PREDICTION OF FAULTY BEHAVIOUR OF A CONVERTER BASED ON TEMPERATURE ESTIMATION WITH MACHINE LEARNING ALGORITHM

Abstract

Disclosed herein is a method for predicting a faulty behaviour of an electrical converter. The method includes receiving an operation point indicator of the electrical converter indicative of an actual operation point of the electrical converter, where the electrical converter is connected to a rotating electrical machine; receiving a measured device temperature of a power semiconductor device of the electrical converter indicative of an actual temperature of the power semiconductor device; inputting the operation point indicator as input data into a machine learning algorithm trained with historical data comprising operation point indicators and associated device temperatures, where the historical data was recorded during normal operation of a power semiconductor device; estimating an estimated device temperature with the machine learning algorithm, where the estimated device temperature represents a device temperature during a normal operation; and predicting the faulty behaviour by comparing the estimated device temperature with the measured device temperature.

Claims

1. A method for predicting a faulty behaviour of an electrical converter, the method comprising: receiving an operation point indicator (I) of the electrical converter indicative of an actual operation point of the electrical converter, wherein the electrical converter is connected to a rotating electrical machine for driving the rotating electrical machine and the operation point indicator (I) comprises at least one of: a torque of the rotating electrical machine and a rotational speed of the rotating electrical machine; receiving a measured device temperature (T.sub.d) of a power semiconductor device of the electrical converter indicative of an actual temperature of the power semiconductor device; inputting the operation point indicator (I) as input data into a machine learning algorithm trained with historical data comprising operation point indicators and associated device temperatures, wherein the historical data was recorded during normal operation of a power semiconductor device; estimating an estimated device temperature (T*.sub.d) with the machine learning algorithm, wherein the estimated device temperature (T*.sub.d) represents a device temperature during a normal operation; and predicting the faulty behaviour (F) by comparing the estimated device temperature (T*.sub.d) with the measured device temperature (T.sub.d).

2. The method of claim 1, wherein the input data of the machine learning algorithm comprises the measured device temperature (T.sub.d(t−1)) of a previous time step and the operation point indicator (I(t)) of an actual time step; and wherein the machine learning algorithm estimates the estimated device temperature (T*.sub.d(t)) for the actual time step.

3. The method of claim 1, wherein the input data comprises measured device temperatures (T.sub.d(t−1), T.sub.d(t−2), . . . ) for a number of previous time steps; and/or wherein the input data comprises operation point indicators (I(t−1), I(t−2), . . . ) for a number of previous time steps; and/or wherein the input data comprises one or more differences of the measured device temperatures (T.sub.d(t)−T.sub.d(t−1), T.sub.d(t−1)−T.sub.d(t−2), . . . ) for a number of previous time steps.

4. The method of claim 1, further comprising: receiving an ambient temperature (T.sub.a) of the electrical converter, the ambient temperature (T.sub.a) being indicative of an ambient temperature of the power semiconductor device and/or the electrical converter; and/or wherein the input data of the machine learning algorithm further comprises the ambient temperature (T.sub.a).

5. The method of claim 4, wherein the input data of the machine learning algorithm further comprises the measured device temperature (T.sub.d): wherein the measured device temperature (T.sub.d) in the input data is provided relative to the ambient temperature (T.sub.a); and wherein the estimated device temperature (T*.sub.d), which is output by the machine learning algorithm, is provided relative to the ambient temperature (T.sub.a).

6. The method of claim 1, further comprising: determining a temperature error (E) from a difference of the measured device temperature (T.sub.d) and the estimated device temperature (T*.sub.d); and comparing the temperature error (E) with a threshold for predicting the faulty behaviour (F).

7. The method of claim 1, wherein the machine learning algorithm has been trained with the historical data of the same converter; and/or wherein the machine learning algorithm has been trained with the historical data of at least one different converter.

8. The method of claim 1, wherein the machine learning algorithm is an artificial neuronal network.

9. The method of claim 1, wherein the operation point indicator (I) comprises at least one of: a converter current, a converter voltage, a switching frequency, and a DC link voltage.

10. The method of claim 1, wherein the power semiconductor device is a power semiconductor switch; and/or wherein the power semiconductor device is an IGBT.

11. A computer program, which, when being executed by a processor, is adapted for performing the method of claim 1.

12. A non-transitory computer-readable medium in which a computer program according to claim 11 is stored.

13. A controller of an electrical converter adapted for performing the method of claim 1.

14. An electrical converter, comprising: a plurality of power semiconductor devices; at least one temperature sensor arranged for measuring a device temperature (T.sub.d) of at least one of the power semiconductor devices; and a controller according to claim 13, adapted for estimating the estimated device temperature (T*.sub.d) of the at least one of the power semiconductor devices and for predicting the faulty behaviour (F).

15. (canceled)

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The subject-matter of the present disclosure will be explained in more detail in the following text with reference to exemplary embodiments which are illustrated in the attached drawings.

[0053] FIG. 1 schematically shows a drive system with an electrical converter according to an embodiment of the present disclosure.

[0054] FIG. 2 shows a block diagram illustrating a method and a controller according to an embodiment of the present disclosure.

[0055] FIG. 3 shows a block diagram illustrating a method and a controller according to a further embodiment of the present disclosure.

[0056] FIGS. 4A and 4B show diagrams with estimated and measured temperature of a power semiconductor device.

[0057] FIGS. 5A, 5B, 6A, and 6B show diagrams with temperature errors as determined with a method according to an embodiment of the present disclosure.

[0058] FIG. 7 schematically shows a structure of a neural network trained to perform a temperature estimation in a method according to an embodiment of the present disclosure.

[0059] The reference symbols used in the drawings, and their meanings, are listed in summary form in the list of reference symbols. In principle, identical parts are provided with the same reference symbols in the figures.

DETAILED DESCRIPTION

[0060] FIG. 1 shows a drive system 10, which includes an electrical converter 12 and a rotating electrical machine 14, such as a motor or generator. The electrical converter 12 is connected to an electrical grid 16 and converts the grid voltage into an output voltage supplied to the rotating electrical machine 14.

[0061] As schematically shown in FIG. 1, the converter 12 includes power semiconductor devices 18, which may be semiconductor switches and in particular IGBTs. The free-wheeling diodes connected in parallel to the semiconductor switches also may be considered as power semiconductor devices.

[0062] As shown, the power semiconductor devices 18 may be connected in series to form half-bridges 20, which may be connected to a DC link 22 and/or may provide a phase output between them. The power semiconductor devices 18 of one half-bridge may be assembled into a power semiconductor module 24.

[0063] For each power semiconductor device 18, a temperature sensor 26 may be present, which also may be assembled into the respective power semiconductor module 24. However, it is also possible that solely one temperature sensor 26 is present per power semiconductor module 24. With the temperature sensor 26, an actual device temperature of an associated power semiconductor device 18 can be measured.

[0064] The drive system 10 also includes a controller 28, which is adapted for controlling the converter 12 and, as indicated, the power semiconductor switches. To this end, gate signals for the power semiconductor switches may be generated by the controller 28. As indicated in FIG. 1, the controller 28 therefore may receive measurement signals from current and/or voltages in the converter 12, such as an input voltage, DC link voltage, output voltage, output current, etc.

[0065] Besides the control of the power and/or torque of the drive system 10, the controller 28 is also adapted for predicting faulty behaviour of the converter 12 and/or its power semiconductor devices 18 based on temperature measurements and estimations.

[0066] For this, the controller 28 is also adapted for receiving measurement signals from the temperature sensors 26. The converter 12 also may include an ambient temperature sensor 30, which may be provided in a housing of the converter 12. The controller 28 also may be adapted for receiving measurement signals from the ambient temperature sensor 30.

[0067] FIGS. 2 and 3 show diagrams for parts of the controller 28, which may perform the prediction of the faulty behaviour. With respect to FIGS. 2 and 3, also a method for predicting a faulty behaviour of the electrical converter 12 is described.

[0068] The method and the controller 28 are based on a machine learning algorithm, which may be executed in module 32 shown in FIGS. 2 and 3.

[0069] As shown in FIG. 2, an operation point indicator I of the electrical converter 12, which is indicative of an actual operation point of the electrical converter 12, is input as input data into the machine learning algorithm 32. The operation point indicator I may include at least one of a converter current, a converter voltage, a switching frequency, a DC link voltage, a torque of the rotating electrical machine 14, a rotational speed of the rotating electrical machine 14, etc.

[0070] In general, the operation point indicator I may be any quantity positively correlated with the device temperature.

[0071] The operation point indicator I may be provided and/or determined by other control functions of the controller 28 and/or may be based on current and/or voltage measurement in the converter 12.

[0072] From the input data, the machine learning algorithm 32 estimates an estimated device temperature T*.sub.d. In general, the machine learning algorithm 32 may be a function with which the estimated device temperature T*.sub.d can be calculated from the input data. The function may include parameters or weights, which have been adjusted during a training phase of the machine learning algorithm 32.

[0073] In particular, the machine learning algorithm 32 has been trained with historical data including operation point indicators I and associated device temperatures T.sub.d, which have been recorded for the same converter 12 and/or from different converters, which may have been of the same type and/or topology. With the aid of the historical data, the parameters of weight of the machine learning algorithm 32 may be adjusted, such that the function outputs similar estimated device temperatures T*.sub.d for similar input data.

[0074] For example, the historical data may be recorded at the beginning of lifetime of the power semiconductor devices 18 and may be used up to the end of the lifetime of these devices 18.

[0075] For example, the machine learning algorithm 32 may be an artificial neuronal network. It has been shown that in a simple artificial neuronal network with an input layer, solely one hidden layer and an output layer is enough to estimate the temperature with high accuracy.

[0076] FIG. 4A shows the measured device temperature T.sub.d of an IGBT, which is part of a converter 12, and the estimated device temperature T*.sub.d produced by such an artificial neuronal network, which has been trained accordingly. The temperature is depicted on the vertical axis in ° C. FIG. 4A has been produced for a set of sample time steps, depicted to the right, at the beginning of the lifetime of the IGBT.

[0077] FIG. 4B is a similar diagram as FIG. 4A, however with the measured device temperature T.sub.d and the estimated device temperature T*.sub.d at the end of the lifetime of the IGBT. As the artificial neural network was trained with normal data, it cannot correctly predict the elevated temperatures at the end of the lifetime. The higher measured device temperatures T.sub.d can be clearly seen in FIG. 4B. The difference between estimation and measurement is clearly visible, showing that the temperature behaviour has changed when compared to training data, which was accumulated at the beginning of the lifetime. This error may be monitored, for example with a moving error index.

[0078] Returning to FIG. 2, the input data of the machine learning algorithm 32 may include the measured device temperature T.sub.d(t−1) of the previous sampling and/or measurement time step t−1 and the operation point indicator I(t−1) of the previous time step t−1. It also may be that the operation point indicator I(t) of the actual time step t is additionally or alternatively included into the input data. The machine learning algorithm 32 then may estimate the estimated device temperature T*.sub.d(t) for the actual time step t.

[0079] By subtracting the actual measured device temperature T.sub.d(t) at the actual time step t from the actual estimated device temperature T*.sub.d(t), a temperature error E(t) at the actual time step t may be determined, which is input into an averaging and/or comparator module 34.

[0080] The averaging and/or comparator module 34 may average the temperature error E over a time horizon of a plurality of time steps. The result may be compared with a threshold and, if the averaged error is higher than the threshold, the faulty behaviour signal F may be changed to 1. Otherwise, the faulty behaviour signal F may be 0.

[0081] It has to be noted that alternatively, a median error E(t) may be determined by the averaging and/or comparator module 34 from the measured device temperature T.sub.d at the end from the estimated device temperature T*.sub.d.

[0082] In FIG. 3 it is shown that the input data for the machine learning algorithm 32 may include operation point indicators I(t−1), I(t−2), . . . for a number of previous time steps t−1, t−2, etc. By using more than one previous operation point indicator I, the accuracy of the machine learning algorithm 32 may be enhanced. It has to be noted that in FIG. 3, also the operation point indicator I(t) of the actual time step t may be included into the input data.

[0083] FIG. 3 also shows that the measured device temperature T.sub.d(t−1) of the previous time step may be part of the input data. It also may be that the input data for the machine learning algorithm 32 includes measured device temperatures T.sub.d(t−1), T.sub.d(t−2), . . . for a number of previous time steps t−1, t−2, etc. Again, this may enhance the prediction accuracy.

[0084] When an ambient temperature T.sub.a is measured, also the ambient temperature T.sub.a(t−1) of the previous time steps and optionally for a number of previous time steps may be included into the input data.

[0085] It also may be that the input data includes differences of the measured device temperatures (T.sub.d(t)−T.sub.d(t−1), T.sub.d(t−1)−T.sub.d(t−2), . . . ) for a number of previous time steps t−1, t−2, . . . . As shown in FIG. 3, the estimation performed by the machine leaning algorithm 32 may be performed relative to the ambient temperature T.sub.a. The ambient temperature T.sub.a may be subtracted from the measured device temperature T.sub.d, before it is input into the machine learning algorithm 32. The machine learning algorithm then estimates the device temperature relative to the ambient temperature T.sub.a, i.e. outputs T*.sub.d−T.sub.a. It has to be noted that then for determining the temperature error E, also the relative measured temperature T.sub.d−T.sub.a has to be subtracted.

[0086] FIGS. 5A and 5B show the moving average Ē of the temperature error E at the beginning and the end of the lifecycle of an IGBT, respectively. Note that the error, which is depicted on the vertical axis, is shown to the value of 2 in FIG. 5A and to the value of 12 in FIG. 5B. The sampling time steps are shown on the horizontal axis.

[0087] FIGS. 6A and 6B are analogously diagrams, however with the median error {tilde over (E)}.

[0088] As can be seen from FIGS. 5A and 5B or from FIGS. 6A and 6B, a threshold may be used for the temperature error E, Ē, {tilde over (E)} to trigger the fault behaviour signal F and to generate a warning about a change in temperature behaviour.

[0089] As can be seen, the temperature error E, Ē, {tilde over (E)} is smaller than 2 at the beginning of the lifecycle. In the end of the lifecycle, it is consistently higher than 4 and may reach values higher than 10 briefly. After the temperature error E, Ē, {tilde over (E)} has been over the threshold for a certain amount of time, the faulty behaviour signal may be triggered and/or a warning about elevated temperatures in a power semiconductor module 24 may be generated. This may be done, even though no overtemperature fault has been triggered up so far. Once a warning has been triggered, troubleshooting may begin to find causes for the faulty behaviour, potentially preventing sudden breakdown or malfunction in the future.

[0090] The following table includes a sample of testing results output by a neural network for implementing the machine learning algorithm 32. The testing was done with Azure Machine Learning Studio, as it provided easy access to stored drive data as well as to suitable machine learning libraries. In this example, data sets of torque values “Torq15” to “Torq20”, indicative of a torque of the electrical machine 14, are used as input data for training the neural network to estimate an IGBT case temperature. The column “Scored Labels” includes the corresponding output data of the neural network, i.e. the estimated values given by the neural network. The column “T.sub.C” includes measured temperature values for each of the estimated values. As can be seen, there are only minor differences between the estimated and the measured values. Alternatively or additionally to the torque, rotational speed of the electrical machine 14 or ambient temperature may be used as input data for the neural network.

TABLE-US-00001 Scored Torq15 Torq16 Torq17 Torq18 Torq19 Torq20 T.sub.C Labels 11.264338 8.273571 9.914753 7.426168 8.828226 8.279785 50.569656 49.605255 30.690624 29.400188 30.600868 31.13504 30.264393 32.505112 47.373535 46.880524 3.79306 2.82437 3.396286 3.914578 3.396516 1.162935 44.121147 44.138519 4.797654 4.65082 4.638852 4.673374 4.664399 4.6812 45.169472 46.162163 33.21373 23.269291 17.212269 10.196453 39.360283 48.53212 47.277687 51.551109 10.339146 8.930412 11.256973 22.446514 29.334827 41.16832 48.744202 48.193192 1.630135 1.568225 1.602057 1.678696 1.641412 1.582724 43.903473 43.640587

[0091] FIG. 7 schematically shows a structure of the above mentioned neural network. The neural network has an input layer 36 with multiple torque input neurons T for inputting the torque values and multiple rotational speed input neurons n for inputting rotational speed values and an output layer 37 with one output neuron for outputting the corresponding estimated value of the temperature. In this example, the input layer 36 and the output layer 37 are interconnected via one hidden layer 38 and ten previous samples from each signal have been used for estimating the temperature.

[0092] While embodiments of the present disclosure have been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The present disclosure is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art from a study of the drawings, the disclosure, and the appended claims. In the claims, the word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality. A single processor or controller or other unit may fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

LIST OF REFERENCE SYMBOLS

[0093] 10 drive system [0094] 12 electrical converter [0095] 14 rotating electrical machine [0096] 16 electrical grid [0097] 18 power semiconductor device [0098] 20 half-bridge [0099] 22 DC link [0100] 24 power semiconductor module [0101] 26 device temperature sensor [0102] 28 controller [0103] 30 ambient temperature sensor 30 [0104] 32 machine learning algorithm/module [0105] 34 averaging and/or comparator module [0106] 36 input layer [0107] 37 output layer [0108] 38 hidden layer [0109] n rotational speed input neuron [0110] t actual time step [0111] t−1, t−2 previous time steps [0112] I operation point indicator [0113] T torque input neuron [0114] T.sub.d measured device temperature [0115] T*.sub.d estimated device temperature [0116] T.sub.a ambient temperature [0117] E error [0118] F fault signal [0119] Ē moving average error [0120] {tilde over (E)} median error