Vacuum pump, purge gas monitoring device, and purge gas monitoring system

Abstract

[Problem] A vacuum pump capable of effectively determining whether an abnormality depending on a purge gas flow rate has occurred is provided [Solution] A turbo molecular pump 100 includes a plurality of temperature sensors 222A to 222C that detect temperatures in a first heat exhaust path (arrow E) to a third heat exhaust path (arrow G) through which a purge gas in the turbo molecular pump 100 flows, a determination model creation unit 232 that creates a determination model (such as an MT method model, a logistic regression model, or a deep learning model) for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature sensors 222A to 222C, and a determination unit 234 that determines the purge gas flow rate state using the determination model, and the temperature sensors 222A to 222C are disposed in different heat exhaust paths.

Description

VACUUM PUMP, PURGE GAS MONITORING DEVICE, AND PURGE GAS MONITORING SYSTEM[Technical Field]

[0001] The present invention relates to, for example, a vacuum pump using a purge gas, a purge gas monitoring device, and a purge gas monitoring system.[Background Art]

[0002] For example, some vacuum pumps such as turbo molecular pumps are supplied with a purge gas for the purpose of protection from the gas to be exhausted. For this purpose, the purge gas is often supplied to a portion in a vacuum space other than a path of the gas to be exhausted. Examples of the gas to be exhausted (exhaust gas) include, for example, a process gas in the manufacture of a semiconductor device. Examples of the "portion other than the path of the gas to be exhausted" to which the purge gas is supplied include a space between a rotor blade and a stator column, a space between a rotor and a stator column, and the like.

[0003] The path of the exhaust gas may be referred to as a "gas exhaust path" below. A portion other than the path of the gas to be exhausted may also be referred to as a "portion other than the gas exhaust path" below. A path through which a purge gas flows may be referred to as a "purge gas path" below.

[0004] As a purge gas, generally, an inert gas such as nitrogen gas is used. The purge gas seals a space between a "gas exhaust path" and a "portion other than the gas exhaust path" to prevent corrosion due to infiltration of the exhaust gas into the "portion other than the gas exhaust path". In addition, the purge gas also performs functions of preventing deposition of product materials (product deposition), promoting heat transfer from rotor blades (hot parts) to a stator column (cold part), and the like.

[0005] PTL 1 (paragraph 0049, and the like), which will be described below, discloses supplying an inert gas to a turbo molecular pump (1). Further, PTL 1 discloses that the amount of inert gas supplied is adjusted in accordance with a target temperature zone, heating conditions, and the like.

[0006] [PTL 1] Japanese Patent Application Publication No. 2024-038591[Summary of Invention][Technical Problem]

[0007] Incidentally, the management of a flow rate of a purge gas depends on a user (customer) using a vacuum pump and is performed by the customer. In general, a desired flow rate of a purge gas is specified in instruction manuals and specifications of a vacuum pump. However, a vacuum pump, such as a turbo molecular pump, is not provided with means for measuring a flow rate of a purge gas. On the other hand, also in peripheral devices held by customers, a flow meter for a purge gas, or the like being installed is usually avoided and is not provided due to a cost increase and an installation space problem. Peripheral devices that are held by customers can include devices and apparatuses that are installed and connected around a vacuum pump.

[0008] In general, a flow rate of a purge gas is adjusted to a predetermined flow rate by the pressure of the purge gas and the design of an orifice through which the purge gas passes (the minimum area in an introduction tube of the purge gas). However, there is a possibility that the flow rate of the purge gas will be lower than a target flow rate (appropriate flow rate) due to slight pressure fluctuations or clogging of the orifice. When the flow rate is lower than the appropriate flow rate, the purge gas may be insufficient and the vacuum pump may not exert its intended function. The insufficient purge gas may then cause a failure or damage of the vacuum pump.

[0009] In particular, in a vacuum pump with a system referred to as a high temperature TMS (Temperature Management System), the temperature of a flow path fora process gas (process gas flow path) is adjusted to a high temperature (for example, approximately 150°C), and the temperature of the side of a flow path for a purge gas (purge gas flow path) is adjusted to a low temperature (for example, approximately 50 to 100°C). For this reason, when the flow rate of the purge gas decreases and sealing performance by the purge gas decreases, product materials (deposits) are fixed in a minute space between a rotating body and a stator column,which may result in an operational failure such as rotating body locking. In vacuum pumps, techniques for detecting an abnormality in the purge gas flow rate (abnormality detection) to enable predictive maintenance and interlock are required.

[0010] PTL 1 discloses that the amount of an inert gas to be supplied is adjusted in accordance with a target temperature zone, heating conditions, and the like, and does not disclose how to effectively determine the state of a flow rate of an inert gas (flow rate state).

[0011] An object of the present invention is to provide a vacuum pump capable of effectively determining whether an abnormality depending on a purge gas flow rate has occurred, a purge gas monitoring device, and a purge gas monitoring device system.[Solution to Problem]

[0012] To achieve the above object, a vacuum pump according to the present invention includesa plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows,a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units, anda determination unit that determines the purge gas flow rate state using the determination model,in which the temperature detection units are disposed in different heat exhaust paths. In addition, the purge gas monitoring device and the purge gas monitoring system according to the present invention have the same configuration as that of the vacuum pump.[Advantageous Effects of Invention]

[0013] According to the present invention, it is possible to provide a vacuum pump capable of effectively determining whether an abnormality depending on a purge gas flow rate has occurred, a purge gas monitoring device, and a purge gas monitoring device system.[Brief Description of Drawings]

[0014] Fig. 1 is a diagram schematically showing the configuration of a vacuum pump according to an embodiment of the present invention.Fig. 2 is a circuit diagram of an amplifier circuit.Fig. 3 is a time chart showing control when a current command value is larger than a detection value.Fig. 4 is a time chart showing control when a current command value is smaller than a detection value.Fig. 5 is an enlarged view showing a portion surrounded by a rectangular frame C in Fig. 1 . Fig. 6 is a diagram schematically showing an exhaust heat path in the portion surrounded by the rectangular frame C in Fig. 1 .Fig. 7 is a graph showing a distribution of measured data in the case of presence of a purge gas and in the case of absence of a purge gas.Fig. 8(a) is a block diagram schematically showing a turbo molecular pump including a model creation unit and a determination unit, and Fig. 8(b) is a block diagram schematically showing a control device including a model creation unit and a determination unit.Fig. 9(a) is a diagram showing a Mahalanobis distance in an MT method, and Fig. 9(b) is a diagram showing a calculation formula of a Mahalanobis distance.Fig. 10 is a diagram showing an x2 distribution of average 1 according to an MT method. Fig. 11 is a diagram showing a calculation formula related to a logistic regression model. Fig. 12 is a diagram showing a neural network related to a deep learning model.Fig. 13 is a diagram showing data for learning (parameter estimation) and data for accuracy verification for each model.Fig. 14(a) is a graph showing a relationship between normal data and abnormal data when a feature value item A is adopted for an MT method, and Fig. 14(b) is a graph showing a relationship between normal data and abnormal data when a feature value item B is adopted for an MT method.A lower diagram in Fig. 15(a) is a graph showing temperature data when a feature value item A is adopted for the MT method, an upper diagram in Fig. 15(a) is a graph showing aMahalanobis distance, a lower diagram in Fig. 15(b) is a graph showing temperature data when a feature value item B is adopted for an MT method, and an upper diagram in Fig. 15(b) is a graph showing a Mahalanobis distance.A lower diagram in Fig. 16(a) is a graph showing temperature data related to a logistic regression model, an upper diagram in Fig. 16(a) is a graph showing an output, a lower diagram in Fig. 16(b) is a graph showing temperature data related to a deep learning model, and an upper diagram in Fig. 16(b) is a graph showing an output.[Description of Embodiments]

[0015] <Basic configuration example of vacuum pump>Fig. 1 shows a schematic vertical cross-section of a vacuum pump (here, a turbo molecular pump 100) according to an embodiment of the present invention. The turbo molecular pump 100 is configured to be connected to a vacuum chamber (not shown) of a target apparatus such as a semiconductor manufacturing device.

[0016] In the turbo molecular pump 100, an inlet port 101 is formed at an upper end of a cylindrical outer cylinder 127. In addition, a rotating body 103 having a plurality of rotor blades 102 (102a, 102b, 102c, ...) formed radially in multiple stages on its circumferential portion, which are turbine blades for suctioning and exhausting a gas, is provided inside the outer cylinder 127. A rotor shaft 113 is attached to the center of the rotating body 103, and the rotor shaft 113 is supported in the air and position-controlled by, for example, a five-axis controlled magnetic bearing.

[0017] Upper radial electromagnets 104 include four electromagnets disposed in pairs on the X axis and the Y axis. Four upper radial sensors 107 are provided adjacent to the upper radial electromagnets 104 and provided corresponding to the upper radial electromagnets 104. As the upper radial sensors 107, for example, an inductance sensor or an eddy-current sensor having a conductive coil is used, and the position of the rotor shaft 113 is detected on the basis of a change in inductance of the conductive coil which is changed depending on the position of the rotor shaft 113. The upper radial sensors 107 are configured to detect a radialdisplacement of the rotor shaft 113, that is, the rotating body 103 fixed thereto, and send a detection result to a control device 200.

[0018] In this control device 200, for example, a compensation circuit having a PID adjusting function generates an excitation control command signal for the upper radial electromagnet 104 on the basis of a position signals detected by the upper radial sensors 107, and an amplifier circuit 150 (to be described below) shown in Fig. 2 performs excitation control on the upper radial electromagnet 104 on the basis of the excitation control command signal, thereby adjusting a radial position of the rotor shaft 113 on an upper side.

[0019] In addition, the rotor shaft 113 is made of a high magnetic permeability material (iron, stainless steel, or the like) or the like, and is suctioned by a magnetic force of the upper radial electromagnet 104. Such adjustments are performed independently in the X axis and Y axis directions. In addition, lower radial electromagnets 105 and lower radial sensors 108 are disposed in the same manner as the upper radial electromagnet 104 and the upper radial sensors 107, and adjust a radial position of a lower side of the rotor shaft 113 in the same manner as the radial position of the upper side.

[0020] Further, axial electromagnets 106A and 106B are disposed to vertically sandwich a disc-shaped metal disc 111 (also referred to as an “armature disc”) provided at a lower portion of the rotor shaft 113. The metal disc 111 is made of a high magnetic permeability material such as iron. An axial sensor 109 is provided to detect an axial displacement of the rotor shaft 113 and is configured to transmit its axial position signal to the control device 200.

[0021] Further, in the control device 200, for example, the compensation circuit having the PID adjusting function generates excitation control command signals for each of the axial electromagnets 106A and 106B on the basis of the axial position signal detected by the axial sensor 109, and the amplifier circuit 150 performs excitation control on each of the axial electromagnet 106Aand the axial electromagnet 106B on the basis of these excitation control command signals, whereby the axial electromagnet 106A suctions the metal disc 111 upward using a magnetic force, the axial electromagnet 106B suctions the metal disc 111 downward, and thus an axial position of the rotor shaft 113 is adjusted.

[0022] Consequently, the control device 200 is configured to appropriately adjust the magnetic force on the metal disc 111 which is generated by the axial electromagnets 106A and 106B, cause the rotor shaft 113 to be magnetically held in the air in the axial direction, and hold the rotor shaft 113 in a space in a non-contact manner. The amplifier circuit 150 that performs the excitation control on the upper radial electromagnet 104, the lower radial electromagnet 105, and the axial electromagnets 106Aand 106B will be described below.

[0023] On the other hand, a motor 121 includes a plurality of magnetic poles disposed circumferentially to surround the rotor shaft 113. Each of the magnetic poles is controlled by the control device 200 to rotate the rotor shaft 113 via electromagnetic forces acting between the magnetic poles and the rotor shaft 113. In addition, the motor 121 incorporates, for example, a rotation speed sensor such as a Hall element, a resolver, or an encoder (not shown), and a rotation speed of the rotor shaft 113 is detected by a detection signal from the rotation speed sensor.

[0024] Further, a phase sensor (not shown) is attached in the vicinity of, for example, the lower radial sensors 108 to detect a phase of rotation of the rotor shaft 113. The control device 200 is configured to use detection signals of both the phase sensor and the rotation speed sensor and detect positions of the magnetic poles.

[0025] A plurality of stator blades 123 (123a, 123b, 123c, ...) are disposed at small gaps (predetermined intervals) from the rotor blades 102 (102a, 102b, 102c, ...). The rotor blades 102 (102a, 102b, 102c, ...) are formed at a predetermined angle from a plane perpendicular to the axis of the rotor shaft 113 in order to transport exhaust gas molecules downward by collision.

[0026] Further, similarly, the stator blades 123 are also formed to be inclined by a predetermined angle from a plane perpendicular to the axis of the rotor shaft 113, and are disposed toward an inside of the outer cylinder 127 alternately with the stages of the rotor blades 102. In addition, outer circumferential ends of the stator blades 123 are supported while they are inserted between stator blade spacers 125 (125a, 125b, 125c, ...) stacked in multiple stages.

[0027] The stator blade spacer 125 is a ring-shaped member and is made of, for example, a metal such as aluminum, iron, stainless steel, or copper, or a metal such as an alloy containing these metals as components. The outer cylinder 127 is fixed to outer circumferences of the stator blade spacers 125 with slight gaps interposed therebetween. A base portion 129 is disposed at a bottom portion of the outer cylinder 127. An outlet port 133 is formed in the base portion 129 and communicates with the outside. The exhaust gas that enters the inlet port 101 from the chamber (vacuum chamber) side and is transported to the base portion 129 is sent to the outlet port 133.

[0028] Further, depending on the use of the turbo molecular pump 100, a screw stator 131 is disposed on the inner side of the base portion 129. The screw stator 131 is a cylindrical member made of a metal such as aluminum, copper, stainless steel, iron, or an alloy containing these metals as components, and has a plurality of helical screw grooves 131a engraved in its inner circumferential surface. A helical direction of the screw grooves 131a is a direction in which the molecules of the exhaust gas are transported toward the outlet port 133 when the molecules move in a rotation direction of the rotating body 103. A rotating body lower cylindrical portion 103b hangs down from the bottom of a rotating main body 103a on which the rotor blades 102 (102a, 102b, 102c, ...) of the rotating body 103 are formed. An outer circumferential surface of the rotating body lower cylindrical portion 103b has a cylindrical shape and protrudes toward the inner circumferential surface of the screw stator 131 , and is close to the inner circumferential surface of the screw stator 131 with a predetermined gap interposed therebetween. The exhaust gas transferred to the screw grooves 131a by the rotor blades 102 and the stator blades 123 is sent to the base portion 129 while being guided by the screw grooves 131a. In this manner, the screw stator 131 and the rotating body lower cylindrical portion 103b opposed to the screw stator configure a Holweck type exhaust mechanism 204. The Holweck type exhaust mechanism 204 imparts directivity to the exhaust gas by rotation of the rotating body lower cylindrical portion 103b relative to the screw stator 131 , and improves the exhaust characteristics of the turbo molecular pump 100.

[0029] The base portion 129 is a disc-shaped member forming a base bottom portion of the turbo molecular pump 100, and is generally made of a metal such as iron, aluminum, or stainless steel. Since the base portion 129 physically holds the turbo molecular pump 100 and also functions as a heat conduction path, it is desirable to use a metal that has rigidity and high thermal conductivity, such as iron, aluminum, or copper, for the base portion 129.

[0030] In the configuration of the turbo molecular pump 100, when the rotor blades 102 are driven to rotate together with the rotor shaft 113 by the motor 121 , the exhaust gas is suctioned from a chamber through the inlet port 101 by the action of the rotor blades 102 and the stator blades 123. The exhaust gas taken in from the inlet port 101 passes between the rotor blades 102 and the stator blades 123 and is transferred to the base portion 129. At this time, the temperature of the rotor blade 102 is increased due to frictional heat generated when the exhaust gas contacts the rotor blade 102 conduction of heat generated by the motor 121 , or the like, but this heat is transmitted to the stator blade 123 side by radiation or conduction through gas molecules of the exhaust gas (gas molecules).

[0031] The stator blade spacers 125 are joined to each other at their outer circumferential portions and transfer heat received by the stator blades 123 from the rotor blades 102, frictional heat generated when the exhaust gas comes into contact with the stator blades 123, or the like to the outside.

[0032] In the above description, it has been described that the screw stator 131 is disposed at the outer circumference of the rotating body lower cylindrical portion 103b of the rotating body 103, and the screw grooves 131a are engraved on the inner circumferential surface of the screw stator 131. However, on the contrary, screw grooves may be engraved in the outer circumferential surface of the rotating body lower cylindrical portion 103b, and a spacer having a cylindrical inner circumferential surface may be disposed around the screw grooves.

[0033] In addition, depending on an application of the turbo molecular pump 100, a periphery of an electrical installation portion is covered with a stator column 122 so that the gas suctioned from the inlet port 101 does not infiltrate the electrical installation portion including the upper radial electromagnet 104, the upper radial sensors 107, the motor 121 , the lower radialelectromagnet 105, the lower radial sensors 108, the axial electromagnets 106A and 106B, the axial sensor 109, and the like, and the inside of the stator column 122 may be kept at a predetermined pressure by a purge gas.

[0034] In this case, a purge gas port 211 is disposed in the base portion 129, and a purge gas (protective gas, inert gas) is introduced through the purge gas port 211. The introduced purge gas is sent to the outlet port 133 through a gap between a protection bearing 120 and the rotor shaft 113, a gap between the rotor and the stator of the motor 121 , and a gap between the inner peripheral side cylindrical portion (rotating body lower cylindrical portion 103b) of the rotor blade 102, the stator column 122 and the base portion 129. In the present embodiment, the state of the flow rate of the purge gas is monitored, but the monitoring of the purge gas will be described later.

[0035] Here, the model of the turbo molecular pump 100 needs to be specified, and the turbo molecular pump 100 needs to be controlled on the basis of individually adjusted unique parameters (for example, characteristics corresponding to the model). In orderto store these control parameters, the turbo molecular pump 100 includes an electronic circuit unit 141 in its main body. The electronic circuit unit 141 is configured of a semiconductor memories such as an EEP-ROM, electronic components such as semiconductor elements for accessing to the semiconductor memories, a substrate 143 for mounting these components, and the like. The electronic circuit unit 141 is housed below the rotation speed sensor (not shown), for example, near a center of the base portion 129 forming a lower portion of the turbo molecular pump 100, and is closed by an airtight bottom cover 145.

[0036] Incidentally, in the semiconductor manufacturing process, some process gases introduced into the chamber have the property of becoming solid when their pressure becomes higher than a predetermined value or their temperature becomes lower than a predetermined value. Inside the turbo molecular pump 100, a pressure of the exhaust gas is lowest at the inlet port 101 and highest at the outlet port 133. When the pressure of the process gas becomes higher than the predetermined value or the temperature thereof becomes lower than the predetermined value in the middle of being transported from the inlet port 101 to the outletport 133, the process gas becomes solid, and adheres to and is deposited inside the turbo molecular pump 100.

[0037] For example, it can be found from the vapor pressure curve that, when SiCI4 is used as a process gas in an Al etching device, solid products (for example, AICI3, also referred to as “reaction products”) precipitate, adhere to and deposit inside the turbo molecular pump 100 at low vacuum (760 torr to 102torr) and low temperature (approximately 20°C). Thus, when precipitates of the process gas are deposited inside the turbo molecular pump 100, the deposits narrow a pump flow path, causing a decrease in the performance of the turbo molecular pump 100. Then, the above-described products are prone to solidifying and adhering to high-pressure portions near the outlet port 133 and the screw stator 131.

[0038] Forthat reason, in order to solve this problem, in the related art, a heater (not shown) and an annular water-cooled pipe 149 are wound around the outer circumference of the base portion 129 or the like, a temperature sensor (not shown) (for example, a thermistor) is embedded, for example, in the base portion 129, and control of heating by the heater or cooling by the water-cooled pipe 149 is performed (hereinafter referred to as a temperature management system (TMS)) so that the temperature of the base portion 129 is maintained at a certain high temperature (set temperature) on the basis of a signal from the temperature sensor. In the present embodiment, the screw stator 131 is heated by a heater (not shown) embedded in the screw stator 131, and the base portion 129 is cooled by the water-cooled pipe 149 embedded in the bottom cover 145.

[0039] Next, the amplifier circuit 150 that performs the excitation control on the upper radial electromagnet 104, the lower radial electromagnet 105, and the axial electromagnets 106A and 106B will be described with respect to the turbo molecular pump 100 configured as described above. Fig. 2 shows a circuit diagram of the amplifier circuit.

[0040] In Fig. 2, an electromagnet coil 151 constituting the upper radial electromagnet 104 or the like has one end connected to a positive electrode 171a of a power source 171 via a transistor 161 and the other end connected to a negative electrode 171b of the power source 171 via a current detection circuit 181 and a transistor 162. In addition, the transistors 161and 162 are so-called power MOSFETs, and have a structure in which a diode is connected between a source and a drain thereof.

[0041] In this case, the transistor 161 has a configuration in which a cathode terminal 161a of its diode is connected to the positive electrode 171a, and an anode terminal 161b thereof is connected to one end of the electromagnet coil 151. In addition, the transistor 162 has a configuration in which a cathode terminal 162a of its diode is connected to the current detection circuit 181 , and an anode terminal 162b thereof is connected to the negative electrode 171 b.

[0042] On the other hand, a current regeneration diode 165 has a configuration in which a cathode terminal 165a thereof is connected to one end of the electromagnet coil 151 , and an anode terminal 165b thereof is connected to the negative electrode 171b. In addition, similarly, a current regeneration diode 166 has a configuration in which a cathode terminal 166a thereof is connected to the positive electrode 171a, and an anode terminal 166b thereof is connected to the other end of the electromagnet coil 151 via the current detection circuit 181 . In addition, the current detection circuit 181 is configured of, for example, a Hall sensor type current sensor or an electric resistance element.

[0043] The amplifier circuit 150 configured as above corresponds to one electromagnet. For that reason, when the magnetic bearing is under five-axis control and a total of ten electromagnets 104, 105, 106A, and 106B are provided, the same amplifier circuits 150 are formed for each of the electromagnets, and the ten amplifier circuits 150 are connected to the power source 171 in parallel.

[0044] Further, an amplifier control circuit 191 is configured of, for example, a digital signal processor unit (hereinafter referred to as a DSP unit) (not shown) of the control device 200, and this amplifier control circuit 191 switches the transistors 161 and 162 on and off.

[0045] The amplifier control circuit 191 compares a current value detected by the current detection circuit 181 (a signal obtained by reflecting the current value is referred to as a current detection signal 191c) with a predetermined current command value. In addition, on the basis of the comparison result, it determines magnitudes of pulse widths (pulse width timesTp1 and Tp2) to be generated within a control cycle Ts, which is one period of PWM control. As a result, gate drive signals 191a and 191 b having the pulse widths are output from the amplifier control circuit 191 to gate terminals of the transistors 161 and 162.

[0046] It is necessary to perform position control on the rotating body 103 at a high speed and in a strong force, when resonance occurs during a rotation speed acceleration operation or a disturbance occurs during a constant speed operation in the rotating body 103, for example. Therefore, for example, a high voltage of approximately 50 V is used as the power source 171 so that a current flowing in the electromagnet coil 151 can be rapidly increased (or decreased). In addition, a capacitor (not shown) is usually connected between the positive electrode 171a and the negative electrode 171b of the power source 171 in order to stabilize the power source 171.

[0047] In such a configuration, a current flowing in the electromagnet coil 151 (hereinafter, referred to as an electromagnet current iL) is increased when both the transistors 161 and 162 are turned on, and the electromagnet current iL is decreased when both the transistors 161 and 162 are turned off.

[0048] In addition, when one of the transistors 161 and 162 is turned on and the other is turned off, a so-called flywheel current is maintained. In addition, by causing the flywheel current to flow through the amplifier circuit 150 in this way, a hysteresis loss in the amplifier circuit 150 can be reduced, and power consumption of the entire circuit can be kept low. Moreover, by controlling the transistors 161 and 162 in this way, high-frequency noise such as harmonics generated in the turbo molecular pump 100 can be reduced. Further, by measuring the flywheel current with the current detection circuit 181 , the electromagnet current iL flowing through the electromagnet coil 151 can be detected.

[0049] In other words, in a case where the detected current value is lower than the current command value, both the transistors 161 and 162 are turned on only once during a control cycle Ts (for example, 100 ps) by a time corresponding to the pulse width time Tp1 as shown in Fig. 3. Therefore, the electromagnetic current iL during this period is increased toward acurrent value iLmax (not shown) that can flow from the positive electrode 171a to the negative electrode 171 b via the transistors 161 and 162.

[0050] Meanwhile, in a case where the detected current value is larger than the current command value, both the transistors 161 and 162 are turned off only once during the control cycle Ts by a time corresponding to the pulse width time Tp2 as shown in Fig. 4. Therefore, the electromagnet current iL during this period is decreased toward a current value iLmin (not shown) that can be regenerated from the negative electrode 171 b to the positive electrode 171a via the diodes 165 and 166.

[0051] In either case, after the pulse width times Tp1 and Tp2 have passed, either one the transistor 161 or 162 is turned on. For this reason, during this period, the flywheel current is maintained in the amplifier circuit 150.

[0052] In the turbo molecular pump 100 having such a basic configuration, the upper side (the side of the inlet port 101) in Fig. 1 serves as an inlet portion connected to the side of a target apparatus, and the lower side (the side of the base portion 129 on which the outlet port 133 is provided) serves as an outlet portion connected to an auxiliary pump (back pump) or the like which is not shown in the drawing. The turbo molecular pump 100 can be used not only in a vertical posture in a vertical direction as shown in Fig. 1 but also in an inverted vertical posture, a horizontal posture, or an inclined posture.

[0053] In the turbo molecular pump 100, the above-mentioned outer cylinder 127 and the base portion 129 are combined to configure a single case. In the following description, both the outer cylinder 127 and the base portion 129 may be collectively referred to as a “casing” or a “main body casing”. In addition, only the outer cylinder 127 and only the base portion 129 may be referred to as a “casing”. The turbo molecular pump 100 is electrically (and structurally) connected to a box-shaped electrical case (not shown), and the above-mentioned control device 200 is incorporated in the electrical case.

[0054] The internal configuration of the main body casing (here, a combination of the outer cylinder 127 and the base portion 129) of the turbo molecular pump 100 can be divided into a rotating mechanism 136 for rotating the rotor shaft 113 and the like by the motor 121 and anexhaust mechanism 137 driven to rotate by the rotating mechanism 136. The exhaust mechanism 137 can also be considered to be divided into a turbo molecular pump mechanism 138 which is configured with the rotor blades 102, the stator blades 123, and the like, and a screw groove pump mechanism (Holwecktype exhaust mechanism 204) which is configured with the rotating body lower cylindrical portion 103b, the screw stator 131 , and the like.

[0055] In addition, the above-mentioned purge gas (protective gas, inert gas) is used to protect the bearing portion, the rotor blades 102, and the like, prevents corrosion due to the exhaust gas (process gas), and performs cooling of the rotor blades 102. The supply of the purge gas can be performed by a general method.

[0056] For example, the purge gas port 211 extending linearly in the radial direction is provided at a predetermined portion (such as a position spaced 90 degrees or 120 degrees away from the outlet port 133) of the base portion 129. Then, the purge gas is supplied to the purge gas port 211 from the outside of the base portion 129 through a purge gas cylinder (N2 gas cylinder or the like), a flow rate regulator (valve device), or the like. In the present embodiment, the state of the flow rate of the purge gas is monitored, but the monitoring of the purge gas will be described later.

[0057] The above-mentioned protective bearing 120 is also referred to as a "touchdown (T / D) bearing", a "backup bearing", or the like. These protective bearings 120 do not greatly change the position and posture of the rotor shaft 113 and prevents the rotor blade 102 and its peripheral portion from being damaged even when, for example, an electric system trouble or a trouble such as atmospheric inrush occurs.

[0058] In Fig. 1 showing the structure of the turbo molecular pump 100 and the rotating body 103, description of a hatching showing the cross-section of the parts is omitted to avoid the complication of the drawings.

[0059] <Monitoring of purge gas>ccOutline of monitoring method»In the turbo molecular pump 100 described above, a purge gas monitoring device 210 is formed in the electronic circuit unit 141. The purge gas monitoring device 210 can monitora state related to the flow rate of the purge gas (purge gas flow rate state). When the purge flow rate is sufficient, it is possible to prevent the temperature of the parts in the turbo molecular pump 100 from excessively rising. The outline of the monitoring of the purge gas flow rate state is as follows.

[0060] (1) The temperature of parts opposed to the rotating body 103 via the purge gas and the temperature of parts existing in a heat path (heat exhaust path) from the parts to the water-cooled pipe 149 are used as measured data (explanatory variables), and the purge gas flow rate is estimated and abnormality is detected by a statistical method and machine learning.(2) In order to improve the accuracy of abnormality detection, the following is performed:(a) A plurality of parts (and / or portions) that are independent of the heat exhaust path are selected as the parts opposed to the rotating body 103.The parts (and / or portions) that are independent of the heat exhaust path include, for example, the motor 121 , the stator column 122, the rotating body 103, the base portion 129, and the like. In addition, one part may be present in part in a plurality of heat exhaust paths. Examples of such a part include a temperature sensor holding member 220, which will be described later, and the like, (b) When a part itself generates heat, data correlated with the amount of heat generation is included in measured data (explanatory variables).Examples of the part that generates heat by itself include the motor 121 , the electrical installation portion, and the like (the electronic circuit unit 141 and the like). The data correlated with the amount of heat generation include a current value of the motor 121 , a temperature value of the electrical installation portion and the like (the electronic circuit unit 141 and the like).(3) As a statistical method, for example, an MT system or the like can be used. As the machine learning, for example, logistic regression, deep learning, and the like can be used.

[0061] «Heat exhaust path and temperature sensor»For example, the rotating body 103 and parts with a flow path for a purge gas interposed between the parts and the rotating body 103 (hereinafter referred to as "opposingparts") will be considered. The "opposing part" mentioned here means a part opposed to the flow path forthe purge gas. Examples ofthe “opposing part” forthe rotating body 103 include the stator column 122, the base portion 129, and the like. The "oppose" can also be rephrased as, for example, "confront", "face", or "look onto".

[0062] Fig. 5 is an enlarged view of a portion surrounded by a rectangular frame C in Fig. 1. As shown in Fig. 5, a space 216 is interposed between an outer periphery 212 of the stator column 122 and an inner periphery 214 ofthe rotating body 103. The space 216 configures a flow path for a purge gas. As the rotating body 103 rotates, the purge gas is introduced into the space 216.

[0063] The purge gas flows into an internal space 218 ofthe base portion 129 from the purge gas port 211 and is introduced into the stator column 122. As described above, the purge gas is sent to the outlet port 133 through a gap between the protective bearing 120 and the rotor shaft 113, a gap between the rotor and the stator in the motor 121 , and a gap between an inner peripheral side cylindrical portion (rotating body lower cylindrical portion 103b) ofthe rotor blade 102, the stator column 122, and the base portion 129.

[0064] Fig. 6 schematically shows the configuration shown in Fig. 5. Furthermore, in Fig. 6, a flow of a purge gas flowing from the inside to the outside ofthe stator column 122, passing through the space 216 between the stator column 122 (and the base portion 129) and the rotating body 103, and heading toward the base portion 129 is indicated by an arrow E. While the purge gas flows from the inside to the outside of the stator column 122, the purge gas exchanges heat with apparatuses such as the protective bearing 120, the motor 121 , and the rotating body 103 shown in Fig. 5. Then, the purge gas is continuously supplied to remove and discharge heat (heat exhaust) from these apparatuses.

[0065] Hereinafter, as indicated by the arrow E in Fig. 6, the flow path ofthe purge gas which includes the space 216 between the stator column 122 (and the base portion 129) and the rotating body 103 will be referred to as a "first heat exhaust path". The "first heat exhaust path" passes through the outer space 216 from an inner space 213 of the stator column 122 and reaches the outer space ofthe rotating body 103.

[0066] The heat of the stator column 122 and the like is also discharged through a cooling apparatus such as the water-cooled pipe 149. An arrow F in Fig. 6 schematically shows a path of heat discharged through the base portion 129 from the motor 121 through the inside of the stator column 122 (the inside of a wall, the inside of a material) by using cooling water (cooling liquid) flowing through the water-cooled pipe 149. The stator column 122 is made of a metal such as iron, aluminum, or stainless steel. The stator column 122 and the base portion 129 are in contact with each other in a state where heat can be transferred. Hereinafter, the heat exhaust path indicated by the arrow F will be referred to as a “second heat exhaust path”.

[0067] Further, in the present embodiment, a "third heat exhaust path" exists between the "first heat exhaust path" and the "second heat exhaust path" as indicated by an arrow G. The "third heat exhaust path" passes through the base portion 129 and reaches the water-cooled pipe 149.

[0068] In the present embodiment, the first heat exhaust path (arrow E), the second heat exhaust path (arrow F), and the third heat exhaust path (arrow G) are considered to be different heat exhaust paths. In the present embodiment, the "different heat exhaust paths" are heat paths that can be distinguished as each independent heat exhaust paths at least in part.

[0069] The "different heat exhaust paths" are formed by a single or combination of the abovedescribed "parts independent of the heat exhaust path" or spaces between a plurality of "parts independent of the heat exhaust path". The "different heat exhaust path" has at least a section where lines drawn along the heat exhaust path (which may be the above-mentioned arrows) do not cross each other on the drawing. Regarding the "different heat exhaust paths", for example, a distinction between them can be made clearer by providing a sufficient distance therebetween.

[0070] Temperature sensors 222A to 222C are provided in the heat exhaust paths such as the first heat exhaust path (arrow E) to the third heat exhaust path (arrow G). In the example of Fig. 6, the temperature sensors 222Ato 222C are provided for the rotating body 103, thestator column 122, and the motor 121. In Fig. 6, portions of the stator column 122 and the base portion 129 on one side from the central axis (portions on the right side from the central axis in the drawing) are integrally simplified and shown. The temperature sensor 222A for the rotating body 103 is provided on the temperature sensor holding member 220 mentioned above and is disposed to face the first heat exhaust path (arrow E) through which the purge gas passes.

[0071] The temperature sensor holding member 220 is formed in a cylindrical shape. The temperature sensor holding member 220 includes a top portion (heat receiving portion) 224 formed in a disc shape to increase an area facing the rotating body 103, and a bottom portion (mounting portion) 226 formed in a flange shape to be fixed to the base portion 129 (or the stator column 122) at an end portion in an axial direction (vertical direction in Fig. 6 and the like). The top portion 224 is located on the side of the inlet port 101 (upper side in Fig. 6 and the like), and the bottom portion 226 is located on the side of the base portion 129 (lower side in Fig. 6 and the like). The temperature sensor holding member 220 is disposed such that the top portion 224 faces the rotating body lower cylindrical portion 103b via a gap, and detects the temperature of the rotating body 103 by the temperature sensor 222A in a non-contact manner. The top portion 224 preferably has a sufficiently large cross-sectional area to detect heat transfer from the rotating body 103 to be measured.

[0072] The temperature sensor holding member 220 detects the temperature of a gap (a gap through which the purge gas flows) between the top portion (heat receiving portion) 224 and the rotating body 103. In addition, only one temperature sensor holding member 220 is disposed in the circumferential direction of the turbo molecular pump 100. For this reason, in a region where the temperature sensor holding member 220 is not disposed, the purge gas flows through a relatively wide space as indicated by a dashed arrow H.

[0073] One temperature sensor 222B out of the temperature sensors 222B and 222C related to the stator column 122 is provided at the bottom portion 226 of the temperature sensor holding member 220. The temperature sensor 222B is not limited to being disposed at the bottom portion 226, but may be disposed, for example, in the stator column 122. The othertemperature sensor 222C is provided on the stator column 122 and used to detect the temperature of the motor 121. The temperature sensor 222C for the motor 121 is located in the second heat exhaust path (arrow F). In the second heat exhaust path (arrow F), the temperature sensor 222C for the motor 121 is located on a side further away from the water-cooled pipe 149 than the temperature sensor 222B at the bottom portion 226 in the temperature sensor holding member 220.

[0074] In the present embodiment, the third heat exhaust path (arrow G) extends from the lower side of the bottom portion 226 of the temperature sensor holding member 220 to the water-cooled pipe 149 through the base portion 129 and the like and reaches the water-cooled pipe 149.

[0075] Various installation locations can be set for the temperature sensors 222Ato 222C as long as the temperature sensors are installed in heat paths that can be distinguished as different heat exhaust paths (here, the first heat exhaust path (arrow E) to the third heat exhaust path (arrow G)) and as long as the heat exhaust paths to which the temperature sensors belong can be distinguished from each other. The temperature sensors 222A to 222C are not limited to being installed on the heat exhaust paths as long as the temperatures in the heat exhaust paths to which the temperature sensors belong can be detected, but the temperature sensors may be disposed, for example, at locations facing (looking onto) the heat exhaust paths to which the temperature sensors belong or at adjacent locations.

[0076] As the temperature sensors (the temperature sensors 222Ato 222C in the example of Fig. 6), a thermistor or the like can be used, but all temperature sensors need not be of the same type or model. Further, various methods such as resin sealing, resin potting, sticking and embedding can be adopted to fix the temperature sensor. In addition, the number of temperature sensors can be increased or decreased as required.

[0077] «Principle of purge gas monitoring»Depending on the presence or absence of a purge gas and a flow rate, a flow rate of heat to a part (opposing part) opposed to a flow path of the purge gas changes. Then, the temperature (including a temperature distribution) of the opposing part changes.Furthermore, not only the temperature of the opposing part but also the temperature of parts (hereinafter referred to as "peripheral parts") disposed around the opposing part changes.

[0078] The "peripheral parts" may include, for example, a part that is in contact with the opposing part in a state where heat transfer is possible, and parts between the opposing part and the water-cooled pipe 149. For example, when the stator column 122 is an opposing part, parts corresponding to the “peripheral parts” may include the base portion 129, a part (including the water-cooled pipe 149) mounted on the base portion 129, and the like. In addition, the "peripheral parts" also include parts in which at least a portion of the inside of a wall (material) is an heat exhaust path.

[0079] A heat flux flows through the heat exhaust path (here, the first heat exhaust path to the third heat exhaust path). Furthermore, detection results of the temperature sensors 222Ato 222C described above are temperature data obtained from the heat flux.

[0080] The inventors plotted and visualized the detection results of the temperature sensors when there is a purge gas (presence of a purge gas) and when there is no purge gas (absence of a purge gas) on a 3-axis (three-dimensional) graph by using a turbo molecular pump with an exhaust volume of 5000 L / s. In a graph shown in Fig. 7, respective axes indicate the temperature of the motor 121 (motor temperature), the temperature of the top portion 224 in the temperature sensor holding member 220 (top portion temperature), and the temperature of the bottom portion 226 in the temperature sensor holding member 220 (bottom portion temperature).

[0081] The motor temperature corresponds to the detection result of the temperature sensor 222C shown in Fig. 6. Further, the top portion temperature corresponds to the detection result of the temperature sensor 222A, and the bottom portion temperature corresponds to the detection result of the temperature sensor 222B. The unit of each axis is °C. The individual points configuring a point group in the graph are represented with, for example, (a motor temperature value, a top portion temperature value, and a bottom portion temperature value) as coordinates.

[0082] Each point in the graph of Fig. 7 can be captured as being divided into two point groups (a cluster P1 and a cluster P2) separated in the axial direction of the top portion temperature (vertical direction in the drawing) in a three-dimensional space. The lower cluster P1 (shown by being surrounded by a solid line) is a detection result acquired when there is a purge gas, and the upper cluster P2 (shown by being surrounded by a dashed line) is a detection result acquired when there is no purge gas.

[0083] In a space between the clusters P1 and P2, there are no points, or the density of a point group is smaller than the density of the point groups in the clusters P1 and P2. The density of the point group can be measured or estimated by various general methods, such as a method of counting the number of points per unit area and a method of statistically processing a distance between each point and the nearest point to estimate the density.

[0084] When there is a purge gas, the supply of a specified amount of purge gas is performed. When there is no purge gas, the supply of a specified amount of purge gas is not performed. When the specified amount of purge gas is not supplied, the flow rate of the purge gas is not sufficient for the specified amount, or no purge gas is supplied.

[0085] It was found from the detection results shown in Fig. 7 that in the three-dimensional space, temperature data based on a predetermined process gas flow rate can be clustered in a separated state depending on the presence or absence of a purge gas. Furthermore, the inventors came up with the idea of using a correlation between the purge gas and the ambient temperature and applying a statistical method and machine learning to results of temperature detection (temperature detection results) to non-linearly model the correlation between the purge gas and the ambient temperature and estimate the flow rate state of the purge gas.

[0086] The estimation related to the flow rate state of the purge gas can be performed by creating a determination model for determining the flow rate state of the purge gas and determining the flow rate state of the purge gas using the determination model. The creation of the determination model can be performed using a statistical method and machine learning. As the statistical method, for example, an MT method, logistic regression, and the like can be applied. Further, as the machine learning, deep learning or the like can be applied.

[0087] Further, when the flow rate of the process gas increases or decreases, the state of heat generation (heat generation state) of the motor 121 itself fluctuates, and the temperatures of the motor 121 and its peripheral parts are affected. In addition, the amount of heat generated by the motor 121 is also related to the value of a motor current, the value of the temperature of the electrical installation circuit, and the like. For this reason, the value of the motor current, the value of the temperature of the electrical installation circuit, and the like are also used as measurement data (explanatory variables related to machine learning, reference data related to a statistical method), and thus it is considered that the flow rate state of the purge gas can be estimated with higher accuracy. Here, the temperature of the electrical installation circuit increases due to heat generated by a motor driver (for example, the amplifier control circuit 191 (Fig. 2) or the like).

[0088] Creation of the determination model and determination using the created determination model can be performed using a computer apparatus in which Al (artificial intelligence) can be constructed. In the present embodiment, the computer apparatus is configured by providing a necessary computer apparatus (central processing unit (CPU) or the like) in the electronic circuit unit 141 shown in Fig. 1.

[0089] When a necessary computer apparatus is provided in the electronic circuit unit 141 , the turbo molecular pump 100 includes a determination model creation unit 232 and a determination unit 234 as shown in Fig. 8(a). The determination model creation unit 232 creates a determination model by executing a statistical method and machine learning. The determination unit 234 performs determination using the determination model created by the determination model creation unit 232. The determination model creation unit 232 and the determination unit 234 are functional parts (functional configurations) configured using a CPU (not shown) and its peripheral apparatuses (storage unit, communication unit, and the like).

[0090] However, this is not limiting, and for example, it is also possible to use the control device 200 as a computer apparatus. In this case, as shown in Fig. 8(b), the control device 200 includes the determination model creation unit 232 and the determination unit 234.

[0091] The electronic circuit unit 141 and the control device 200 including the determination model creation unit 232 and the determination unit 234 configure a purge gas monitoring device. It is also possible to capture the turbo molecular pump 100 equipped with the electronic circuit unit 141 and the control device 200 as a purge gas monitoring device and a purge gas monitoring system. Furthermore, a combination of the mechanical structure of the turbo molecular pump 100 and the control device 200 may also be collectively referred to as a "turbo molecular pump".

[0092] The mechanical structure of the turbo molecular pump 100 includes mechanical structure parts such as the outer cylinder 127, the base portion 129, the rotating body 103, the rotor shaft 113, the motor 121 , the stator column 122, and the screw stator 131 , the electronic circuit unit 141 , and the like.

[0093] Further, a control apparatus other than the control device 200 includes the determination model creation unit 232 and the determination unit 234. When the control apparatus is combined with the turbo molecular pump 100, a combination of the control apparatus and the turbo molecular pump 100 configures a purge gas monitoring device and a purge gas monitoring system. Also in this case, the machine structure of the turbo molecular pump 100, and the combination of the control device 200 and the control apparatus can also be collectively referred to as a "turbo molecular pump".

[0094] «MT method»As a statistical method in product design or the like, an MT system (Mahalanobis-Taguchi system) is known. The MT system is a general term for a group of methods that combine multivariate analysis with the theory of quality engineering, and is used for abnormality detection, pattern recognition, and specifying factors that affect response variables.

[0095] In an MT method in the MT system, it is determined whether it is normal or abnormal on the basis of a Mahalanobis distance (MD = D2, MD value). A normal population is referred to as a "unit space", which is used as a criterion. Then, quantitative determination is made on a sample to be determined in accordance with the degree of separation from the unit space.

[0096] Fig. 9(a) schematically shows the idea of such a Mahalanobis distance. Data having a relatively small distance with respect to a unit space indicated by an ellipse is treated as normal data, and data having a relatively large distance is treated as abnormal data (data indicating abnormality).

[0097] The Mahalanobis distance (multivariate Mahalanobis distance: D2) is defined by a formula shown in Fig. 8(b). In a formula shown in Fig. 9(b), the number of variables (feature items) is k and the number of pieces of data is n. The formula shown in Fig. 9(b) shows the square of the Mahalanobis distance D of the p-th data.

[0098] As shown in Fig. 9(b), ui is a feature item (standardized feature item), rij is a correlation coefficient between feature items ui and uj.

[0099] The MD value (= D2) in the unit space in the MT method follows a chi-squared distribution with mean 1 (Fig. 10, a 1 -cumulative density function (a complement of a cumulative density function)). The horizontal axis in Fig. 10 shows a Mahalanobis distance (MD), and the vertical axis shows a significance level. The values "3" to "8" in an upper right legend in a graph of Fig. 10 indicate the degree of freedom. The degrees of freedom for respective curves shown in Fig. 10 are "8" to "3" in order from left, for example, when described with reference to the range of 1% to 10% significance levels. The value of the degree of freedom is a numerical value obtained by subtracting the number of constraints from a feature value.

[0100] In the present embodiment, for evaluation of a determination model using the MT method, the following two types of feature value items (feature value item A, feature value item B) are used as feature value items related to feature values.

[0101] Feature value item A (the number of measurement items is 3);Motor temperature (m_temp)Rotor blade temperature sensortop temperature (rotor_top)Rotor blade temperature sensor base temperature (rotor_base)

[0102] Feature value item B (the number of measurement items is 3);Motor temperature (m_temp)Rotor blade temperature sensortop temperature (rotor_top)Rotor blade temperature sensor base temperature (rotor_base)Circuit temperature (control_temp)Motor current (m_current)

[0103] For the feature value item A, the number of measurement items is 3 (three measurement items, that is, the motor temperature, the rotor blade temperature sensor top temperature, and the rotor blade temperature sensor base temperature). In addition, for the feature value item B, the number of measurement items is 5 (five measurement items, that is, the motor temperature, the rotor blade temperature sensor top temperature, the rotor blade temperature sensor base temperature, the circuit temperature, and the motor current).

[0104] For a case where feature value items are determined in this manner, in accordance with a general idea, it is determined that data is normal when MD = 4 or less, and it is determined that data is abnormal when MD = 4 or more. Test data (determination result data using the created determination model) when the feature value items A and B are used will be described later with reference to Figs. 15(a) and 15(b).

[0105] «Logistic regression»Fig. 11 shows a logistic regression (logistic regression analysis) model. Logistic regression is one of machine learning methods. In general, logistic regression is a method making it possible to describe and predict the probability that binary (0 or 1) results (objective variables) occur from a plurality of factors (explanatory variables), as shown in Fig. 11.

[0106] In the logistic regression according to the present embodiment, as shown in Fig. 11 , measurement items (five measurement items, that is, a motor temperature, a rotor blade temperature sensor top temperature, a rotor blade temperature sensor base temperature, a circuit temperature, and a motor current) similar to the above-mentioned feature value item B are used as explanatory variables to be input. Then, the motor temperature, the rotor blade temperature sensor top temperature, the rotor blade temperature sensor base temperature, the circuit temperature, and the motor current are substituted into explanatory variables x1 , x2, ..., and xm.

[0107] For each measurement item, when pieces of measurement data to be acquired (corresponding to the number of temperature sensors used for each measurement item) are one by one, there are five explanatory variables of x1 to x5. However, when one feature value is measured at a plurality of locations, the number of explanatory variables is 6 or more.

[0108] An operation result of a summation portion (Z) is substituted into an activation function to obtain an output (yA: the meaning of y hat in Fig. 11 , ratio explanatory variable). In the present embodiment, "0" obtained as the output (yA) is related to the presence of purge, and "1 " is related to the absence of purge. Test data (determination result data using the created determination model) when using the explanatory variables x1 , x2, ..., and xm will be described later on the basis of Fig. 16(a).

[0109] «Deep learning»As another method for machine learning, for example, a deep learning model as shown in Fig. 12 can be adopted. In the deep learning model shown in Fig. 12, intermediate layers configuring a neural network together with an input layer and an output layer are shown as an intermediate layer 1 (hidden layer 1) and an intermediate layer 2 (hidden layer 2).

[0110] As an input to the input layer, measurement items (five measurement items, that is, a motor temperature, a rotor blade temperature sensor top temperature, a rotor blade temperature sensor base temperature, a circuit temperature, a motor current) similar to the above-mentioned feature value item B are used. For each measurement item, when pieces of measurement data to be acquired (corresponding to the number of temperature sensors used for each measurement item) are one by one, the number of inputs is five. However, when one measurement item is measured at a plurality of locations, the number of inputs is 6 or more.

[0111] In the example shown in Fig. 12, the intermediate layer 1 and the intermediate layer 2 are all bonding layers. Activation functions in the intermediate layer 1 and the intermediate layer 2 are all Relu (Rectified Linear Unit) functions, and the number of nodes is 10. An activation function in the output layer is a softmax function.

[0112] For the neural network in the example shown in Fig. 12, in the creation stage (learning stage) for the determination model, when measurement data acquired in a state of presence of purge is input, an output having a value relatively close to "1" is obtained. When measurement data acquired in a state of absence of purge is input, learning is performed so that an output having a value relatively close to "0" is obtained.

[0113] «Learning of each model (parameter estimation) and accuracy verification data» Fig. 13 shows accuracy verification data for models of an MT method, logistic regression (logistic), and deep learning. Data used for the models includes normal data, abnormal data, and test data as shown on the left side (front side) in Fig. 13.

[0114] The normal data in Fig. 13 is measurement data obtained by setting a flow rate of a purge gas (purge flow rate) supplied to the turbo molecular pump 100 to a "specified amount" which is a normal value.

[0115] The abnormality data in Fig. 13 is measured data obtained by setting the flow rate (purge amount) of the purge gas supplied to the turbo molecular pump 100 to 0 (“purge amount 0”) which is an abnormal value.

[0116] The test data in Fig. 13 is measurement data obtained when the flow rate (purge flow rate) of the purge gas supplied to the turbo molecular pump 100 is changed from the “specified amount” to “0 (zero)”.

[0117] A "content" in an upper part (table head) in Fig. 13 more specifically shows a data acquisition method. For the above-mentioned normal data and abnormal data, a flow rate of a gas to be exhausted (a gas load related to a process gas) and the temperature of water (water temperature) supplied to the water-cooled pipe 149 are changed in a predetermined manner, and measured data per second is used as it is (in a so-called continuous time series). Each of the number of pieces of normal data and the number of pieces of abnormal data is approximately 5000.

[0118] As shown in the column of "MT method" in Fig. 13, in the MT method, the normal data is used as data in a learning stage (learning data) to set a distance in a normal space.Furthermore, in the MT method, the abnormal data and the test data are used in a verification stage using the created model (determination model).

[0119] As shown in the column of "logistic" and "deep learning" in Fig. 13, the normal data and the abnormal data are used as data in a learning stage (learning data) in logistic regression and deep learning. Furthermore, in the logistic regression or the deep learning, test data is used in a verification stage using the created model (determination model).

[0120] Regarding the normal data and the abnormal data in the MT method, as shown in the column of "content", using the measured data as it is in a so-called incontinent time-series is a method slightly deviating from the main point of catching a change in a feature value item due to a noise factor (here, a fluctuation in the purge flow rate) in the MT method.

[0121] In addition, considering that there is a possibility that a deviation will occur in the number of pieces of data due to a measurement time, there is a possibility that using the measurement data as it is in the so-called incontinent time-series will affect all analysis models (here, an MT method, logistic regression, and deep learning). However, it is assumed that data acquisition at the time of product development is facilitated, and that so-called incontinent time-series data is used as it is in consideration of a possibility that measurement data will be acquired online via a public communication line (public communication environment) such as a cloud and learning for creating a model will be performed in the future.

[0122] «Setting of normal space in MT method»Figs. 14(a) and 14(b) show an example of the idea of setting a normal space in an MT method. In both graphs in Figs. 14(a) and 14(b), a normal space is set on the basis of normal data, and MDs (Mahalanobis distance) of normal data and abnormal data are calculated to plot calculation results in a histogram of normal and abnormal data.

[0123] Fig. 14(a) shows calculation results based on the above-mentioned feature value item A (the number of measurement items is 3), and Fig. 14(b) shows calculation result based on the above-mentioned feature value item B (the number of measurement items is 5). In either graph, calculation results based on normal data are shown in a set manner on a relatively leftside, and calculation results based on abnormal data are shown in a set manner on a relatively right side.

[0124] For both Figs. 14(a) and 14(b), it was possible to separate normal data and abnormal data from each other by setting MD = 4 as shown in the drawing. Thus, by setting MD = 4, a purge gas flow rate state can be easily divided into normal and abnormal states.

[0125] Comparing the case of the feature value item Ain Fig. 14(a) with the case of the feature value item B in Fig. 14(b), it is not clearto distinguish a boundary portion between a calculation result based on the normal data and a calculation result based on the abnormal data in the case of the feature value item A in Fig. 14(a). Then, as shown in Fig. 14(a), a mixture (mixed portion) occurs at a boundary portion in which the calculation result based on the normal data and the calculation result based on the abnormal data overlap each other.

[0126] On the other hand, in the case of the feature value item B in Fig. 14(b), there is no mixed portion, and it is clear to distinguish a boundary portion between a calculation result based on normal data and a calculation result based on abnormal data.

[0127] Thus, by increasing the number of feature value items, it is possible to clearly separate the calculation result from the normal data and the calculation result from the abnormal data. By increasing the feature value items, a purge gas flow rate state can be more easily divided into normal and abnormal states.

[0128] «Verification of test data according to MT method»Figs. 15(a) and 15(b) show verification results of test data according to an MT method. In Fig. 15(a), measurement data related to the temperature in the case of the feature value item A (the number of measurement items is 3) is shown in the lower stage, and a Mahalanobis distance (MD = D2) in the case of the feature value item A is shown in the upper stage.

[0129] In Fig. 15(b), measured data related to the temperature in the case of the feature value item B (the number of measurement items is 5) is shown in the lower stage, and a Mahalanobis distance (MD = D2) in the case of the feature value item B is shown in the upper stage.

[0130] Horizontal axes in lower and upper graphs in Fig. 15(a) and lower and upper graphs in Fig. 15(b) all represent a time. In addition, vertical axes in the lower graphs in Figs. 15(a)and 15(b) represent a temperature (unit: °C), and vertical axes in the upper graphs in Figs.15(a) and 15(b) all represent a Mahalanobis distance (MD = D2).

[0131] In the lower graph in Fig. 15(a), waveforms of a motor temperature, a top portion temperature, and a base portion temperature are shown as indicated by arrows in the drawing. The lower graph in Fig. 15(b) shows the waveform of a circuit unit temperature in addition to the motor temperature, the top portion temperature, and the base portion temperature. Here, although the feature value item B (the number of measurement items is 5) in Fig. 15(b) includes current data (data of a motor current), the lower graph in Fig. 15(b) shows changes in temperature, and thus current data is not shown.

[0132] Further, in Figs. 15(a) and 15(b), a purge cut line (purge cut) is shown by a dashed line so as to extend over the upper and lower graphs. The purge cut line shows the timing when the flow rate of a purge gas (purge flow rate) changes from a specified amount to 0 (zero) as shown in the column (cell) where the row of the “test data” and the column of the “purge flow rate” in the table of Fig. 13 cross each other.

[0133] In the upper graphs in Figs. 15(a) and 15(b), the line of MD = 4 is shown by a bold line. As shown in the upper graphs, the Mahalanobis distance (MD) is approximately MD = 1 as a whole and is largely below the line of MD = 4 for normal times before the purge cut.

[0134] However, for abnormal times after the purge cut, the Mahalanobis distance exceeds the line of MD = 4 in approximately several tens of seconds, and the tendency is that it continues to increase thereafter. A deviation of the Mahalanobis distance (MD) between the normal time and the abnormal time reaches approximately several ten times.

[0135] From such verification results, it is possible to distinguish between normal and abnormal times based on a threshold value by determining an appropriate threshold value such as MD = 4. Thus, it is considered that the MT method can be used to create a determination model related to a purge gas flow rate state or to perform determination using the determination model.

[0136] Further, comparing the upper graphs (the graph of the feature value item A and the graph of the feature value item B) in Figs. 15(a) and 15(b), a time required to exceed MD = 4from the purge cut is longer in the case of the feature value item B shown in Fig. 15(b). Further, the Mahalanobis distance (MD) has reached a larger value after the purge cut in the case of the feature value item B. Thus, it can be said that detection sensitivity in the case of the feature value item B is higher than that in the case of the feature value item A.

[0137] «Verification of test data according to logistic regression and deep learning» Lower and upper graphs in Fig. 16(a) show verification results of test data according to logistic regression, and lower and upper graphs in Fig. 16(b) show verification results of test data according to deep learning.

[0138] Horizontal axes in the lower and upper graphs in Figs. 16(a) and 16(b) all represent a time. Vertical axes in the lower graphs in Figs. 16(a) and 16(b) represent a temperature (unit: °C). Furthermore, vertical axes in the upper graphs in Figs. 16(a) and 16(b) show numerical values of 0 to 1 . Among these numerical values, "1 " corresponds to the presence of purge, and "0" corresponds to the absence of purge. A method of distinguishing between the case of the presence of purge and the case of the absence of purge will be described later.

[0139] Verification of test data according to logistic regression is performed under the condition ofthe feature value itemA(the number of measurement items is 3). Forthis reason, in the lowergraph in Fig. 16(a), waveforms of a motor temperature, a top portion temperature, and a base portion temperature are shown as indicated by arrows in the drawing.

[0140] Verification of test data related to deep learning is performed under the condition of the feature value item B (the number of measurement items is 5). For this reason, the lower graph in Fig. 16(b) shows the waveform of a circuit unit temperature in addition to the motor temperature, the top portion temperature, and the base portion temperature. Here, although the feature value item B in Fig. 16(b) includes current data, the lowergraph in Fig. 16(b) shows changes in temperature, and thus current data is not shown.

[0141] Further, in Figs. 16(a) and 16(b), similar to Figs. 15(a) and 15(b) according to the MT method, a purge cut line (purge cut) is shown by a dashed line. Regarding the purge cut line, description ofthe same matters as those in Figs. 15(a) and 15(b) according to the MT method will be omitted.

[0142] In the upper graph in Fig. 16(a) according to logistic regression, the waveform of verification data maintains "1 " for normal times before the purge cut. However, for abnormal times after the purge cut, “0” is indicated beyond the line of 0.5 (indicated by a bold line in the graph). This line of 0.5 is a line with a value (0.5) determined as a threshold value for convenience.

[0143] From such verification results, it is possible to distinguish between normal and abnormal times based on a threshold value by determining an appropriate threshold value such as 0.5. Thus, it is considered that the logistic regression can be used to create a determination model related to a purge gas flow rate state or to perform determination using the determination model.

[0144] Further, in the upper graph in Fig. 16(b) according to deep learning, the waveform of verification data maintains "1 " for normal times before the purge cut. However, for abnormal times after the purge cut, it reaches "0" beyond the line of 0.5 (indicated by a bold line in the graph) with stepwise lowering. This line of 0.5 is a line with a value (0.5) determined as a threshold value for convenience.

[0145] From such verification results, even for deep learning, it is possible to distinguish between normal and abnormal times based on a threshold value by determining an appropriate threshold value such as 0.5. Accordingly, it is considered that the deep learning can be used to create a determination model related to a purge gas flow rate state or to perform determination using the determination model.

[0146] «Notification of abnormality»As shown in Figs. 8(a) and 8(b), the determination unit 234 is provided with an alarm output unit 236. The alarm output unit 236 is a functional portion (functional configuration) configured with a CPU (not shown) and its peripheral apparatuses (storage unit, communication unit, and the like). The alarm output unit 236 outputs a command for outputting an alarm when the determination unit 234 determines that a purge gas flow rate state is inappropriate (it determines that the purge gas flow rate state is abnormal).

[0147] The alarm is output by an alarm output apparatus (not shown). Examples of the alarm output apparatus include an LED, a display device, a speaker, and the like provided in the control device 200. In addition, examples of the alarm output apparatus include a PC (personal computer) communicatively connected to the turbo molecular pump 100 and the control device 200, a portable information terminal device such as a tablet terminal and a smartphone, and the like.

[0148] «Basic effects of invention according to embodiment»As described above, the turbo molecular pump 100 according to the embodiment includes the temperature sensors 222Ato 222C disposed in the first heat exhaust path (arrow E) to the third heat exhaust path (arrow G) different from each other, the determination model creation unit 232, and the determination unit 234. The determination model creation unit 232 creates a determination model for determining a purge gas flow rate state indicating the state of a flow rate of a purge gas on the basis of a detection result of a temperature detection unit. The determination unit 234 determines a purge gas flow rate state using the determination model.

[0149] Thus, the turbo molecular pump 100 itself can determine the flow rate state of the purge gas. Then, it is possible to provide a vacuum pump, a purge gas monitoring device, and a purge gas monitoring system which are capable of more accurately determining the presence or absence of an abnormality depending on a purge gas flow rate.

[0150] It is also conceivable to monitor the temperature of one location to determine the suitability of a flow rate of a purge gas, but it is difficult to make a distinction for matters, such as whether the flow rate of the purge gas is sufficient orwhetherthe temperature of the purge gas is low, unless a plurality of temperatures (particularly the temperatures of a plurality of heat exhaust paths) are monitored comprehensively. In such a sense, it is also important to use temperature data of a plurality of independent heat exhaust paths.

[0151] In addition, the determination model created by the determination model creation unit 232 is created on the basis of a correlation between detection results of the temperature sensors 222Ato 222C and a purge gas flow rate indicating the flow rate of the purge gas whenthe detection results of the temperature sensors 222A to 222C are acquired. Thus, the acquired correlation between the temperature data (detection results) and the purge gas flow rate can be appropriately modeled. The purge gas flow rate state can be determined more appropriately.

[0152] In addition, the determination model is created using at least one of an MT method, logistic regression, and deep learning. Thus, this also makes it possible to more appropriately determine the purge gas flow rate state.

[0153] In addition, since temperatures in the independent heat exhaust paths (first to third heat exhaust paths) are detected, separated clusters (for example, the clusters P1 and P2 in Fig. 7) can be relatively easily formed. Thus, a statistical method and machine learning can be easily introduced. It is considered that the larger the number of independent heat exhaust paths, the easier it is to form separate clusters and the easier it is to improve the accuracy of determination.

[0154] Although the temperature sensor holding member 220 is provided with the temperature sensors 222A and 222B in the above-described embodiment, this is not limiting, and, for example, the temperature sensors 222A and 222B may be directly attached to the stator column 122.

[0155] In addition, since items (for example, a motor current, and the like) otherthan detection results of the temperature sensors are adopted as feature value items and the number of feature value items is increased as much as possible, it is possible to distinguish between normal data and abnormal data more clearly as shown in Fig. 14(b), Fig. 15(b), and Fig. 16(b).

[0156] Although 0.5 is adopted as a threshold value in the example of the logistic regression in Fig. 16(a) or the example of the deep learning in Fig. 16(b), this is not limiting, and a value (for example, 0.2) otherthan 0.5 may be adopted as a threshold value.

[0157] In addition, a purge gas is supplied to the turbo molecular pump 100, for example, at room temperature, and the temperature of the purge gas can be increased depending on the temperature of a contacting part in the turbo molecular pump 100.

[0158] Further, the determination unit 234 includes the alarm output unit 236 that outputs an alarm when it determines that a purge gas flow rate state is inappropriate. Thus, when the purge gas flow rate state is inappropriate, this can be more reliably reported. clnvention that can be extracted from embodimentThe following invention can be extracted from the above-described embodiments. (1) A vacuum pump including:a plurality of temperature detection units (such as the temperature sensors 222Ato 222C) that detect temperatures in heat exhaust paths (such as the first to third heat exhaust paths) through which a purge gas in the vacuum pump (such as the turbo molecular pump 100) flows;a determination model creation unit (such as the determination model creation unit 232) that creates a determination model (such as an MT method model, a logistic regression model, or a deep learning model) for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit (such as the determination unit 234) that determines the purge gas flow rate state using the determination model,in which the temperature detection units are disposed in different heat exhaust paths. (2) The vacuum pump according to (1), in which the determination model is created based on a correlation between the detection results of the temperature detection units and a purge gas flow rate indicating the flow rate of the purge gas when the detection results of the temperature detection units are acquired.(3) The vacuum pump according to (2), in which the determination model is created using any one of an MT method, logistic regression, and deep learning in order to non-linearly model the correlation.(4) The vacuum pump according to (2), in which the temperature detection units are disposed in at least one of a rotor blade portion, a stator portion (such as the stator column 122), and a driving unit (such as the motor 121) of the rotor blade portion.(5) The vacuum pump according to any one of (1) to (4), in which the determination unit includes an alarm output unit (such as the alarm output unit 236) that outputs an alarm when the determination unit determines that the purge gas flow rate state is inappropriate (such as that the purge gas flow rate state is abnormal).(6) A purge gas monitoring device including:a plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows;a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit that determines the purge gas flow rate state using the determination model,in which the temperature detection units are disposed in different heat exhaust paths. (7) A purge gas monitoring system including:a plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows;a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit that determines the purge gas flow rate state using the determination model,in which the temperature detection units are disposed in different heat exhaust paths.

[0159] The present invention is not limited to the above-described embodiments, and various modifications and combinations of the embodiments can be made without departing from the gist.[Reference Signs List]

[0160] 100 Turbo molecular pump101 Inlet port102 Rotor blade103 Rotating body113 Rotor shaft121 Motor122 Stator column123 Stator blade127 Outer cylinder133 Outlet port141 Electronic circuit unit149 Water-cooled pipe200 Control device210 Purge gas monitoring device211 Purge gas port216 Space between rotating body and stator column 220 Temperature sensor holding member 222Ato 222C Temperature sensor222 Temperature sensor holding member224 Top portion226 Bottom portion232 Determination model creation unit234 Determination unit236 Alarm output unitE Arrow indicating first exhaust pathF Arrow indicating second exhaust pathG Arrow indicating third exhaust pathP1 ClusterP2 Cluster

Claims

CLAIMS1. A vacuum pump comprising:a plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows;a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit that determines the purge gas flow rate state using the determination model,wherein the temperature detection units are disposed in different heat exhaust paths.

2. The vacuum pump according to claim 1 , wherein the determination model is created based on a correlation between the detection results of the temperature detection units and a purge gas flow rate indicating the flow rate of the purge gas when the detection results of the temperature detection units are acquired.

3. The vacuum pump according to claim 2, wherein the determination model is created using any one of an MT method, logistic regression, and deep learning in order to non-linearly model the correlation.

4. The vacuum pump according to claim 2, wherein the temperature detection units are disposed in at least one of a rotor blade portion, a stator portion, and a driving unit of the rotor blade portion.

5. The vacuum pump according to any one of claims 1 to 4, wherein the determination unit includes an alarm output unit that outputs an alarm when the determination unit determines that the purge gas flow rate state is inappropriate.

6. A purge gas monitoring device comprising:a plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows;a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit that determines the purge gas flow rate state using the determination model,wherein the temperature detection units are disposed in different heat exhaust paths.

7. A purge gas monitoring system comprising:a plurality of temperature detection units that detect temperatures in heat exhaust paths through which a purge gas in a vacuum pump flows;a determination model creation unit that creates a determination model for determining a purge gas flow rate state indicating a state of a flow rate of the purge gas based on detection results of the temperature detection units; anda determination unit that determines the purge gas flow rate state using the determination model,wherein the temperature detection units are disposed in different heat exhaust paths.

Images