KR20200064197A - Machine learning method for diagnising fault of inverter and inverter for diagnosing fault through learning data - Google Patents

Abstract

The present invention relates to a machine learning method for diagnosing a failure of an inverter, and to an inverter which self-diagnoses a failure through learning data obtained through the corresponding method. According to one embodiment of the present invention, the machine learning method for diagnosing a failure of an inverter comprises the steps of: driving a motor while changing an operating state of the inverter; and generating learning data for each operation state based on an error value which is a difference between an output value and a command value of the inverter, and storing the learning data in a database. The operating state includes a state in which at least one switching element among a plurality of switching elements provided in the inverter is opened and a state in which the gain or offset of a current sensor for measuring the output current of the inverter is changed.

Description

Inverter for diagnosing faults through machine learning methods and learning data for fault diagnosis of inverters TECHNICAL FIELD

The present invention relates to a machine learning method for diagnosing a malfunction of an inverter, and an inverter that self-diagnoses a fault through learning data acquired through the method.

In recent years, inverters and electric motors supplied with power have been used in a wide variety of fields. Since these electric motors are driven by electrical signals, there are various defects that may occur during operation.

If there are some fatal defects in the entire system while the motor is running, the inverter needs to stop the motor immediately. On the other hand, when a non-fatal defect occurs, such as a decrease in torque due to the switch opening state in the motor, the user can handle the defect in a simple way when the user recognizes the defect.

The basic method for diagnosing non-fatal defects is to measure the current or voltage output to the motor while driving the motor under specific conditions, and compare the measured current or voltage with a reference value. However, this method has a problem that requires a very wide look-up table (LUT) to implement the algorithm.

In addition, another method of diagnosing a defect is to diagnose a defect based on a change in electric characteristics of the electric motor. However, this method relies heavily on the parameters of the electric motor, which is changed by various causes, so there is a problem of low accuracy of diagnosis.

An object of the present invention is to provide a method for performing machine learning through supervised learning according to an operation state of an inverter for diagnosing a failure of the inverter.

In addition, an object of the present invention is to provide an inverter that diagnoses a failure by itself using learning data for each failure state obtained by machine learning.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention not mentioned can be understood by the following description, and will be more clearly understood by embodiments of the present invention. In addition, it will be readily appreciated that the objects and advantages of the present invention can be realized by means of the appended claims and combinations thereof.

The machine learning method for diagnosing an inverter failure according to an embodiment of the present invention is a step of driving a motor while changing an operation state of an inverter and learning data for each operation state based on an error value that is a difference between an output value and a command value of the inverter And generating and storing in a database, wherein the operating state is a state in which at least one switching element among the plurality of switching elements provided in the inverter is opened, and a gain of a current sensor for measuring the output current of the inverter. ) Or an offset is changed.

In addition, the inverter according to an embodiment of the present invention converts a DC voltage into an alternating current and outputs it to a motor, a switching control unit for controlling a plurality of switching elements in the power conversion unit according to a command value, and a fault condition. It includes a memory in which learning data is stored, and includes a fault diagnosis unit that identifies a fault condition by comparing an error value between the output value of the power conversion unit and the command value with the learning data, and the fault condition is the power conversion unit. It characterized in that it comprises a state in which the gain or offset of the current sensor for measuring the output current of at least one switching element of the plurality of provided switching element is open and the power converter.

In the machine learning method of the present invention, by performing machine learning through supervised learning according to the operating state of the inverter, driving of the inverter is possible, but learning data for non-fatal defects requiring user action can be secured and secured learning When data is applied to the inverter, the inverter can self-diagnose itself.

In addition, the inverter of the present invention can diagnose the failure by itself using the learning data for each failure state obtained by machine learning, so that the user can quickly and accurately identify the non-fatal defect that is difficult to recognize whether the defect is present.

In addition to the above-described effects, the concrete effects of the present invention will be described together while describing the specific matters for carrying out the invention.

1 is a flowchart illustrating a machine learning method for diagnosing inverter failure according to an embodiment of the present invention.
2 is a view showing a state in which a machine learning server controls an inverter and a motor and receives an output current according to a control result.
FIG. 3 is a view showing a state in which a plurality of switching elements and a motor provided in the inverter shown in FIG. 2 are connected.
4A is a graph showing a plurality of operating points defined by a speed command value and a torque command value.
4B is a graph showing a speed command value and a torque command value of a motor over time when the motor is driven according to the operating point shown in FIG. 4A.
5A and 5B are graphs showing error values over time when the upper and lower arm switching elements of the a phase are open, respectively.
Figure 6 is a graph showing the error value over time when the gain of the current sensor of a phase is 0.5.
7 is a graph showing an error value over time when the offset of the current sensor of a phase is +150[A].
8 is a view showing an inverter according to an embodiment of the present invention.
9 is an internal configuration diagram of the switching control unit shown in FIG. 8.
FIG. 10 is an enlarged graph of the vicinity R where the speed of the motor is zero crossing in FIG. 5A.
11 is a flowchart illustrating an example of a process in which the failure diagnosis unit illustrated in FIG. 8 diagnoses a failure state.

The above-described objects, features, and advantages will be described in detail below with reference to the accompanying drawings, and accordingly, a person skilled in the art to which the present invention pertains can easily implement the technical spirit of the present invention. In the description of the present invention, when it is determined that detailed descriptions of known technologies related to the present invention may unnecessarily obscure the subject matter of the present invention, detailed descriptions will be omitted. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings are used to indicate the same or similar components.

In the following, when a component is described as being "connected", "coupled" or "connected" to another component, the components may be directly connected to or connected to each other, but other components between each component may be " It will be understood that intervening, or that each component may be "connected", "coupled" or "connected" through other components.

The present invention relates to a machine learning method for diagnosing a malfunction of an inverter, and an inverter that self-diagnoses a fault through learning data acquired through the method.

First, a machine learning method according to an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 7.

1 is a flowchart illustrating a machine learning method for diagnosing inverter failure according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a state in which a machine learning server controls an inverter and a motor and is provided with an output current according to a control result, and FIG. 3 is a view in which a plurality of switching elements and a motor provided in the inverter shown in FIG. 2 are connected It is a diagram showing.

4A is a graph showing a plurality of operating points defined by a speed command value and a torque command value, and FIG. 4B is a speed command value and a torque command value of a motor over time when the motor is driven according to the operation point shown in FIG. 4A It is a graph showing.

5A and 5B are graphs showing error values over time when the upper and lower arm switching elements of the a phase are open, respectively.

6 is a graph showing an error value over time when the gain of the current sensor of a phase is 0.5. 7 is a graph showing an error value over time when the offset of the current sensor on the a phase is +150 [A].

Referring to FIG. 1, a machine learning method (hereinafter, a machine learning method) for diagnosing an inverter failure according to an embodiment of the present invention includes changing the operating state of the inverter 10 (S10) and the motor 20. Driving (S20), generating learning data for each operation state based on an error value that is a difference between an output value and a command value of the inverter 10 (S30) and storing the generated learning data in a database (S40) ).

The machine learning method illustrated in FIG. 1 is according to an embodiment, and each step constituting the method is not limited to the embodiment illustrated in FIG. 1, and some steps may be added, changed, or deleted as necessary.

The machine learning method illustrated in FIG. 1 may be performed before the inverter 10 is applied to a specific electronic product. In other words, the machine learning method of the present invention can be performed before the inverter 10 is driven under a specific load condition after the inverter 10 is first produced.

The machine learning method may be performed in the machine learning server 1. The machine learning server 1 may control the inverter 10 and the motor 20 connected to the inverter 10, and generate learning data by receiving an output value according to the control result.

The machine learning server 1 and the inverter 10 and the motor 20 may be connected by wire or wirelessly, and the inverter 10 and the motor 20 may be controlled according to a control signal provided from the machine learning server 1 Can be.

The motor 20 to which the present invention can be applied may be any motor including a stator in which a coil is wound and a rotor rotating by a magnetic field generated in the coil. For example, the motor 20 may be a brushless DC motor (BLDC) or a permanent magnet synchronous motor (PMSM), which is a type of permanent magnet synchronous motor.

The machine learning server 1 may drive the motor 20 while changing the operating state of the inverter 10. In other words, the machine learning server 1 may change the operating state of the inverter 10 and may drive the motor 20 based on the operating state whenever the operating state of the inverter 10 is changed.

Referring to FIG. 2, the machine learning server 1 provides a first control signal Sc1 to the inverter 10 to control operating conditions of the inverter 10, and a second control signal Sc2 to the motor 20. By providing the can control the operating conditions of the motor 20. The machine learning server 1 can change the operating state of the inverter 10 by controlling the operating conditions of the inverter 10 and the motor 20.

The inverter 10 may convert and output the input voltage Vin into a three-phase AC current according to the first control signal Sc1, and the motor 20 receives the three-phase AC current and receives the second control signal Sc2 ) May generate a constant torque.

The inverter 10 may generate different output values according to a command value included in the first control signal Sc1. In addition, the inverter 10 may generate different output values according to the operating conditions of the motor 20 controlled according to the second control signal Sc2.

Here, the command value may include a speed command value, a current command value, a voltage command value, and the like, and the output value may include an output current (ia, ib), an output voltage, and the like. A detailed description of each setpoint will be described later with reference to FIG. 10.

On the other hand, in the present invention, the operating state of the inverter 10 may include a state in which any non-fatal defect that the inverter 10 operates abnormally does not need to be stopped to protect the system. .

For example, the operating state may include a state in which at least one switching element among a plurality of switching elements provided in the inverter 10 is open. In addition, the operating state may include a state in which the gain or offset of the current sensors Sa and Sb for measuring the output currents ia and ib of the inverter 10 is changed.

When the inverter 10 is applied to an electronic product, if any switching element in the inverter 10 is opened regardless of the switching control signal, the inverter 10 may provide an abnormal output value to the motor 20.

On the other hand, if the gain or offset of the current sensors Sa and Sb that measure the output currents ia and ib of the inverter 10 is changed even if the switching elements are operating normally, each switching element is switched according to the output currents ia and ib. The feedback controlled inverter 10 may provide an abnormal output value to the motor 20.

Accordingly, the machine learning server 1 may drive the motor 20 while controlling the open state of the switching element of the inverter 10 and the gain or offset change state of the current sensors Sa and Sb.

Subsequently, the machine learning server 1 may generate learning data for each operation state and store it in a database based on an error value that is a difference between an output value of the inverter 10 and a command value whenever the operation state is changed.

Hereinafter, as illustrated in FIG. 3, it is assumed that the inverter 10 of the present invention includes upper and lower arm switching elements Sha, Sla, Shb, Slb, Shc, and Slc for phases a, b, and c, respectively. The operation process of the machine learning server 1 will be described in detail.

In one example, the machine learning server 1 may drive the motor 20 while sequentially opening a plurality of switching elements Sha, Sla, Shb, Slb, Shc, Slc in the inverter 10 one by one.

More specifically, the machine learning server 1 is a lower arm switching element (Sla) of the a phase, the lower arm switching element (Sla), b and c phase switching elements (Shb, Slb, Shc, Slc) of the a phase By controlling it, the motor 20 can be driven.

Next, the machine learning server 1 controls the a-phase switching arm (Sha), the b-phase and c-phase switching elements (Shb, Slb, Shc, Slc) while the a-phase lower arm switching element (Sla) is opened. By doing so, the motor 20 can be driven.

According to this method, the machine learning server 1 may drive the motor 20 while sequentially opening each of the six switching elements Sha, Sla, Shb, Slb, Shc, and Slc.

According to the above-described change in the operating state, the inverter 10 can provide six different output values to the motor 20, and the machine learning server 1 has a current sensor Sa, which is provided at the output terminal of the inverter 10. Six output values can be provided through Sb).

In another example, the machine learning server 1 may drive the motor 20 while sequentially opening two or more of the switching elements Sha, Sla, Shb, Slb, Shc, and Slc.

More specifically, the machine learning server 1 may sequentially open any two switching elements among six switching elements (Sha, Sla, Shb, Slb, Shc, Slc). For example, the motor 20 may be driven by controlling the switching elements (Shb, Slb, Shc, and Slc) of the b and c phases while the two switching elements (Sha, Sla) of the a phase are opened. Subsequently, the machine learning server 1 controls the a-phase and c-phase switching elements (Sha, Sla, Shc, and Slc) while the two switching elements (Shb, Slb) of the b phase are opened to drive the motor 20. Can be. In addition, the machine learning server 1 controls the a-phase and b-phase switching elements (Sha, Sla, Shb, and Slb) with the two c-phase switching elements (Shc, Slc) open to drive the motor (20). can do.

The two switching elements controlled by the open state may be switching elements of the same phase or switching elements provided in different phases. Accordingly, the machine learning server 1 may drive the motor 20 while sequentially opening the switching elements according to fifteen cases.

In accordance with the above-described operation state change, the inverter 10 can provide fifteen different output values to the motor 20, and the machine learning server 1 is a current sensor Sa provided at the output terminal of the inverter 10 , Sb) can receive fifteen output values.

In addition, it is natural that the machine learning server 1 may sequentially open any three or more switching elements among six switching elements (Sha, Sla, Shb, Slb, Shc, Slc).

The machine learning server 1 may drive the motor 20 while changing the gain of the current sensors Sa and Sb provided at the output terminal of the inverter 10.

To set the operating state for the increased gain, the gain can be set in the range of 1.8 to 2.2. In addition, the gain can be set in the range of 0.3 to 0.7 to set the operating state for the reduced gain.

In one example, the machine learning server 1 may drive the motor 20 while changing the gain of the current sensors Sa and Sb provided on the a-phase or the b-phase to 2.0 or 0.5.

When the gain of the current sensors Sa and Sb is changed to 2.0, the output currents ia and ib of the inverter 10 provided to the motor 20 are doubled when the current sensors Sa and Sb are in a normal state. Can be measured. In addition, when the gain of the current sensors Sa and Sb is changed to 0.5, the output currents ia and ib of the inverter 10 provided to the motor 20 are when the current sensors Sa and Sb are in a normal state. It can be measured in half.

The machine learning server 1 may be provided with output values that are doubled or reduced in half through the current sensors Sa and Sb.

In another example, the machine learning server 1 may drive the motor 20 while changing the offset of the current sensors Sa and Sb provided on the a-phase or the b-phase.

To set the operating state for the increased offset, the offset may be set in the range of +100 [A] to +200 [A]. In addition, in order to set the operating state for the reduced offset, the offset may be set in the range of -200 [A] to -100 [A].

In one example, the machine learning server 1 drives the motor 20 while changing the offset of the current sensors Sa and Sb provided on the a-phase or the b-phase to +150 [A] or -150 [A]. can do.

When the offsets of the current sensors Sa and Sb are changed to +150 [A], the output currents ia and ib of the inverter 10 provided to the motor 20 are the current sensors Sa and Sb. It can be measured 150[A] higher than when. In addition, when the offsets of the current sensors Sa and Sb are changed to -150 [A], the output currents ia and ib of the inverter 10 provided to the motor 20 are normal to the current sensors Sa and Sb. It can be measured 150[A] lower than in the state.

The machine learning server 1 may be provided with an output value to which an offset is applied through current sensors Sa and Sb.

Meanwhile, the process of driving the motor 20 while changing the above-described operation state may be performed according to a plurality of operation points defined by the speed command value ω _r and the torque command value T _e .

More specifically, the machine learning server 1 may set an operation point of the inverter 10 for driving the motor 20 in advance. The operating point may be determined by the speed command value ω _r of the motor 20 and the torque command value T _e of the motor 20, and a speed command value ω _r and a torque command value T _e to determine the operating point. Can be set arbitrarily.

Referring to FIG. 4A, in one example, the machine learning server 1 has eight operating points in a motoring area on the right and 8 in a generating area on the left based on the base speed ω _base . You can set the dog's operating point.

Accordingly, as illustrated in FIG. 4B, the speed command value ω _r of the motor 20 over time may increase and decrease in the motoring area and the generating area. In addition, the torque command value T _e of the motor 20 with time may be increased and decreased in the motoring area by using a point at which the size is 0 as an inflection point, and may decrease and increase in the generating area.

The machine learning server 1 may drive the motor 20 based on a total of 16 operating points according to the arrow sequence shown in FIG. 4A.

In one example, when the machine learning server 1 drives the motors 20 while sequentially opening a plurality of switching elements Sha, Sla, Shb, Slb, Shc, Slc one by one, the machine learning server 1 is The motor 20 can be driven according to 16 operating points while the phase-arm switching element Sha is opened.

In another example, when the machine learning server 1 drives the motor 20 while changing the gains of the current sensors Sa and Sb, the machine learning server 1 of the current sensor Sa provided on a In the state where the gain is changed to 2.0, the motor 20 can be driven according to 16 operating points.

Accordingly, the machine learning server 1 may be provided with output values for each of the 16 operation points in a single operation state.

The number and position of operation points are not limited to those shown in FIG. 4A, and can be freely set in the motoring area and the generating area according to the user's setting.

On the other hand, when the inverter 10 is applied to an electronic product, the inverter 10 may be driven in various environments, and a speed command value ω of the motor 20 for determining an operating point in order to diagnose a failure in various environments _r ) and the torque command value (T _e ) may be preferred as they are subdivided.

The machine learning server 1 may calculate an error value that is a difference between an output value and a command value provided from the current sensors Sa and Sb, and generate learning data for each operation state based on the calculated error value. Accordingly, a plurality of learning data may be generated as many as the number of operational states that can be changed.

When the inverter 10 is in a normal state, the command value and the output value of the inverter 10 may be the same. However, as described above, when the operating state is changed, the output value cannot follow the command value, so a difference may occur between the command value and the output value. The machine learning server 1 may calculate an error value that is a difference between the command value and the output value, and generate learning data based on this.

In other words, the machine learning server 1 may generate learning data using supervised learning among various methods of machine learning.

When learning data is generated, the machine learning server 1 may store the generated learning data in a database in the server to correspond to each operation state.

For example, the learning data for each operation state may be represented as [Table 1] below.

[Operation status] [Learning Data] a phase arm switching element open LD1 a phase lower arm switching element open LD2 b-phase upper arm switching element open LD3 b-phase lower arm switching element open LD4 c-phase upper arm switching element open LD5 c-phase lower arm switching element open LD6 a phase current sensor gain 2.0 LD7 a phase current sensor gain 0.5 LD8 b-phase current sensor gain 2.0 LD9 b-phase current sensor gain 0.5 LD10 a-phase current sensor offset +150[A] LD11 a-phase current sensor offset -150[A] LD12 b-phase current sensor offset +150[A] LD13 b-phase current sensor offset -150[A] LD14

In [Table 1], each operation state and learning data corresponding thereto are shown independently. However, two or more individual operating states shown in [Table 1] may be simultaneously applied to the inverter 10, and learning data corresponding thereto may also be generated.

In calculating the error value for generating the learning data for each of the above-described operating states, the machine learning server 1 may calculate the error value according to a preset sampling period or phase of the motor 20.

More specifically, the machine learning server 1 may be provided with output values from current sensors Sa and Sb according to a preset sampling period. Subsequently, the machine learning server 1 may calculate an error value that is a difference between the provided output value and the command value, and generate learning data based on the calculated error value.

In addition, the machine learning server 1 may be provided with output values from the current sensors Sa and Sb according to the phase of the motor 20. The phase of the motor 20 can be measured or estimated. When the phase of the motor 20 is measured, the machine learning server 1 may be provided with a phase value of the motor 20 through an arbitrary Hall sensor (not shown). Alternatively, when the phase of the motor 20 is estimated, the machine learning server 1 may estimate the phase of the motor 20 based on the output currents ia and ib provided from the current sensors Sa and Sb. .

The method of estimating the phase of the motor 20 based on the output currents ia and ib is performed according to a general method used in the art, so a detailed description thereof will be omitted here.

For example, the machine learning server 1 may be provided with output values from the current sensors Sa and Sb whenever the phase of the motor 20 is changed by π/3. Subsequently, the machine learning server 1 may calculate an error value that is a difference between the provided output value and the command value, and generate learning data based on the calculated error value.

Accordingly, the learning data can be generated based on error values when the phase of the motor 20 is π/3, (2π)/3, π, (4π)/3, (5π)/3 and 2 π. have.

Meanwhile, the learning data may be defined as an arbitrary parameter using an error value. However, hereinafter, the learning data is defined as an error value over time, and the error value is defined as a difference (Verr) between the q-axis output voltage and the command voltage of the rotary coordinate system.

In addition, hereinafter, the error value Verr is composed of first to sixth error values Verr1 to Verr6 calculated according to the phase of the motor 20, and the first to sixth error values Verr1 to Verr6 are motors. The phases of (20) are assumed to be error values when π/3, (2π)/3, π, (4π)/3, (5π)/3 and 2π are respectively.

In FIG. 5A, the first to sixth errors calculated by the machine learning server 1 when the motor 20 is driven according to the operating point illustrated in FIG. 4A in a state where the phase-a-phase switching element Sha is open. Values (Verr1 to Verr6) are shown.

In addition, in FIG. 5B, when the motor 20 is driven according to the operating point shown in FIG. 4A in a state where the lower arm switching element Sla of a is open, the first to the second calculated by the machine learning server 1 6 error values (Verr1 to Verr6) are shown.

Meanwhile, in FIG. 6, when the motor 20 is driven according to the operating point shown in FIG. 4A while the gain of the current sensor Sa provided on the a phase is set to 0.5, it is calculated by the machine learning server 1 The first to sixth error values (Verr1 to Verr6) were shown.

In addition, in FIG. 7, when the motor 20 is driven according to the operating point shown in FIG. 4A while the offset of the current sensor Sa provided on the a is set to +150[A], the machine learning server 1 ) To the first to sixth error values (Verr1 to Verr6).

As shown in FIGS. 5A to 7, error values Verr1 to Verr6 according to each operation state have different values, and accordingly learning data according to each operation state may also be different.

In addition to the operating states described in FIGS. 5A to 7, the machine learning server 1 may store learning data corresponding to each operating state and combinations thereof shown in [Table 1] described above in a database.

The machine learning method of the present invention described above performs machine learning through supervised learning according to the operating state of the inverter 10, so that the drive of the inverter 10 is possible, but the learning data for non-fatal defects requiring user action is provided. When the secured learning data is applied to the inverter 10, there is an effect that the inverter 10 can self-diagnose itself.

Next, an inverter 10 according to an embodiment of the present invention will be described in detail with reference to FIGS. 8 to 11.

8 is a diagram illustrating an inverter according to an embodiment of the present invention, and FIG. 9 is an internal configuration diagram of the switching controller illustrated in FIG. 8.

FIG. 10 is an enlarged graph of the vicinity R where the speed of the motor is zero crossing in FIG. 5A.

11 is a flowchart illustrating an example of a process in which the failure diagnosis unit illustrated in FIG. 8 diagnoses a failure state.

Referring to FIG. 8, the inverter 10 according to an embodiment of the present invention may include a power converter 11, a switching controller 12, a fault diagnosis unit 13 and a current sensor S. The inverter 10 shown in FIG. 8 is according to an embodiment, and components constituting the invention are not limited to the embodiment shown in FIG. 8, and some components may be added, changed, or deleted as necessary. have.

The power converter 11 may convert a DC voltage (V _DC ) into an AC current and output it to the motor.

The power converter 11 may be connected to any voltage source that supplies an input voltage, and the input voltage may be stored as a DC voltage (V _DC ) in the DC link capacitor (C _DC ). The power converter 11 may convert the DC voltage (V _DC ) stored in the DC link capacitor (C _DC ) into AC current, and for this purpose, a plurality of switching elements (Sha, Sla, Shb, Slb, Shc, Slc) It may include. For example, the power converter 11 is a three-phase switching element, and may include upper and lower arm switching elements (Sha, Sla, Shb, Slb, Shc, Slc) corresponding to each phase.

The switching control unit 12 may control a plurality of switching elements Sha, Sla, Shb, Slb, Shc, and Slc in the power conversion unit 11 according to the command value.

Hereinafter, a method of controlling a switching element of the switching control unit 12 will be described in detail with reference to FIG. 9.

Referring to FIG. 9, the switching control unit 12 includes a 3-phase/2-phase axis conversion unit 121, a speed calculation unit 122, a current command generation unit 120a, a voltage command generation unit 120b, and a 2-phase/3 It may include an upper axis conversion unit 126 and a switching control signal output unit 127.

The three-phase/two-phase axis conversion unit 121 displays the a-phase and b-phase output currents i _a and i _b detected by the current sensor S and the c-phase output currents i _c calculated therefrom. It can be converted to the two-phase current (i _α ,i _β ) of the, and the two-phase current (i _α ,i _β ) of the stationary coordinate system to the two-phase current (i _d , i _q ) of the rotary coordinate system.

_r)를 추정하고, 추정된 위치를 미분하여 모터의 속도(

_r)를 연산할 수 있다.The speed calculating unit 122 is based on the output current (i _a , i _b , i _c ) of the power converter 11 (position of the motor (rotor) (

_r ) and differentiate the estimated position to determine the speed of the motor (

_r ).

_r)와 속도 지령치(

)에 기초하여 전류 지령치(i_q ^*)를 생성할 수 있다.On the other hand, the current command generator 120a calculates the speed of the motor (

_r ) and speed setpoint (

), a current command value (i _q ^* ) can be generated.

_r)와 속도 지령치(

)의 차이에 대해 PI 제어기(123)를 통해 PI 제어를 수행할 수 있고, 이에 따라 전류 지령치(i_q ^*)가 생성될 수 있다.For example, the current command generation unit 120a is the speed of the motor (

_r ) and speed setpoint (

), the PI control may be performed through the PI controller 123, and thus a current command value (i _q ^* ) may be generated.

In FIG. 9, only the q-axis current command value (i _q ^* ) is shown to be generated. Alternatively, the d-axis current command value (i _d ^* ) may also be generated, and at this time, the d-axis current command value (i _d ^* ) The value of can be set to 0.

Meanwhile, the current command generation unit 120a may further include a limiter (not shown) that limits the current command value i _q ^* from exceeding a predetermined range.

The voltage command generation unit 120b is generated by the d-axis and q-axis currents (i _d , i _q ), which are axis-converted from the three-phase/two-phase axis conversion unit 121 to a rotation coordinate system, and the current command generation unit 120a. The d-axis and q-axis voltage command values (V _d ^* , V _q ^* ) may be generated based on the generated current command values (i _d ^* , i _q ^* ).

For example, the voltage command generator 120b may perform PI control through the PI controller 124 for the difference between the q-axis current (i _q ) and the q-axis current command value (i _q ^* ), and accordingly The q-axis voltage command value V _q ^* may be generated. In addition, the voltage command generator 120b may perform PI control through the PI controller 125 for the difference between the d-axis current (i _d ) and the d-axis current command value (i _d ^* ), and accordingly the d-axis The voltage command value V _d ^* may be generated.

Meanwhile, the voltage command generation unit 120b may further include a limiter (not shown) that limits the voltage command values V _d ^* and V _q ^* of the d-axis and q-axis to not exceed a predetermined range.

_r)와 d축 q축의 전압 지령치(V_d ^*, V_q ^*)를 이용하여 축변환을 수행할 수 있다.The voltage command values (V _d ^* , V _q ^* ) of the d-axis q-axis generated by the voltage command generation unit 120b may be input to the 2-phase/3-phase axis conversion unit 126, and the 2-phase/3-phase axis conversion The unit 126 is the position of the motor calculated by the speed calculating unit 122 (

_The axis conversion can be performed using the voltage command values (V _d ^* , V _q ^* ) of the _r- axis and the q-axis of the d-axis.

_r)가 사용될 수 있다. 다음으로, 2상/3상 축변환부(126)는 2상 정지 좌표계에서 3상 정지 좌표계로 축변환을 수행하여 3상 전압 지령(V_a ^*, V_b ^*, V_c ^*)을 출력할 수 있다.First, the two-phase/three-phase axis conversion unit 126 may perform axis conversion from a two-phase rotational coordinate system to a two-phase stationary coordinate system, where the motor position (

_r ) can be used. Next, the 2-phase/3-phase axis conversion unit 126 outputs a 3-phase voltage command (V _a ^* , V _b ^* , V _c ^* ) by performing an axis conversion from a 2-phase stationary coordinate system to a 3-phase stationary coordinate system. Can be.

The switching control signal output unit 127 may generate a switching control signal S _C according to a pulse width modulation (PWM) method according to a three-phase voltage command (V _a ^* , V _b ^* , V _c ^* ), The generated switching control signal S _C may be output to the power converter 11.

The switching control signal S _C may be converted into a gate driving signal by a gate driver (not shown), and the gate driving signal may be input to the gate terminal of each switching element in the inverter 10. Accordingly, the switching element may perform a switching operation according to the gate driving signal.

The failure diagnosis unit 13 includes a memory in which learning data about a failure state is stored, and the failure diagnosis unit 13 is a difference between an output value and a command value of the power conversion unit 11 during the above-described operation of the switching control unit 12 The error state can be identified by comparing the error value with the learning data stored in the memory.

), 전류 지령치(i_q ^*, i_d ^*) 및 전압 지령치(V_d ^*, V_q ^*, V_a ^*, V_b ^*, V_c ^*)중 적어도 하나일 수 있다. 한편, 고장상태는 도 1 내지 도 7을 참조하여 설명한 동작상태와 대응되는 것으로서, 인버터(10)가 비정상적으로 동작되는 비치명적인 임의의 결함이 발생한 상태로 정의될 수 있다.Here, the output value may be the output voltage or output current of the power converter 11, and the command value is the speed command value (

), the current command value (i _q ^* , i _d ^* ) and the voltage command value (V _d ^* , V _q ^* , V _a ^* , V _b ^* , V _c ^* ). Meanwhile, the failure state corresponds to an operation state described with reference to FIGS. 1 to 7, and may be defined as a state in which any non-fatal defect in which the inverter 10 operates abnormally occurs.

Accordingly, the fault condition is a state in which at least one switching element among the plurality of switching elements provided in the power converter 11 is opened and the gain or offset of the current sensor for measuring the output current of the power converter 11 is changed. State.

As illustrated in FIGS. 5A to 7, learning data is also defined as an error value over time, and the error value is the q-axis output voltage (V _q ) and the command voltage (voltage command value, V _q ^* ) of the rotation coordinate system. It is defined as the difference of (Verr).

The failure diagnosis unit 13 may calculate an output voltage of the inverter 10 and calculate an error value Verr, which is a difference between the calculated output voltage and the command voltage. Subsequently, the failure diagnosis unit 13 may compare the calculated error value Verr with learning data stored in the memory.

Since the fault state corresponds to the operation state described with reference to FIGS. 1 to 7, the fault state may also be represented as described in [Table 1]. In the memory, learning data for each failure state may be stored, and each learning data may be error value (Verr) data according to time described with reference to FIGS. 5A to 7.

The failure diagnosis unit 13 may compare the previously calculated error value Verr with a plurality of learning data stored in the memory, and determine one learning data corresponding to the error value Verr.

For example, when the inverter 10 of the present invention is applied to any electronic product and driven, when the error value Verr with time appears as shown in FIG. 5A, the failure diagnosis unit 13 is applicable The error value Verr can be compared with a plurality of learning data stored in the memory.

In the memory, an error value (Verr) according to the time shown in FIG. 5A is stored as learning data for the operating state of the'open phase-a switching device', and the failure diagnosis unit 13 learns the data shown in FIG. 5A. It can be determined to correspond to the currently calculated error value (Verr).

Subsequently, the failure diagnosis unit 13 may identify the failure state by referring to the operation state corresponding to the learning data. In other words, the fault diagnosis unit 13 may identify the current fault state of the inverter 10 as the phase-arm switching element of a phase open with reference to the operating state corresponding to the learning data shown in FIG. 5A.

Meanwhile, since the inverter 10 is driven in various environments, some parameters (eg, stator resistance, stator inductance, etc.) in the inverter 10 may be partially changed.

Accordingly, the failure diagnosis unit 13 may determine one learning data most similar to the error value Verr among a plurality of learning data stored in the memory, and identify a failure state corresponding to the determined learning data.

To this end, the failure diagnosis unit 13 may measure similarity, and for this purpose, various regression analysis such as an R ² (r square) method may be performed. The failure diagnosis unit 13 may determine one learning data most similar to the error value Verr through regression analysis, and identify a failure state corresponding to the determined learning data.

Alternatively, the failure diagnosis unit 13 may apply a margin to the learning data, and if an error value (Verr) is included in the learning data range to which the margin is applied, the failure state corresponding to the learning data may be identified. Can be.

For example, when the learning data shown in FIG. 5A is stored in the memory, the failure diagnosis unit 13 may apply a margin of -5% to +5% to the learning data. The learning data may be defined as a range, and the failure diagnosis unit 13 may determine whether an error value (Verr) is included in the range of the learning data, and determine learning data corresponding to the error value (Verr).

Meanwhile, in the memory, learning data for each failure state may be stored for a plurality of operation points. For example, learning data for each operation state for each operation point illustrated in FIG. 4 may be stored in the memory.

The failure diagnosis unit 13 may identify one operation point that matches the current operation state among the plurality of operation points, and compare the learning data corresponding to the identified operation point and an error value (Verr) to identify the failure state. have.

More specifically, the failure diagnosis unit 13 may identify the speed and torque of the motor and compare them with a plurality of operating points stored in the memory. For example, the failure diagnosis unit 13 may identify the speed and torque of the current motor as 2ω _base and T _fw , respectively.

Since the corresponding speed and torque correspond to one of the plurality of operating points shown in FIG. 4, the failure diagnosis unit 13 can identify the failure state by comparing the learning data and the error value (Verr) corresponding to the operating point. have.

In order to increase the accuracy of the above-described failure state identification, the failure diagnosis unit 13 may identify a failure state corresponding to the determined learning data if any one learning data corresponding to the error value Verr is continuously determined more than a reference number of times. Can be.

More specifically, the failure diagnosis unit 13 may calculate an error value Verr according to a preset sampling period or phase of the motor. For example, when the failure diagnosis unit 13 calculates an error value Verr according to a preset sampling cycle, in the first cycle, the failure diagnosis unit 13 identifies learning data corresponding to the error value Verr as LD1. Can be.

Assuming that the above-mentioned reference frequency is 2 times, the failure diagnosis unit 13 does not immediately identify the failure state corresponding to LD1 even if the learning data is identified as LD1 in the first period, and the error value Verr in the second period. The corresponding learning data can be identified again.

If the identified learning data is not LD1 again, the failure diagnosis unit 13 may not identify a failure state corresponding to LD1. On the other hand, if the re-identified learning data is LD1, it corresponds to a case in which the corresponding learning data is determined two or more times in succession, so the failure diagnosis unit 13 can identify the failure state corresponding to LD1.

In addition, in order to increase the accuracy of the fault condition identification, the fault diagnosis unit 13 may compare the error value Verr with the learning data when the motor speed is higher than the reference speed.

Referring to FIG. 10, learning data according to a plurality of operation points illustrated in FIG. 4 may include first to sixth error values (Verr1 to Verr6). At this time, the first to sixth error values (Verr1 to Verr6) may be calculated as 0 in the vicinity R where the speed of the motor is zero-crossed according to the change of the torque command value T _e .

At this point, since the output value of the power converter 11 seems to follow the command value, the faulty inverter 10 can be identified as a normal state. The failure diagnosis unit 13 may compare the error value Verr when the speed of the motor is greater than or equal to the reference speed to the learning data in order not to identify a failure state in the vicinity where the speed of the motor is zero crossing.

Hereinafter, an operation process of the failure diagnosis unit 13 according to an example of the present invention will be described with reference to FIG. 11.

The failure diagnosis unit 13 may receive an output value from the current sensor S and calculate an error value Verr, which is the difference between the output value and the command value (S11). Before comparing the error value Verr with the learning data, the failure diagnosis unit 13 may determine whether the current motor speed is greater than or equal to the reference speed (S12).

If the motor speed is less than the reference speed, the failure diagnosis unit 13 may calculate the error value Verr again (S11), and if the motor speed is greater than the reference speed, the failure diagnosis unit 13 may calculate the error value ( Verr) and learning data (S13).

Subsequently, the failure diagnosis unit 13 may determine any one learning data corresponding to the error value Verr, and count the number of determinations for the corresponding learning data (S14). The failure diagnosis unit 13 can continuously compare the error value Verr and the learning data until the determination number exceeds the reference number (S13), and when the determination number exceeds the reference number, the failure corresponding to the corresponding learning data The state can be identified (S16).

The inverter 10 of the present invention described above can diagnose a failure by itself using learning data for each failure state obtained by machine learning, so that a user can quickly and accurately identify a non-fatal defect that is difficult to recognize whether a defect is present or not. It has an effect.

The above-described present invention, the above-described embodiments and the accompanying drawings because it is possible for a person having ordinary knowledge in the technical field to which the present invention pertains to various substitutions, modifications and changes without departing from the technical spirit of the present invention It is not limited by.

Claims (11)

Driving a motor while changing an operating state of the inverter; And
And generating learning data for each operation state based on an error value that is a difference between the output value of the inverter and a command value, and storing it in a database.
The operation state
It includes a state in which at least one of the plurality of switching elements provided in the inverter is open and the gain or offset of the current sensor for measuring the output current of the inverter is changed.
Machine learning method for inverter fault diagnosis.
According to claim 1,
The step of driving the motor while changing the operating state of the inverter is
And driving the motor while sequentially opening the switching elements one by one.
According to claim 1,
The step of driving the motor while changing the operating state of the inverter is
And driving the motor while sequentially opening two or more switching elements among the plurality of switching elements.
According to claim 1,
The step of driving the motor while changing the operating state of the inverter is
And driving the motor according to a plurality of operating points defined by a speed setpoint and a torque setpoint.
According to claim 1,
The step of generating learning data for each operation state based on an error value that is the difference between the output value and the command value of the inverter and storing it in a database is
And calculating the error value according to a preset sampling period or the phase of the motor.
A power converter for converting a DC voltage into an AC current and outputting it to a motor;
A switching control unit controlling a plurality of switching elements in the power conversion unit according to a command value; And
It includes a memory in which learning data for a fault condition is stored, and includes a fault diagnosis unit for identifying a fault condition by comparing an error value between the output value of the power conversion unit and the command value with the learning data,
The fault condition
And a state in which at least one switching element among the plurality of switching elements provided in the power converter is open and a gain or offset of a current sensor measuring the output current of the power converter is changed.
inverter.
The method of claim 6,
The fault diagnosis unit is an inverter that calculates the output voltage of the inverter and compares the error value, which is the difference between the calculated output voltage and the command voltage, with the learning data.
The method of claim 6,
Learning data for each failure state is stored in the memory,
The failure diagnosis unit determines one learning data most similar to the error value, and an inverter for identifying a failure state corresponding to the determined learning data.
The method of claim 6,
The failure diagnosis unit is an inverter for identifying a failure state corresponding to the determined learning data when any one of the learning data corresponding to the error value is continuously determined more than a reference number.
The method of claim 6,
The memory stores learning data for each failure state for a plurality of operating points,
The failure diagnosis unit identifies an operation point that matches the current operation state among the plurality of operation points, and compares the learning data corresponding to the identified operation point and the error value to identify the failure state.
The method of claim 6,
The fault diagnosis unit is an inverter that compares the error value with the learning data when the speed of the motor is higher than a reference speed.

Images