processing, thereby controlling the operation of each unit of the insulation condition diagnostic device 5. The control unit 131 may have multiple processors, and the various processes of this embodiment may be executed by multiple processors. In this case, the multiple processors may be involved in a common process, or the multiple processors may independently execute different processes in parallel.

[0082] The operation unit 132 is configured with, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc., and functions as an input unit. The display unit 133 is configured with, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc., and functions as an output unit. The operation unit 132 and the display unit 133 may be configured integrally, such as a touch panel display.

[0083] The communication unit 134 is connected to the measurement device 3 via the network 7 and functions as a communication interface for sending and receiving various data (measurement data, etc.).

[0084] The media input / output unit 135 is configured with, for example, a DVD drive, a CD drive, a USB port, etc., and reads and writes data from and to media (non-transitory storage media) such as DVDs, CDs, and USBs.

[0085] The RAM 136 stores various data and programs and is configured from a volatile memory (DRAM, SRAM, etc.) that functions as a main memory.

[0086] The storage unit 137 is configured by a storage device such as an HDD or SSD, and stores various data required for the execution of the operating system and programs.

[0087] The program may be stored in the storage unit 137 instead of the RAM 136. The program may be recorded on a medium in an installable file format or an executable file format and provided to the insulation condition diagnosis device 5 via the media input / output unit 135. The program may also be provided to the insulation condition diagnosis device 5 by being downloaded via the communication unit 134 over the network 7.

[0088] The control unit 131 executes an insulation state diagnosis program (not shown) to function as an input data acquisition unit 141, an inference unit 142, and an output processing unit 143. The storage unit 137 also includes a trained model storage unit 144 that functions as a database for storing trained learning models 6 that have been subjected to machine learning.

[0089] The input data acquisition unit 141 is an interface unit that is connected to the measuring device 3 (motor voltage / current sensor 101, winding heat generation state measuring sensor 102, handled liquid measuring sensor 103) provided in the actual canned motor pump 2 via the communication unit 134 and the network 7, and acquires input data based on the measurements taken by the measuring device 3 (data on motor voltage and current values ​​at a specified timing, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated thereto, and data on the inlet temperature and flow rate of the handled liquid, or data on motor voltage and current values ​​over a specified period, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated thereto, and data on the inlet temperature and flow rate of the handled liquid).

[0090] Here too, an example is shown in which the input data is raw data (unprocessed data) of the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the windings 31 of the stator 23 or a temperature, temperature difference, or pressure correlated thereto, and the inlet temperature and flow rate of the handled liquid. However, in accordance with the learning data, the input data may be data in which at least a portion of the voltage and current values ​​supplied to the motor unit 20 (motor 24), the temperature of the windings 31 of the stator 23 or a temperature, temperature difference, or pressure correlated thereto, and the inlet temperature and flow rate of the handled liquid has been preprocessed.

[0091] The inference unit 142 inputs the input data acquired by the input data acquisition unit 141 into the learning model 6, and performs inference processing to infer diagnostic information about the insulation state of the windings 31. For the inference processing, the learned learning model 6 that has been generated by the machine learning device 4 and has undergone supervised learning is used.

[0092] The inference unit 142 may not only have the function of performing inference processing using the learning model 6, but may also have a preprocessing function of adjusting the input data acquired by the input data acquisition unit 141 into a desired format, etc., and inputting it into the learning model 6 as preprocessing of the inference processing, and a postprocessing function of applying a predetermined logical formula or calculation formula to the value of the output data output from the learning model 6 as postprocessing of the inference processing, thereby finally diagnosing the insulation state of the winding 31.

[0093] The trained model storage unit 144 is a database that stores trained learning models 6 used in the inference process of the inference unit 142. The number of learning models 6 stored in the trained model storage unit 144 is not limited to one. For example, multiple learning models 6 may be stored for different input data, different numbers of input data, or different machine learning techniques, and may be selected and used as appropriate.

[0094] An example of a case where the input data is different is when, in a learning model that uses data on motor voltage and current values, the temperature of the stator winding, and the inlet temperature and flow rate of the handled liquid, data on temperature, temperature difference, or pressure that is correlated with the temperature of the stator winding is used instead of the temperature of the stator winding.An example of a case where the number of input data is different is when data that affects the amount of heat dissipation is added to a model of the heat balance of the motor section (this will be described later).

[0095] The output processing unit 143 performs output processing to output the inference result of the inference unit 142, i.e., diagnostic information on the insulation state of the windings 31. Various means can be used as specific output means. For example, the output processing unit 143 may notify the operator of the diagnostic information by display or email, transmit the diagnostic information to the motor control device 26 or a higher-level management device as a diagnostic history of the canned motor pump 2 for storage, or use it for drive control of the motor unit 20 and the pump unit 10.

[0096] (Insulation State Diagnostic Method) Fig. 6 is a flowchart showing an example of an insulation state diagnostic method using the insulation state diagnostic device 5. In this example, a case will be described in which the diagnostic information on the insulation state is defined as either "0" if normal or "1" if abnormal. First, in step S200, the input data acquisition unit 141 acquires input data (motor voltage and current data at a predetermined timing, stator winding temperature or a temperature, temperature difference, or pressure data correlated thereto, and inlet temperature and flow rate data of the treated liquid, or motor voltage and current data, stator winding temperature or a temperature, temperature difference, or pressure data correlated thereto, and inlet temperature and flow rate data of the treated liquid) based on measurements taken by the measurement devices 3 (motor voltage / current sensor 101, winding heat generation state measurement sensor 102, and treated liquid measurement sensor 103.

[0097] Next, in step S210, the inference unit 142 inputs the input data into the input layer of the trained learning model 6 and performs inference. Then, in step 220, the inference unit 142 determines whether or not there is an abnormality in the insulation state based on the inference result. That is, as an example of post-processing of supervised learning, the inference unit 142 compares the value of the output data (a number between 0 and 1) with a predetermined threshold value to determine whether the insulation state is normal or abnormal. For example, if the value of the output data is less than the predetermined threshold value, the insulation state is determined to be "normal" (NO), and if the value is equal to or greater than the predetermined threshold value, the insulation state is determined to be "abnormal" (YES).

[0098] Next, if the inference result of the inference unit 152 in step 220 (diagnosis result of the insulation state of the windings) is no abnormality (NO), output processing unit 154 outputs information indicating "normal" in step S230, and if the inference result is abnormal (YES), output processing unit 154 outputs information indicating "abnormal" in step 240 (step S240). Then, after outputting the diagnostic information in step S230 or S240, the diagnosis process of the insulation state of the windings ends.

[0099] (Operational Effects of the Insulation State Diagnostic Device and Insulation State Diagnostic Method According to the First Embodiment) The insulation state diagnostic device 5 and insulation state diagnostic method according to the first embodiment described above enable highly accurate diagnosis of the insulation state of the windings 31 without relying on the experience or intuition of an operator and without stopping the canned motor pump 2. That is, in order to properly determine the insulation state under complexly changing conditions, monitoring each input data individually is insufficient. Conventionally, therefore, the canned motor pump has been stopped to measure insulation resistance and inter-terminal resistance, and trend management has been performed, which relies heavily on the experience and intuition of an operator. However, the present diagnostic device uses a learning model generated using input data including data on the voltage and current values ​​supplied to the motor unit, data on the winding temperature or a temperature, temperature difference, or pressure correlated therewith, and data on the inlet temperature and flow rate of the treated liquid. This makes it possible to accurately diagnose the insulation state of the windings without relying on the experience or intuition of an expert, even when the insulation state of the windings changes due to various influences.

[0100] In the above example, when considering the heat balance that affects the insulation state of the windings of the canned motor pump, changes in the amount of heat dissipation were not actively taken into account, but if the motor section 20 is equipped with a cooling device, changes in the amount of heat dissipation cannot be ignored.

[0101] 7 shows an example of a canned motor pump equipped with a cooling device 70 in the motor section 20. In this example, a cooling jacket 71 is disposed around the motor section 20, i.e., on the outside of the outer casing 41. This cooling jacket 71 is formed by the outer casing 41, a jacket outer periphery 71a disposed outside the outer casing 41, and two end plates 71b disposed on both ends of the jacket outer periphery 71a. A cooling pipe 72 is disposed inside the cooling jacket 71, and the space other than the cooling pipe is filled with a heat transfer medium. Here, the heat transfer medium may be, for example, oil, silica, aluminum powder, or the like.

[0102] The coolant supplied to the cooling pipe 72 in the cooling jacket 71 may be a coolant from a system separate from the treated liquid of the pump section 10, or may be the treated liquid flowing through the pump section 10. When the treated liquid flowing through the pump section 10 is supplied, the treated liquid passes through the cooling pipe 72 and is then sent to the rotor chamber 50. The other components of the canned motor pump 2 are the same as those shown in Figure 2, and therefore the same parts are designated by the same reference numerals and their description will be omitted.

[0103] When such a cooling device 70 is provided, the reference conditions to which the windings 31 are exposed change depending on the temperature and flow rate of the coolant. Therefore, when such a cooling device 70 is provided, a coolant measurement sensor 104 for measuring the inlet temperature and flow rate of the coolant is provided in the pipe 73 that supplies the coolant to the cooling jacket 71. The coolant measurement sensor 104 is composed of, for example, an inlet temperature sensor 104a and a flow rate sensor 104b that are provided in the pipe that supplies the coolant to the cooling jacket 71. These inlet temperature sensor 104a and flow rate sensor 104b may be the same types as the inlet temperature sensor 103a and flow rate sensor 103b of the treated liquid measurement sensor 103.

[0104] As shown in FIGS. 8A and 8B , the inlet temperature and flow rate of the coolant measured by the coolant measurement sensor 104 may be added to the data measured by the measuring device 3 used in the machine learning device 4 and the insulation condition diagnosis device 5. That is, the inlet temperature and flow rate of the coolant may be included in the input data included in the learning data in the learning phase, and the inlet temperature and flow rate of the coolant may be included in the input data for diagnosis in the inference phase. Even in this case, the inlet temperature and flow rate of the coolant used as input data are not limited to raw data (unprocessed data) obtained by the inlet temperature sensor 103 a and the flow rate sensor 103 b, and preprocessed data may be used. Furthermore, the coolant temperature and flow rate data may be data obtained at a predetermined timing, or data obtained over a predetermined period may be used to improve correlation accuracy and inference accuracy. Therefore, when the motor unit 20 includes a cooling device 70, the insulation condition of the windings 31 can be more accurately inferred (diagnosed) by further taking the inlet temperature and flow rate of the coolant into account in the learning phase and inference phase.

[0105] Second Embodiment In the first embodiment described above, a case where "supervised learning" is adopted as machine learning has been described, but in this embodiment, a case where "unsupervised learning" is adopted will be described. Note that the basic configurations and operations of the measuring device 3, the machine learning device 4, and the insulation condition diagnosis device 5 that constitute the insulation condition diagnosis system 1 for the canned motor pump 2 and the windings 31 according to the second embodiment are the same as those of the first embodiment, so a description thereof will be omitted and the following description will focus on the differences from the first embodiment.

[0106] As in the first embodiment, the machine learning device 4 includes a learning data acquisition unit 121, a learning data storage unit 123, a machine learning unit 122, and a trained model storage unit 124, as shown in FIG. 3. When "unsupervised learning" is adopted as the machine learning, the learning data uses only input data when the diagnostic information indicates that the insulation state of the windings at a predetermined timing or that the insulation state of the windings over a predetermined period of time is in a predetermined state. In other words, the learning data does not include output data (teaching data).

[0107] When diagnosing whether there is an abnormality in the insulation state of the windings 31, the diagnostic information is configured as information indicating whether the insulation state is normal or abnormal. Here, the diagnostic information is configured only with input data when the insulation state of the windings is normal (data on the voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings or temperatures, temperature differences, or pressures correlated thereto, data on the inlet temperature and flow rate of the liquid being handled, and, if a cooling device is provided, data on the inlet temperature and flow rate of the cooling liquid).

[0108] When acquiring the learning data, the learning data acquisition unit 121 acquires as input data the voltage and current values ​​supplied to the motor at a predetermined timing or for a predetermined period, the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, the inlet temperature and flow rate of the treated liquid, and, if a cooling device is provided, the inlet temperature and flow rate of the coolant, all measured by the measuring device 3 provided in the canned motor pump 2 serving as a tester. At the same time, the operator diagnoses the insulation condition at the timing or for the predetermined period when the input data was measured and inputs, via the operation unit 112, that the diagnosis result is normal, thereby forming a set of learning data and storing it in the learning data storage unit 123. By repeating this process, multiple sets of learning data are stored in the learning data storage unit 123.

[0109] Even in this case, the learning data acquisition unit 121 may collect the input data included in the learning data by attaching each sensor (motor voltage / current sensor 101, winding heat generation state measurement sensor 102, and handled liquid measurement sensor 103) serving as a measuring device 3 to each part of the canned motor pump 2 serving as a testing machine and connecting this measuring device 3 to the machine learning device 4 via the network 7. Alternatively, if the collected learning data (motor voltage / current value data as input data, winding temperature or temperature, temperature difference, or pressure data correlated therewith, and handled liquid inlet temperature and flow rate data) already exists, the learning data acquisition unit 121 may load the existing learning data into the learning data storage unit 123 via the media input / output unit 115, the communication unit 114, and the network 7. Also, in this embodiment, the input data is not limited to raw data (unprocessed data), and at least a portion of the data may be subjected to preprocessing similar to that described above.

[0110] The machine learning unit 122 inputs learning data into the learning model, causing the learning model to learn the correlation between the input data included in the learning data and the diagnostic information indicating that the insulation state of the winding is normal, and generates a trained model. As a method of unsupervised learning by the machine learning unit, for example, an autoencoder model may be adopted.

[0111] This autoencoder model is a well-known model that inputs input data contained in training data into an input layer and compares this input data with output data output from the output layer as inference results to learn patterns and trends in the input data. This series of processes is repeatedly performed on the training model, and when a predetermined training termination condition is met, the machine learning is terminated and a trained autoencoder model is generated.

[0112] 9 is a flowchart illustrating an example of a machine learning method using the machine learning device according to the second embodiment. In this example, as a preliminary preparation for having the machine learning device 4 perform machine learning, the learning data acquisition unit 121 acquires a desired number of sets of learning data and stores the acquired sets of learning data in the learning data storage unit 123. The number of sets of learning data to be prepared here is set appropriately taking into account the inference accuracy required for the learning model 6.

[0113] Various methods can be used to acquire the learning data, but for example, input data constituting the learning data is prepared by using the measuring device 3 to acquire various measured values ​​at predetermined timings or for a predetermined period when the insulation state of the windings 31 in the canned motor pump 2 serving as a test machine is normal. Then, by repeating this process, multiple sets of learning data are prepared.

[0114] Furthermore, before starting machine learning, a pre-learning learning model 6 composed of an autoencoder model is prepared. The input layer of this learning model 6 is associated with data on the voltage and current values ​​supplied to the motor unit as input data contained in the learning data, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, and data on the inlet temperature and flow rate of the handled liquid, and if a cooling device is used, data on the inlet temperature and flow rate of the cooling liquid is further associated with the input data.

[0115] After the advance preparation is completed, in step S300, the machine learning unit 122 first acquires one set of learning data from the multiple sets of learning data stored in the learning data storage unit 123. The learning data may be acquired in a predetermined order or randomly.

[0116] Next, in step S310, the machine learning unit 122 performs unsupervised machine learning. That is, the machine learning unit 122 inputs input data included in the acquired set of training data into a prepared training model, causes the model to output an inference result, compares the input data included in the training data with output data output from the output layer as an inference result, and performs machine learning using a known autoencoder technique.

[0117] Thereafter, the machine learning unit 122 continues learning by performing the above-described process on the remaining prepared learning data using the learning model under training. At this time, the machine learning unit 122 determines whether or not to continue machine learning based on the number of times of training and the remaining amount of untrained learning data stored in the learning data storage unit 123 (step S320). That is, if the machine learning unit 122 determines in step S320 to continue machine learning (YES), it performs steps S300 to S310 on the learning model 6 under training using untrained learning data. On the other hand, if the machine learning unit 122 determines in step S320 not to continue machine learning (NO), then in step S330 the machine learning unit 122 stores the generated trained learning model 6 in the trained model storage unit 124 and ends machine learning.

[0118] (Action and effect of the machine learning device and machine learning method according to the second embodiment) By using the machine learning device and machine learning method according to the second embodiment described above, it is possible to provide a learning model 6 that can infer (estimate) diagnostic information on the insulation state of the windings with high accuracy from data on motor voltage values ​​and current values ​​at a predetermined timing, data on the temperature of the stator windings or a temperature, temperature difference, or pressure correlated thereto, and data on the inflow temperature and flow rate of the handled liquid, or data on motor voltage values ​​and current values ​​over a predetermined period, data on the temperature of the stator windings or a temperature, temperature difference, or pressure correlated thereto, and data on the inflow temperature and flow rate of the handled liquid.

[0119] The insulation state diagnosis device 5 in this embodiment has the same configuration as that of the first embodiment in Fig. 5. That is, the insulation state diagnosis device 5 has an input data acquisition unit 141, an inference unit 142, an output processing unit 143, and a trained model storage unit 144. As in the first embodiment, the inference unit 152 inputs input data acquired by the input data acquisition unit 141 to the trained learning model 6 and performs inference processing to infer diagnostic information about the insulation state of the windings, but the inference processing uses the trained model that has been trained in the machine learning device 4 through unsupervised learning.

[0120] 10 is a flowchart showing an example of a method for diagnosing the insulation state of the winding 31 by the insulation state diagnosis device 5 according to the second embodiment. In this example, a case will be described in which the diagnostic information on the insulation state is defined as either "0" if normal or "1" if abnormal.

[0121] First, in step S400, the input data acquisition unit 141 acquires input data (motor voltage and current data at a predetermined timing or for a predetermined period, stator winding temperature or a temperature, temperature difference, or pressure data correlated therewith, handled liquid inlet temperature and flow rate data, and, if a cooling device 70 is provided, coolant inlet temperature and flow rate data) based on measurements taken by the measurement devices 3 (motor voltage / current sensor 101, winding heat generation state measurement sensor 102, handled liquid measurement sensor 103, and, if a cooling device 70 is provided, coolant measurement sensor 104). Here, too, the input data may be partially preprocessed.

[0122] Next, in step S410, the inference unit 142 inputs input data to the input layer of the learning model 6 to perform inference, and obtains output data output from the output layer of the learning model 6.

[0123] Then, in step S420, the inference unit 142 determines whether or not there is an abnormality in the insulation state based on the inference result. That is, as an example of post-processing of unsupervised learning, the inference unit 142 calculates the difference between the feature amount based on the input data and the feature amount based on the output data, and if the difference is less than a predetermined threshold, determines that the insulation state is "normal" (NO), and if the difference is equal to or greater than the predetermined threshold, determines that the insulation state is "abnormal" (YES).

[0124] Next, in step S430, if the output processing unit 154 determines that the diagnostic information on the insulation state of the windings, which is the inference result of the inference unit 152, is normal (NO), it outputs information indicating "normal" (step S430), and if the diagnostic information on the insulation state of the windings, which is the inference result of the inference unit 152, it outputs information indicating "abnormal" (YES) (step S440). After outputting the diagnostic information of step S430 or S440, the process of diagnosing the insulation state of the windings ends.

[0125] (Actions and Effects of the Insulation State Diagnostic Device and Insulation State Diagnostic Method According to the Second Embodiment) The insulation state diagnostic device 5 and insulation state diagnostic method according to the second embodiment described above also make it possible to diagnose the insulation state of the windings with high accuracy without relying on the experience or intuition of the operator and without stopping the canned motor pump 2.

[0126] (Other Embodiments) Note that in the above-described embodiments, the cases where a neural network (first embodiment) and an autoencoder (second embodiment) are employed as specific machine learning techniques by the machine learning unit 122 have been described, but the machine learning unit 122 may employ any other machine learning technique. For example, tree-type techniques such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network-type techniques (including deep learning) such as recurrent neural networks and convolutional neural networks, clustering-type techniques such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analysis such as principal component analysis, factor analysis, and logistic regression, support vector machines, etc. may be employed.

[0127] In addition, the above-mentioned machine learning device 4 and insulation condition diagnosis device 5 can also be provided in the form of a program (machine learning program or insulation condition diagnosis program) for executing each step of the above-mentioned machine learning method or insulation condition diagnosis method.

[0128] In the above example, measurement data from each sensor at a predetermined timing or over a predetermined period of time was used as input data for the learning data as data on the temperature of the windings 31 of the stator 23 or data on temperatures, temperature differences, or pressures correlated therewith. However, temperature difference data correlated with the temperature of the windings 31 may also be used as appropriate as input data. For example, even if the voltage and current supplied to the motor unit 20 are constant, imbalance in the winding resistance values ​​of each phase due to insulation deterioration can cause variations in the temperatures of the windings of each phase. Therefore, data on the temperature difference between the windings of each phase may also be used. Furthermore, as temperature difference data correlated with the temperature of the windings 31, the difference (Δt) between the temperature of each phase of the winding and the temperature of the operating liquid, or the difference (T5 - T4) between the temperatures T4 and T5 of the front and rear backup sleeves 34a, 34b can also be monitored to detect abnormal heat generation in the motor (windings 31). Therefore, data on these temperature differences may also be used as input data.

[0129] (Embodiments of the Present Invention) Next, embodiments of the present invention that can be understood from the above-described embodiments will be described below, using the terms and symbols described in the respective embodiments.

[0130] A first embodiment of the present invention includes a motor unit (e.g., motor unit 20) and a pump unit (e.g., pump unit 10) driven by the motor unit, wherein the motor unit includes a rotor (e.g., rotor 22) rotatable around a shaft (e.g., shaft 21), a stator (e.g., stator 23) facing the rotor with a gap therebetween, and a cylindrical can (e.g., can 33) forming a rotor chamber (e.g., rotor chamber 50) for accommodating the rotor on the inside and forming a stator chamber (e.g., stator chamber 45) for accommodating the stator together with a motor housing (e.g., motor housing 25) on the outside, and a machine learning device (e.g., machine learning device 4) that generates a learning model (e.g., learning model 6) to be used in an insulation state diagnosis device (e.g., insulation state diagnosis device 5) for diagnosing an insulation state of a winding (e.g., winding 31) of the stator used in a canned motor pump (e.g., canned motor pump 2) that circulates a portion of the handled liquid in the pump unit to the rotor chamber, a learning data acquisition unit (e.g., learning data acquisition unit 121) that acquires multiple sets of learning data including, as input data, unprocessed data or at least partially preprocessed data, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings (e.g., windings 31, coil ends 31a, 31b) or temperatures correlated therewith (e.g., temperatures of the stator core 30, backup sleeves 34a, 34b, end bells 42, 43, or bearing holders 44, 46), data on temperature differences (e.g., temperature differences of the windings of each phase, differences (Δt) between the temperatures of the windings of each phase and the temperature of the treated liquid, and differences (T5-T4) between the temperatures T4, T5 of the front and rear backup sleeves 34a, 34b) or pressures (e.g., pressures in the stator chamber 45 or coil end accommodating portions 45a, 45b), and data on the inflow temperature and flow rate of the treated liquid; The machine learning device includes: a machine learning unit (e.g., machine learning unit 122) that inputs the learning data into the learning model, thereby causing the learning model to learn the correlation between the input data and diagnostic information on the insulation state of the winding; and a trained model storage unit (e.g., trained model storage unit 124) that stores the learning model trained by the machine learning unit.

[0131] Here, the input data used as learning data is the following unprocessed data (1) to (3) or at least partially preprocessed data: (1) data on the voltage and current values ​​supplied to the motor unit, (2) data on the temperature of the stator windings or data on temperatures, temperature differences, or pressures correlated therewith, and (3) data on the inlet temperature and flow rate of the handled liquid. While (1) to (3) are essential input data, (2) can be any of the following: (i) data on the temperature of the stator windings, (ii) data on temperatures correlated with the temperature of the stator windings, (iii) data on temperature differences correlated with the temperature of the stator windings, or (iv) data on pressures correlated with the temperature of the stator windings.

[0132] This configuration provides a learning model that can accurately infer (estimate) diagnostic information for the motor winding insulation from unprocessed or at least partially preprocessed data, such as data on the voltage and current values ​​supplied to the motor of a canned motor pump, data on the temperature of the stator winding of the motor or data on temperatures, temperature differences, or pressures correlated therewith, and data on the inlet temperature and flow rate of the pump-handled liquid. It also provides a learning model that can accurately diagnose the insulation condition of the windings without stopping the canned motor pump.

[0133] A second embodiment of the present invention is the machine learning device of the first embodiment, further comprising a cooling jacket (e.g., cooling jacket 71) on at least a portion of a motor housing of the motor unit, and the input data further includes data on the inlet temperature and flow rate of coolant supplied into the cooling jacket, the data being unprocessed data or data at least partially preprocessed. With this configuration, in addition to the input data of the first embodiment, data on the inlet temperature and flow rate of coolant is further added as input data included in the learning data, the unprocessed data or data at least partially preprocessed data, making it possible to provide a learning model that can more accurately infer (estimate) diagnostic information about the winding insulation of the motor unit.

[0134] A third aspect of the present invention is the machine learning device of the first aspect, wherein the data of voltage and current values ​​supplied to the motor unit is data supplied to the motor unit over a predetermined period, the data of the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith is data over the predetermined period, and the data of the inlet temperature and flow rate of the treated liquid is data over the predetermined period. With this configuration, data over a predetermined period is used as the input data used in the first aspect, thereby improving the accuracy of the correlation of the learning data.

[0135] A fourth aspect of the present invention is the machine learning device of the second aspect, wherein the data on the voltage and current values ​​supplied to the motor unit is data supplied to the motor unit over a predetermined period of time, the data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith is data over the predetermined period of time, the data on the inlet temperature and flow rate of the treated liquid is data over the predetermined period of time, and the data on the inlet temperature and flow rate of the coolant supplied into the cooling jacket is data over the predetermined period of time. With this configuration, data over a predetermined period of time is used as input data used in the second aspect, thereby improving the accuracy of the correlation of the learning data.

[0136] A fifth aspect of the present invention is a machine learning device according to any one of the first to fourth aspects, wherein the learning data further includes, as output data, the diagnostic information associated with the input data and indicating that the insulation state of the winding is one of a plurality of states, and the machine learning unit causes the learning model to learn the correlation between the input data and the output data by supervised learning. With this configuration, it is possible to generate a learning model that has learned the correlation between the input data and the diagnostic information about the insulation state of the winding by supervised learning.

[0137] A sixth aspect of the present invention is a machine learning device in any one of the first to fourth aspects, wherein the learning data includes only input data when the diagnostic information indicates that the insulation state of the winding is in a predetermined state, and the machine learning unit causes the learning model to learn, by unsupervised learning, a correlation between the input data and the diagnostic information indicating that the insulation state of the winding is in the predetermined state. With this configuration, it is possible to generate a learning model that has learned, by unsupervised learning, the correlation between the input data and the diagnostic information about the insulation state of the winding.

[0138] A seventh embodiment of the present invention is an insulation state diagnosis device (e.g., insulation state diagnosis device 5) for diagnosing an insulation state of the stator winding used in the canned motor pump using a learning model generated by the machine learning device in any of the first to sixth embodiments, the insulation state diagnosis device comprising: an input data acquisition unit (e.g., input data acquisition unit 141) that acquires input data including unprocessed data or data that has been at least partially preprocessed, such as data on voltage values ​​and current values ​​supplied to the motor unit, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, data on the inflow temperature and flow rate of the handled liquid, and, as necessary, data on the inflow temperature and flow rate of the coolant; and an inference unit (e.g., inference unit 142) that inputs the input data acquired by the input data acquisition unit into the learning model and infers diagnostic information on the insulation state of the stator winding. According to this configuration, diagnostic information for the insulation condition of the stator windings can be inferred using input data including unprocessed data or data that has been at least partially preprocessed, such as data on the voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings or temperatures, temperature differences, or pressures correlated thereto, data on the inlet temperature and flow rate of the handled liquid, and, as necessary, data on the inlet temperature and flow rate of the cooling liquid.This makes it possible to diagnose the insulation condition of the windings with high accuracy without shutting down the plant.

[0139] An eighth embodiment of the present invention is a machine learning method for learning a learning model used in an insulation condition diagnosis device for diagnosing an insulation condition of a winding of a stator used in a canned motor pump that circulates a portion of a handled liquid in the pump section to the rotor chamber, the machine learning method comprising: a motor section; and a pump section driven by the motor section; wherein the motor section comprises a rotor rotatable around a shaft, a stator facing the rotor across a gap, and a cylindrical can that forms a rotor chamber for accommodating the rotor on the inside and, together with a motor housing, forms a stator chamber for accommodating the stator on the outside; and a learning data acquisition step (e.g., steps S100 and S300) for acquiring a plurality of sets of learning data comprising at least this input data, the learning data being data on voltage values ​​and current values ​​supplied to the motor section, data on the temperature of the winding of the stator or data on a temperature, temperature difference or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid, and the data being unprocessed data or data that has been at least partially preprocessed; The machine learning method includes a machine learning process (e.g., processes S110 and S310) for inputting the learning data into the learning model to cause the learning model to learn a correlation between the input data and diagnostic information about the insulation condition of the windings; and a learned model storage process (e.g., processes S130 and S330) for storing the learning model learned by the machine learning process in a learned model storage unit. This configuration provides a learning model that can accurately infer (estimate) diagnostic information about the winding insulation of the motor unit from unprocessed data or at least partially preprocessed data, including data on voltage and current values ​​supplied to the motor unit of a canned motor pump, data on the temperature of the stator winding of the motor unit or data on temperatures, temperature differences, or pressures correlated therewith, and data on the inlet temperature and flow rate of the pump-handled liquid. The machine learning method also provides a learning model that can accurately diagnose the insulation condition of the windings without stopping the canned motor pump.

[0140] A ninth aspect of the present invention is the machine learning method of the eighth aspect, wherein the motor unit further includes a cooling jacket on at least a part of an outer periphery of the motor housing, and the input data further includes data on the inlet temperature and flow rate of coolant supplied into the cooling jacket, the data being unprocessed data or data having at least a portion thereof subjected to preprocessing. With this configuration, in addition to the input data of the eighth aspect, data on the inlet temperature and flow rate of coolant is further added as input data included in the learning data, the unprocessed data or data having at least a portion thereof subjected to preprocessing, so that a learning model capable of more accurately inferring (estimating) diagnostic information on the winding insulation of the motor unit can be provided.

[0141] A tenth aspect of the present invention is a machine learning program for causing a computer to execute each step of the machine learning method according to the eighth or ninth aspect. This configuration provides a learning model capable of highly accurately inferring (estimating) diagnostic information for the winding insulation of the motor unit from unprocessed data or at least partially preprocessed data, including data on voltage and current values ​​supplied to the motor unit of a canned motor pump, data on the temperature of the stator winding of the motor unit or data on a temperature, temperature difference, or pressure correlated therewith, data on the inlet temperature and flow rate of the liquid handled by the pump unit, and, as necessary, data on the inlet temperature and flow rate of the coolant. Furthermore, a learning model capable of accurately diagnosing the insulation condition of the windings without stopping the canned motor pump can be provided.

[0142] An eleventh aspect of the present invention is an insulation condition diagnosis method for diagnosing an insulation condition of a winding of the stator used in the canned motor pump using a learning model generated by the machine learning device according to any one of the first to sixth aspects, the insulation condition diagnosis method comprising: an input data acquisition step (e.g., steps S200 and S400) for acquiring input data including unprocessed data or data having been at least partially preprocessed, such as data on voltage values ​​and current values ​​supplied to the motor unit, data on the temperature of the stator winding or a temperature, temperature difference, or pressure correlated therewith, data on the inflow temperature and flow rate of the handled liquid, and, as necessary, data on the inflow temperature and flow rate of the cooling liquid; and an inference step (steps S210 and S410) for inputting the input data acquired by the input data acquisition step into the learning model and inferring diagnostic information on the insulation condition of the winding of the stator. This configuration makes it possible to infer (estimate) diagnostic information for the motor winding insulation with high accuracy using input data including data on the voltage and current values ​​supplied to the motor, data on the stator winding temperature or a temperature, temperature difference, or pressure correlated thereto, data on the inlet temperature and flow rate of the treated liquid, and, as necessary, data on the inlet temperature and flow rate of the coolant, and the input data may be unprocessed or at least partially preprocessed. Furthermore, the insulation condition of the windings can be diagnosed with high accuracy without stopping the canned motor pump.

[0143] A twelfth embodiment of the present invention is an insulation condition diagnosis program for causing a computer to execute each step of the insulation condition diagnosis method according to the eleventh embodiment. This configuration enables highly accurate inference (estimation) of diagnostic information for the motor winding insulation from input data including unprocessed data or at least partially preprocessed data, such as data on voltage and current values ​​supplied to the motor of a canned motor pump, data on the temperature of the stator winding of the motor or data on a temperature, temperature difference, or pressure correlated thereto, data on the inlet temperature and flow rate of the liquid handled by the pump, and, as necessary, data on the inlet temperature and flow rate of the coolant. Furthermore, the insulation condition of the winding can be accurately diagnosed without stopping the canned motor pump.

[0144] REFERENCE SIGNS LIST 1 Insulation state diagnosis system 2 Canned motor pump 3 Measuring device 4 Machine learning device 5 Insulation state diagnosis device 6 Learning model 10 Pump section 20 Motor section 21 Shaft 22 Rotor 23 Stator 25 Motor housing 31 Winding 33 Can 45 Stator chamber 50 Rotor chamber 71 Cooling jacket 101 Motor voltage / current sensor 102 Winding heat generation state measurement sensor 103 Handling liquid measurement sensor 104 Cooling liquid measurement sensor 121 Learning data acquisition unit 122 Machine learning unit 123 Learning data storage unit 124 Trained model storage unit 141 Input data acquisition unit 142 Inference unit 143 Output processing unit 144 Trained model storage unit

Claims

1. A machine learning device for generating a learning model for use in an insulation condition diagnosis device for diagnosing an insulation condition of a winding of a stator used in a canned motor pump that circulates a portion of a handled liquid in the pump section to the rotor chamber, the machine learning device comprising: a learning data acquisition unit that acquires a plurality of sets of learning data including at least the input data, the input data being data on voltage and current values ​​supplied to the motor section, data on the temperature of the winding of the stator or a temperature, temperature difference or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; a machine learning unit that inputs the learning data to the learning model to cause the learning model to learn a correlation between the input data and diagnosis information on the insulation condition of the winding; a trained model storage unit that stores the learning model trained by the machine learning unit.

2. The machine learning device according to claim 1, wherein the motor section further comprises a cooling jacket on at least a portion of the motor housing, and the input data further includes data on an inlet temperature and a flow rate of a coolant supplied into the cooling jacket.

3. The machine learning device of claim 1, wherein the data on voltage and current values ​​supplied to the motor unit is data supplied to the motor unit during a predetermined period of time, the data on the temperature of the stator winding or a temperature, temperature difference or pressure correlated therewith is data during the predetermined period of time, and the data on the inlet temperature and flow rate of the handled liquid is data during the predetermined period of time.

4. The machine learning device of claim 2, wherein data on voltage and current values ​​supplied to the motor unit is data supplied to the motor unit during a predetermined period of time, data on the temperature of the stator winding or a temperature, temperature difference or pressure correlated therewith is data during the predetermined period of time, data on the inlet temperature and flow rate of the handled liquid is data during the predetermined period of time, and data on the inlet temperature and flow rate of the cooling liquid supplied into the cooling jacket is data during the predetermined period of time.

5. A machine learning device as described in any one of claims 1 to 4, wherein the learning data is associated with the input data and further includes, as output data, the diagnostic information indicating that the insulation state of the winding is one of a plurality of states, and the machine learning unit causes the learning model to learn the correlation between the input data and the output data through supervised learning.

6. A machine learning device as described in any one of claims 1 to 4, wherein the learning data includes only input data when the diagnostic information indicates that the insulation state of the winding is in a predetermined state, and the machine learning unit causes the learning model to learn the correlation between the input data and the diagnostic information indicating that the insulation state of the winding is in the predetermined state by unsupervised learning.

7. An insulation condition diagnosis device for diagnosing the insulation condition of the stator winding used in the canned motor pump using a learning model generated by the machine learning device according to claim 1 or 3, comprising: an input data acquisition unit that acquires input data including unprocessed data or data that has been at least partially preprocessed, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator winding or a temperature, temperature difference or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; and an inference unit that inputs the input data acquired by the input data acquisition unit into the learning model and infers diagnostic information for the insulation condition of the stator winding.

8. An insulation condition diagnosis device for diagnosing an insulation condition of a stator winding used in the canned motor pump using a learning model generated by the machine learning device according to claim 2 or 4, comprising: an input data acquisition unit that acquires input data including unprocessed data or data that has been at least partially preprocessed, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator winding or data on a temperature, temperature difference or pressure correlated therewith, data on the inlet temperature and flow rate of the handled liquid, and data on the inlet temperature and flow rate of a cooling liquid; and an inference unit that inputs the input data acquired by the input data acquisition unit into the learning model and infers diagnostic information for the insulation condition of the stator winding.

9. A machine learning method for learning a learning model used in an insulation condition diagnosis device for diagnosing an insulation condition of a winding of a stator used in a canned motor pump that circulates a part of a handled liquid in the pump section to the rotor chamber, the machine learning method comprising: a learning data acquisition step of acquiring a plurality of sets of learning data including at least the input data, the input data being data on voltage and current values ​​supplied to the motor section, data on the temperature of the winding of the stator or a temperature, temperature difference or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; and a machine learning step of inputting the learning data into the learning model to cause the learning model to learn a correlation between the input data and diagnosis information on the insulation condition of the winding. a trained model storage step of storing the learning model trained by the machine learning step in a trained model storage unit.

10. The machine learning method according to claim 9, wherein the motor section further comprises a cooling jacket on at least a portion of an outer periphery of the motor housing, and the input data further includes data on an inlet temperature and a flow rate of a coolant supplied into the cooling jacket.

11. A machine learning program for causing a computer to execute each step of the machine learning method according to claim 9 or 10.

12. An insulation condition diagnosis method for diagnosing the insulation condition of the stator windings used in the canned motor pump using a learning model generated by the machine learning device described in claim 1 or 3, comprising: an input data acquisition step of acquiring input data including unprocessed data or data that has been at least partially preprocessed, the input data being data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator windings or a temperature, temperature difference or pressure correlated therewith, and data on the inflow temperature and flow rate of the handled liquid; and an inference step of inputting the input data acquired by the input data acquisition step into the learning model and inferring diagnostic information for the insulation condition of the stator windings.

13. An insulation condition diagnosis method for diagnosing the insulation condition of the stator winding used in the canned motor pump using a learning model generated by the machine learning device described in claim 2 or 4, comprising: an input data acquisition step of acquiring input data including unprocessed data or data that has been at least partially preprocessed, such as data on voltage and current values ​​supplied to the motor unit, data on the temperature of the stator winding or data on a temperature, temperature difference or pressure correlated therewith, data on the inlet temperature and flow rate of the handled liquid, and data on the inlet temperature and flow rate of the cooling liquid; and an inference step of inputting the input data acquired by the input data acquisition step into the learning model and inferring diagnostic information for the insulation condition of the stator winding.

14. An insulation condition diagnosis program for causing a computer to execute each step of the insulation condition diagnosis method according to claim 12.

15. An insulation condition diagnosis program for causing a computer to execute each step of the insulation condition diagnosis method according to claim 13.

Images