Diagnostic system and method for switch open faults

Abstract

A system for diagnosing open-circuit failures in switches using an artificial neural network model is disclosed. [Solution] In the system for diagnosing open switch failures, the data input to the input layer of the artificial neural network model includes the DC component and RMS value of the three-phase current.

Description

Technical Field

[0001] The present invention relates to a system and method for diagnosing an open failure of a switch.

Background Art

[0002] When an open failure of a switch occurs, distortion occurs in the current flowing through the inverter and converter, and its form and pattern vary depending on the switch in which the failure has occurred. In order to diagnose an open failure, it is necessary to classify the current distortion pattern caused by the failed switch, but if a human directly performs this, there is a possibility of an error occurring in the classification process (Human Error).

[0003] In Korean Patent Publication No. 10-2021-0142883 of the prior art, a step of obtaining a DC component (IV_dc) of d-q axis current in a stationary coordinate system and estimating a total harmonic distortion (THD) (THDv) of the current, applying the DC component (IV_dc) of the d-q axis current to a pre-trained first artificial neural network (Artificial Neural Networks, ANN), and grouping types of open failures according to the current vector angle in a plurality of sectors, and applying the DC component (IV_dc) of the d-q axis current and the total harmonic distortion (THDv) of the current to a pre-trained second artificial neural network to diagnose a switch in which an open failure has occurred within the plurality of grouped sectors are included, and a method for diagnosing a multi-switch open failure of a three-phase PWM converter using artificial intelligence is disclosed.

[0004] However, the method of using two artificial neural networks and using the DC component of the current and the total harmonic distortion as input values has the disadvantages of requiring a high-specification computer and a long learning period.

[0005] <Prior Art Document> <Patent Document> Korean Patent Publication No. 10-2021-0142883

Summary of the Invention

[0006] The present invention aims to provide a system and method for diagnosing open-circuit failures in switches using an artificial neural network. [Means for solving the problem]

[0007] To achieve the above objective, in one embodiment of the present invention, a system for diagnosing open-circuit faults of a switch using an artificial neural network model is provided, wherein the data input to the input layer of the artificial neural network model includes the DC component and RMS value of the three-phase current.

[0008] The aforementioned switch open fault diagnostic system can diagnose a switch open fault during regenerative motor operation.

[0009] The number of neurons in the input layer is 6, and the number of neurons in the output layer of the artificial neural network model may be "number of switches attempting to diagnose open faults + 1".

[0010]

number

[0011] The above formula can be used to calculate the DC component of the three-phase current.

[0012]

number

[0013] The above formula can be used to calculate the effective value of the three-phase current.

[0014] In another embodiment of the present invention for achieving the above object, in a method for diagnosing an open fault of a switch by an artificial neural network model, it includes a step of processing and preprocessing the data input to the input layer of the artificial neural network model, and the data input to the input layer of the artificial neural network model includes the DC component and the effective value of three-phase current, and a method for diagnosing an open fault of a switch is provided.

[0015] The method for diagnosing an open fault of a switch further includes a step of measuring and storing the operation data of a motor to generate the data input to the input layer, and the operation data can be measured during the regenerative operation of the motor.

[0016] The method for diagnosing an open fault of a switch further includes a step of designing the artificial neural network model, the number of neurons in the input layer is 6, and the number of neurons in the output layer of the artificial neural network model can be "the number of switches to be diagnosed for open faults + 1".

[0017]

Number

[0018] The above formula can be used to calculate the DC component of the three-phase current.

[0019]

Number

[0020] The above formula can be used to calculate the effective value of the three-phase current.

Advantages of the Invention

[0021] According to the present invention, an artificial neural network model can be learned even by a relatively low-specification computer, and the learning time can also be shortened.

[0022] According to the present invention, by applying a simplified mathematical formula to calculate the DC component and the root mean square (RMS) value, the calculation can be further reduced.

[0023] According to the present invention, during the regenerative operation of the motor, the open failure of the switch is diagnosed, and the artificial neural network learns and distinguishes the current pattern caused by the faulty switch for each faulty switch, so that the presence or absence of failure of all switches can be diagnosed.

[0024] According to the present invention, since the input values of the input layer are sufficiently characteristic to distinguish various failure modes that are the output values of the output layer, there is an advantage that the depth (number of layers) of the hidden layer and / or the number of neurons in the hidden layer can be effectively reduced.

Brief Description of the Drawings

[0025] [Figure 1] It is a block diagram that briefly shows an example of a motor drive circuit used in an elevator. [Figure 2] An example of the fault current pattern of the switch included in the converter section during the motoring of the motor is shown. [Figure 3] An example of the fault current pattern of the switch included in the converter section during the regenerative operation of the motor is shown. [Figure 4] The structure of an artificial neural network model according to an embodiment of the present invention is shown. [Figure 5] It is a schematic flowchart of a method for diagnosing a switch open failure according to an embodiment of the present invention.

Modes for Carrying Out the Invention

[0026] Hereinafter, the configuration and operation of the preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The present invention can be variously applied and applied in the fields of switch fault diagnosis, elevator fault diagnosis, and artificial neural networks. Note that the following examples can be modified into various other forms, and the scope of the present invention is not limited by the following examples.

[0027] Figure 1 is a simple block diagram showing an example of a motor drive circuit used in an elevator.

[0028] Referring to Figure 1, the motor drive circuit used in the elevator may include one or more of the following: grid section 110, filter section 120, converter section 130, connection section 140, inverter section 150, and motor section 160.

[0029] The grid section 110 supplies power to the circuit and may be equipped with an AC power supply. The grid section 110 may be equipped with a three-phase AC power supply.

[0030] The filter section 120 plays a role in smoothly adjusting the power factor of the power supplied from the grid section 110 in the converter section 130. The filter section 120 may include one or more inductors. The filter section 120 may include three inductors connected in series to each power source of the three-phase AC power supply.

[0031] The converter unit 130 adjusts the active and reactive power of the power supplied from the filter unit 120. The converter unit 130 may be equipped with multiple switches. The switches may be power semiconductor switches or IGBTs (Insulated Gate Bipolar Transistors).

[0032] The converter section 130 may include multiple switches. The converter according to this embodiment is a 3-level PWM converter and may include 12 switches Ta1, Ta2, Ta3, Ta4, Tb1, Tb2, Tb3, Tb4, Tc1, Tc2, Tc3, Tc4.

[0033] Ta1 and Ta4 are connected in series, and the point between the two switches is called the "pole." Tb1 and Tb4, and Tc1 and Tc4 are connected in the same way. Each of (Ta1, Ta4), (Tb1, Tb4), and (Tc1, Tc4) can be connected in parallel with each other. Ta2 and Ta3 are connected in series, and are connected between the pole between Ta1 and Ta4 and the Y point of the connection. Tb2 and Tb3, and Tc2 and Tc3 can also be connected in the same way.

[0034] When the grid unit 110 supplies a three-phase AC power supply, the pole between Ta1 and Ta4 is connected to the first power supply Va, the pole between Tb1 and Tb4 is connected to the second power supply Vb, and the pole between Tc1 and Tc4 is connected to the third power supply Vc.

[0035] Ta1, Tb1, and Tc1 can each be connected in parallel to the first node X of the connection unit 140. (Ta2+Ta3), (Tb2+Tb3), and (Tc2+Tc3) can each be connected in parallel to the second node Y of the connection unit 140. Ta4, Tb4, and Tc4 can each be connected in parallel to the third node Z of the connection unit 140.

[0036] This type of switch arrangement can reduce voltage and current harmonics and improve efficiency.

[0037] The converter unit 130 can adjust the power supplied to the connection unit 140 and operate so that the power factor is 1 (high-efficiency operation).

[0038] The connection section 140 connects the converter section 130 and the inverter section 150. The connection section 140 converts DC current to AC current, or AC current to DC current. The connection section 140 may include multiple (e.g., two) capacitors (DC-Link) connected in series or parallel. The connection section 140 can store the three-phase AC power supplied from the grid section 110 and passed through the filter section 120 and the converter section 130 as DC power.

[0039] The first node X, connected in parallel to Ta1, Tb1, and Tc1 of the converter section 130, is connected to the + terminal of the first capacitor VDC1. The second node Y, connected in parallel to (Ta2+Ta3), (Tb2+Tb3), and (Tc2+Tc3) of the converter section 130, may be connected to the - terminal of the first capacitor VDC1 and the + terminal of the second capacitor VDC2. The third node Z, connected in parallel to Ta4, Tb4, and Tc4 of the converter section 130, may be connected to the - terminal of the second capacitor VDC2.

[0040] The inverter unit 150 converts the current that has passed through the connection unit 140 to a format suitable for driving the motor M. The inverter unit 150 varies the frequency and size of the current converted to DC at the connection unit 140 and converts it into the AC current required to drive the motor M.

[0041] The inverter unit 150 may be equipped with multiple switches. The switches may be power semiconductor switches or IGBTs (Insulated Gate Bipolar Transistors).

[0042] The inverter unit 150 may be equipped with 12 switches Ta1, Ta2, Ta3, Ta4, Tb1, Tb2, Tb3, Tb4, Tc1, Tc2, Tc3, Tc4.

[0043] Ta1 and Ta4 are connected in series, with the point between the two switches designated as the pole. Tb1 and Tb4, and Tc1 and Tc4 are connected in the same manner. Each of (Ta1, Ta4), (Tb1, Tb4), and (Tc1, Tc4) can be connected in parallel with each other. Ta2 and Ta3 are connected in series, with the connection between the pole between Ta1 and Ta4 and the Y point of the connection. Tb2 and Tb3, and Tc2 and Tc3 can be connected in the same manner.

[0044] Ta1, Tb1, and Tc1 can each be connected in parallel to the first node X of the connection section 140. (Ta3+Ta2), (Tb3+Tb2), and (Tc3+Tc2) can each be connected in parallel to the second node Y of the connection section 140. Ta4, Tb4, and Tc4 can each be connected in parallel to the third node Z of the connection section 140.

[0045] By arranging the switches in this way, it is possible to reduce harmonics in voltage and current and improve efficiency.

[0046] The motor unit 160 includes a motor M. The motor M is driven by a current supplied by the inverter unit 150 that has been converted to a state suitable for driving the motor M.

[0047] An open switch failure restricts the current flow through the faulty switch, causing distortion in the current flowing through the circuit. This can lead to a decrease in the power factor and efficiency of the converter section 130, and a decrease in the control performance of the motor M in the motor section 160.

[0048] If an open circuit failure occurs in one of the switches in the inverter unit 150, vibration and noise may be generated in the motor M. Therefore, it is possible to externally confirm that an open circuit failure has occurred in one or more of the switches in the inverter unit 150.

[0049] However, if an open circuit failure occurs in the switch provided in the converter unit 130, it is difficult to detect the failure externally, and operation may continue with the failure unaddressed. This can lead to a decrease in operating efficiency and the risk of secondary failures such as burnout and short circuits due to overheating. Therefore, there is a need for technology to diagnose open circuit failures in the switch provided in the converter unit 130, which are difficult to detect externally.

[0050] Figure 2 shows an example of a fault current pattern of a switch included in the converter section 130 during motoring of motor M. Figure 3 shows an example of a fault current pattern of a switch included in the converter section 130 during regenerative operation of motor M. In Figures 2 and 3, (a) shows the fault current pattern of Ta1, (b) shows the fault current pattern of Ta2, (c) shows the fault current pattern of Ta3, and (d) shows the fault current pattern of Ta4.

[0051] Referring to Figure 2, it can be seen that when motor M is motoring, current distortion occurs in Ta2 and Ta3 of the converter section 130, but no current distortion occurs in Ta1 and Ta4 of the converter section 130.

[0052] Referring to Figure 1, when motor M is motoring, Ta2, Ta3, Tb3, Tb3, Tc2, Tc3 of the converter section 130 are power semiconductors ( JPEG2026513711000006.jpg8170

[0053] ) and diode ( JPEG2026513711000007.jpg10170

[0054] Although the motor is operated using both diodes, the Ta1, Tb1, Tc1, Ta4, Tb4, and Tc4 of the converter section 130 are operated only via diodes. Therefore, even if an open-circuit fault occurs in the Ta1, Tb1, Tc1, Ta4, Tb4, and Tc4 of the converter section 130, steady-state operation is possible and no current distortion occurs. Consequently, in the motoring of the motor M, it is possible to diagnose only open-circuit faults in the Ta2, Ta3, Tb2, Tb3, Tc2, and Tc3 of the converter section 130, and it is impossible to diagnose open-circuit faults in the Ta1, Tb1, Tc1, Ta4, Tb4, and Tc4 of the converter section 130.

[0055] On the other hand, referring to Figure 3, it can be seen that during regenerative operation of motor M, current distortion occurs in all of the converters Ta1, Ta2, Ta3, and Ta4 of the converter section 130.

[0056] Referring to Figure 1, during regenerative operation of motor M, all switches Ta1, Tb1, Tc1, Ta2, Tb2, Tc2, Ta3, Tb3, Tc3, Ta4, Tb4, and Tc4 of the converter unit 130 are operated using both power semiconductors and diodes. Therefore, when an open circuit failure occurs in any of the switches, a current distortion occurs. Thus, by diagnosing open circuit failures in all switches through the current distortion, it is possible to diagnose open circuit failures in the switches during regenerative operation of motor M.

[0057] Figure 4 shows the structure of an artificial neural network model according to one embodiment of the present invention.

[0058] Referring to Figures 1 and 4, the artificial neural network model applicable to the present invention includes an input layer I, a hidden layer H, and an output layer O.

[0059] Input layer I is the layer into which data for training the artificial neural network model is input, and it has one or more neurons. When a three-phase current is supplied from the grid section 110, the number of neurons in input layer I can be six (three DC components of the three-phase current + three RMS values ​​of the three-phase current).

[0060] The hidden layer H is a layer that processes the data input to the input layer I to generate output data, and it has one or more neurons. The number of neurons in the hidden layer H can be determined to the optimal number through iterative learning, taking into account computation time, accuracy, etc.

[0061] The output layer O is the layer that outputs the output data generated in the hidden layer H, and comprises one or more neurons. When the target of diagnosis is a switch included in the converter unit 130, the number of neurons in the output layer O can be (number of switches included in the converter unit 130 + 1) (for each failure mode of the switch + when all switches are in a steady state).

[0062] Since the current distortion pattern changes depending on the switch that fails, there are as many failure modes as there are switches. If the converter section 130 has 12 switches, there will be 12 failure modes in the converter section 130. Therefore, the number of neurons in the output layer O will be 13. The design of the number of neurons in the input layer I, output layer O, and hidden layer H will be described later.

[0063] Figure 5 is a schematic flowchart of a method for diagnosing a switch open fault according to one embodiment of the present invention.

[0064] Referring to Figure 5, the method for diagnosing a switch open fault in the present invention may include one or more of the following steps: step S110 of measuring and storing operating data; step S120 of processing and pre-processing the data; step S130 of designing an artificial neural network model; step S140 of the artificial neural network model learning the data; step S150 of designing the algorithm of the artificial neural network model; step S160 of determining whether the optimization of the artificial neural network model is sufficient; and step S170 of applying the artificial neural network model to diagnose a switch open fault.

[0065] 1) Step S110: Measure and store the operating data. To generate data to be input to the input layer I of the artificial neural network, operating data of the system to which the motor M and switch are applied (e.g., an elevator) is measured and stored. The operating data may include three-phase AC current values ​​supplied by the three-phase AC power supply of the grid unit 110.

[0066] 2) Step S120: Process and pre-process the data. The stored operating data is processed and pre-treated to make it suitable for input to the input layer I of an artificial neural network. The steps of processing and pre-treating the data may include calculating the DC component and RMS value of the three-phase AC current values.

[0067] The following formula can be used to calculate the DC component of a three-phase current.

[0068]

number

[0069] When applying an artificial neural network model to diagnose open-circuit faults in a switch, the relative magnitudes of the values ​​obtained from the formula across different fault conditions may be more important than the magnitudes of the values ​​themselves. The relative magnitudes remain unchanged even if the division operation is omitted, and since the division operation places a heavy load on the computer, the DC component of the three-phase current can be calculated more simply by omitting the division operation as shown below.

[0070]

number

[0071] This invention has the advantage of reducing the computational load on a computer by using a formula that excludes division operations to calculate the DC component of a three-phase current, allowing even relatively low-spec computers to train artificial neural network models and shortening training time.

[0072] The following formula can be used to calculate the effective value (RMS) of the three-phase current.

[0073]

number

[0074] Similarly, the relative magnitudes remain unchanged even if division and square root operations are omitted. Since division and square root operations place a heavy load on the computer, the effective value (RMS) of the three-phase current can be calculated more easily by omitting them, as shown below.

[0075]

number

[0076] This invention has the advantage of reducing the computational load on a computer by using a formula that excludes division and square root operations to calculate the effective value (RMS) of a three-phase current, allowing even relatively low-spec computers to train artificial neural network models and shortening training time.

[0077] 3) Step S130: Design an artificial neural network model. An artificial neural network model is designed by determining the types of input data and the number of neurons in the input layer I, and the types of output data and the number of neurons in the output layer O.

[0078] The number of neurons in input layer I can be determined by the number of inputs used for diagnosis, and the number of neurons in output layer O can be determined by the number of faults to be diagnosed.

[0079] When a three-phase current is supplied from the grid section 110, the number of neurons in the input layer I can be 6 (3 DC components of the three-phase current + 3 RMS values ​​of the three-phase current). When the target of diagnosis is a switch included in the converter section 130, the number of neurons in the output layer O can be (number of switches included in the converter section 130 + 1) (each failure mode of the switch + when all switches are in a steady state).

[0080] 4) Step S140: The artificial neural network model learns from the data. An artificial neural network model learns from the input data input to the input layer I and generates output data.

[0081] Learning is the process of determining the weights W that exist between layers. The weights W act as synapses between layers and determine the transfer ratio of the values ​​output from the activation function. The weights W are optimized in a direction that minimizes the error between the predicted value and the correct answer of the artificial neural network model.

[0082] The Back Propagation Algorithm can be used to optimize the weight (W) values. The error can be calculated using the Cross Entropy Error function.

[0083] The number of layers and neurons in the hidden layer H can initially be set using approximate values ​​(empirical values). Then, weights are determined through learning, and the diagnostic performance (error, cross-entropy error) is checked. If the diagnostic performance decreases, the number of neurons in the hidden layer H is increased to improve it. Conversely, if the diagnostic performance is excellent when learning with the initially set number of hidden layers H and neurons, the computational load can be reduced while maintaining performance by decreasing the number of hidden layers and neurons. Through this iterative learning process, the optimal number of layers, neurons, and weights for the hidden layer H that satisfy both diagnostic performance and computational load can be determined. Based on the results of this iterative learning, the optimal number of neurons in the hidden layer H can be determined.

[0084] According to the present invention, the input value (DC component, RMS value) of input layer I is sufficiently distinctive to distinguish between multiple failure modes (each failure mode of the switch + the case where all switches are in a steady state) which are output values ​​of output layer O. This has the advantage of effectively reducing the depth (number of layers) and / or the number of neurons in hidden layer H. In the present invention, the depth of hidden layer H may be 1 layer and the number of neurons in the hidden layer may be 5, which is a very simple configuration compared to typical artificial neural networks.

[0085] 5) Step S150: Design the algorithm for the artificial neural network model. By applying an artificial neural network model, we can design algorithms necessary for diagnosing open switch failures. As an example, we will use C-language based program code for algorithm design.

[0086] 6) Step S160 to determine whether the artificial neural network model has been sufficiently optimized, and Step S170 to apply the artificial neural network model to diagnose an open fault in the switch. The system determines whether the artificial neural network model is sufficiently optimized to be suitable for diagnosing open switch failures. If the system determines that the artificial neural network model is sufficiently optimized (YES in S160), it applies the artificial neural network model to diagnose open switch failures (S170). If the system determines that the artificial neural network model is not sufficiently optimized (NO in S160), it repeats steps 1) to 5) to optimize the artificial neural network model.

[0087] The computer specifications and training time required for an artificial neural network model vary greatly depending on the type of input data and the number of neurons in the input layer I, as well as the type of output data and the number of neurons. To actually apply an artificial neural network model, realistic computer specifications and training time are necessary. However, if a model can obtain more efficient and appropriate output data even with a low-spec computer and limited training time, it can be considered a more well-designed artificial neural network model.

[0088] According to the present invention, when a three-phase AC power supply is supplied from the grid 110, the number of hidden layers H in the artificial neural network model and the number of neurons in the input layer I and hidden layer H can be dramatically reduced by using DC components and RMS values ​​as input data for the artificial neural network model. By reducing the number of hidden layers H and the number of neurons in the input layer I and hidden layer H of the artificial neural network model, the artificial neural network model can be trained even on a relatively low-spec computer, and the training time can also be shortened.

[0089] According to the present invention, the number of calculations can be further reduced by applying a simplified formula to calculate the DC component and the effective value RMS.

[0090] According to the present invention, switch open faults are diagnosed during regenerative operation, not during motoring of motor M. The artificial neural network learns and distinguishes the current pattern caused by each faulty switch, thus enabling the diagnosis of faults in all switches.

[0091] It will be obvious to those skilled in the art that the present invention is not limited to the embodiments described above, and that various modifications or variations can be made without exceeding the technical essence of the invention. [Explanation of Symbols]

[0092] 110: Grid section 120: Filter section 130: Converter section 140: Connection part 150: Inverter section 160: Motor section I: Input layer H: Hidden layer O: Output layer

Claims

1. In a system that diagnoses open-circuit failures of switches using an artificial neural network model, The data input to the input layer of the aforementioned artificial neural network model includes the DC component and RMS value of the three-phase current. A diagnostic system for detecting open switch malfunctions.

2. To diagnose a switch open fault during motor regenerative braking, A diagnostic system for switch open faults according to claim 1.

3. The number of neurons in the input layer is 6. The number of neurons in the output layer of the aforementioned artificial neural network model is "the number of switches attempting to diagnose open faults + 1". A diagnostic system for switch open faults according to claim 1 or claim 2.

4. To calculate the DC component of the three-phase current, [Math 1] Use the formula A diagnostic system for switch open faults according to claim 1 or claim 2.

5. To calculate the effective value of the three-phase current, [Math 2] Using the formula, A diagnostic system for switch open faults according to claim 1 or claim 2.

6. In a method for diagnosing open-circuit failures in switches using an artificial neural network model, The step includes processing and pre-processing the data input to the input layer of the artificial neural network model, The data input to the input layer of the aforementioned artificial neural network model includes the DC component and RMS value of the three-phase current. How to diagnose a switch open malfunction.

7. The method further includes the step of measuring and storing motor operating data in order to generate data to be input to the input layer, The aforementioned operating data is measured during the regenerative operation of the motor. The method for diagnosing a switch open fault according to claim 6.

8. The process further includes the step of designing the aforementioned artificial neural network model, The number of neurons in the input layer is 6. The number of neurons in the output layer of the aforementioned artificial neural network model is "the number of switches attempting to diagnose open faults + 1". A method for diagnosing a switch open fault according to claim 6 or claim 7.

9. To calculate the DC component of the three-phase current [Math 3] Use the formula A method for diagnosing a switch open fault according to claim 6 or claim 7.

10. To calculate the effective value of the three-phase current [Math 4] Using the formula, A method for diagnosing a switch open fault according to claim 6 or claim 7.

Images