Power conversion device and manufacturing method

Abstract

A power conversion device (1) comprises: a first switch (AH); a second switch (AL); a third switch (BH); a fourth switch (BL); an isolation transformer (T1); a resonant capacitor (Cr) and a resonant inductor (Lr); and a control unit (10) that controls the switching of the first switch (AH), the second switch (AL), the third switch (BH), and the fourth switch (BL), using a trained model that outputs an output parameter when receiving an input parameter, or using table data that has the function of the trained model.

Description

Power conversion device and manufacturing method

[0001] The present disclosure relates to a power conversion device and a method for manufacturing a power conversion device.

[0002] Patent Document 1 describes a technology in which the gain is first adjusted by adjusting only the frequency, and if the gain does not reach the target value at the maximum frequency, the gain is adjusted to the target value by adjusting the phase at the maximum frequency.

[0003] Non-Patent Document 1 describes a technique for adjusting the gain to a target value by fixing the frequency to a value according to the output voltage and determining the phase from the output current.

[0004] JP 2016-63745 A

[0005] J.-H. Kim, et al, "Analysis on Load Adaptive Phase-Shift Control for High Efficiency Full-Bridge LLC Resonant Converter in Light Load Conditions", IEEE Transactions on Power Electronics, 29 July 2015, pp. 4942 - 4955, vol.31, No.7, IEEE.

[0006] There are countless combinations of parameters (e.g., frequency and phase) for controlling the switch to achieve a specific output power and gain. In the technology described in Patent Document 1, the switch is controlled by the parameter combination that results in the maximum frequency among these countless combinations, so the efficiency of the power conversion device is not taken into consideration and high efficiency may not be achieved. In the technology described in Non-Patent Document 1, the frequency is determined only from the output voltage and the phase is determined only from the output current, so if the range of the power or gain to be controlled is wide, it may not be possible to satisfy both the desired power and gain.

[0007] Therefore, the present disclosure provides a power conversion device and the like that can improve efficiency while realizing a desired gain.

[0008] a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node on the first path between the first switch and the second switch and a second node on the second path between the third switch and the fourth switch; a resonant capacitor and a resonant inductor connected between the first node and the primary winding or between the second node and the primary winding; and a control unit that controls switching of the first switch, the second switch, the third switch, and the fourth switch using a trained model that outputs an output parameter when an input parameter is input, or table data having the function of the trained model.

[0009] A manufacturing method according to the present disclosure is a method for manufacturing a power conversion device, the power conversion device including: a first switch provided on a first path connecting a first input terminal and a second input terminal; a second switch provided on the first path and connected in series with the first switch; a third switch provided on a second path different from the first path connecting the first input terminal and the second input terminal; a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node on the first path between the first switch and the second switch and a second node on the second path between the third switch and the fourth switch; and a resonant capacitor and a resonant inductor connected between the first node and the primary winding or between the second node and the primary winding. and a control unit that controls switching of the first switch, the second switch, the third switch, and the fourth switch, and the manufacturing method includes the steps of: generating, using pre-prepared parameters, a trained model that outputs output parameters for controlling switching of the first switch, the second switch, the third switch, and the fourth switch so that when input parameters are input, the efficiency of the power conversion device becomes equal to or greater than a predetermined value; and incorporating the trained model into the power conversion device, or generating table data from the trained model that has the functions of the trained model, and incorporating the table data into the power conversion device.

[0010] These comprehensive or specific aspects may be realized as a system, a method, an integrated circuit, a computer program, or a computer-readable recording medium such as a CD-ROM, or may be realized as any combination of a system, a method, an integrated circuit, a computer program, and a recording medium.

[0011] According to a power conversion device according to an aspect of the present disclosure, it is possible to improve efficiency while realizing a desired gain.

[0012] 1 is a circuit configuration diagram showing an example of a power conversion device according to an embodiment; FIG. 2 is a diagram showing an example of training data used when generating a trained model; FIG. 3 is an example of a neural network for collecting combinations of efficiencies equal to or greater than a predetermined value and output parameters; FIG. 4 is a diagram showing an example of collected combinations; FIG. 5 is a diagram for explaining the efficiency of a power conversion device to which the technology of Patent Document 1 is applied; FIG. 6 is a diagram for explaining the efficiency of a power conversion device to which the technology of Non-Patent Document 1 is applied; FIG. 7 is a diagram for explaining the efficiency of a power conversion device according to an embodiment; FIG. 8 is a flowchart showing an example of a method for manufacturing a power conversion device according to another embodiment; FIG. 9 is a flowchart showing a specific example of a method for manufacturing a power conversion device according to another embodiment;

[0013] Hereinafter, the embodiments will be specifically described with reference to the drawings.

[0014] The embodiments described below are all comprehensive or specific examples, and the numerical values, shapes, materials, components, arrangement and connection of the components, steps, and order of steps shown in the following embodiments are merely examples and are not intended to limit the present disclosure.

[0015] (Embodiment) Hereinafter, a power conversion device according to an embodiment will be described.

[0016] FIG. 1 is a circuit configuration diagram showing an example of a power conversion device 1 according to an embodiment.

[0017] The power conversion device 1 is an isolated DC-DC converter that boosts or bucks an input voltage to a predetermined output voltage. For example, the power conversion device 1 is an LLC converter that utilizes LLC resonance due to the leakage inductance, excitation inductance, and resonant capacitor of a transformer. The LLC converter performs frequency control to change the switching frequency of each switch on the primary side, and phase shift control to change the phase difference between the switching of each switch on the primary side, thereby changing the input / output voltage ratio (gain), and thereby enabling the desired output power.

[0018] The power conversion device 1 has terminals t1, t2, t3, and t4. Terminal t1 is an example of a first input terminal. Terminal t2 is an example of a second input terminal, specifically a ground terminal. Terminal t3 is an output terminal, and terminal t4 is a ground terminal. Note that, since the power conversion device 1 is an isolated DC-DC converter, terminals t2 and t4 are electrically isolated. An input voltage and an input current are input to terminal t1. The input voltage is the voltage between terminals t1 and t2. An output voltage and an output current are output from terminal t3. The output voltage is the voltage between terminals t3 and t4.

[0019] For example, the power conversion device 1 is mounted on a vehicle and applied to an electric vehicle system that drives auxiliary equipment. For example, a high-voltage lithium-ion battery or the like is connected to terminals t1 and t2, and a low-voltage lead-acid battery and auxiliary equipment are connected to terminals t3 and t4. For example, the voltage of a lithium-ion battery is 250 V to 450 V, and the voltage of a lead-acid battery is 10 V to 16 V. To convert the high voltage of 250 V to 450 V to a low voltage of 10 V to 16 V, the power conversion device 1, such as an LLC converter that supports a wide input / output voltage ratio, is used.

[0020] The power conversion device 1 includes switches AH, AL, BH, BL, CH, CL, DH, and DL, a transformer T1, a capacitor Cr, an inductor Lr, a control unit 10, and a primary gate drive circuit 20.

[0021] The switch AH is a switch provided on the path P1 connecting the terminal t1 and the terminal t2. The switch AH is an example of a first switch. The path P1 is an example of a first path. The switch AH is, for example, an N-channel metal oxide semiconductor field effect transistor (MOSFET). The drain of the switch AH is connected to the terminal t1, and the source of the switch AH is connected to the drain of the switch AL.

[0022] The switch AL is provided on the path P1 and is connected in series with the switch AH. The switch AL is an example of a second switch. The switch AL is, for example, an N-channel MOSFET. The drain of the switch AL is connected to the source of the switch AH, and the source of the switch AL is connected to the terminal t2.

[0023] The switch BH is a switch provided on a path P2 that connects the terminal t1 and the terminal t2 and is different from the path P1. The switch BH is an example of a third switch. The path P2 is an example of a second path. The switch BH is, for example, an N-channel MOSFET. The drain of the switch BH is connected to the terminal t1, and the source of the switch BH is connected to the drain of the switch BL.

[0024] The switch BL is provided on the path P2 and is connected in series with the switch BH. The switch BL is an example of a fourth switch. The switch BL is, for example, an N-channel MOSFET. The drain of the switch BL is connected to the source of the switch BH, and the source of the switch BL is connected to the terminal t2.

[0025] The transformer T1 is an example of an isolation transformer, and has a primary winding and a secondary winding that are insulated from each other.

[0026] The primary winding of the transformer T1 is connected between a node N1 on the path P1 between the switch AH and the switch AL, and a node N2 on the path P2 between the switch BH and the switch BL. The node N1 is an example of a first node, and the node N2 is an example of a second node.

[0027] The capacitor Cr and the inductor Lr are connected between the node N1 and the primary winding of the transformer T1 or between the node N2 and the primary winding of the transformer T1. The capacitor Cr is an example of a resonant capacitor, and the inductor Lr is an example of a resonant inductor. For example, the capacitor Cr is connected to the node N1, the inductor Lr is connected to the node N1 via the capacitor Cr, and the capacitor Cr and the inductor Lr are connected in series. The inductor Lr may be provided as a separate inductor, or may be provided by utilizing the leakage inductance of the transformer T1.

[0028] The capacitor Cr and the inductor Lr may be connected to the node N2. In this case, one end of the primary winding of the transformer T1 is connected to the node N1, and the other end of the primary winding of the transformer T1 is connected to the node N2 via the capacitor Cr and the inductor Lr. Although the example in which the capacitor Cr is connected to the transformer T1 via the inductor Lr has been shown, the inductor Lr may be connected to the transformer T1 via the capacitor Cr. For example, the capacitor Cr and the inductor Lr shown in FIG. 1 may be interchanged. However, if the inductor Lr is provided by utilizing the leakage inductance of the transformer T1, the inductor Lr is placed closer to the transformer T1 than the capacitor Cr (in other words, the capacitor Cr is connected to the transformer T1 via the inductor Lr).

[0029] Alternatively, the capacitor Cr may be connected to the node N1 and the inductor Lr may be connected to the node N2, or the capacitor Cr may be connected to the node N2 and the inductor Lr may be connected to the node N1. In other words, the primary winding of the transformer T1 may be connected between the capacitor Cr and the inductor Lr.

[0030] In this way, the order in which the capacitor Cr, the inductor Lr, and the primary winding of the transformer T1 are connected between the node N1 and the node N2 is not particularly limited.

[0031] In FIG. 1, the excitation inductance of the transformer T1 is indicated by an inductor Lm.

[0032] The switch CH is provided on the path P3 connecting the terminal t3 and the terminal t4. The switch CH is, for example, an N-channel MOSFET. The drain of the switch CH is connected to the terminal t3, and the source of the switch CH is connected to the drain of the switch CL.

[0033] The switch CL is provided on the path P3 and is connected in series with the switch CH. The switch CL is, for example, an N-channel MOSFET. The drain of the switch CL is connected to the source of the switch CH, and the source of the switch CL is connected to the terminal t4.

[0034] The switch DH is a switch provided on a path P4 that connects the terminal t3 and the terminal t4 and is different from the path P3. The switch DH is, for example, an N-channel MOSFET. The drain of the switch DH is connected to the terminal t3, and the source of the switch DH is connected to the drain of the switch DL.

[0035] The switch DL is provided on the path P4 and is connected in series with the switch DH. The switch DL is, for example, an N-channel MOSFET. The drain of the switch DL is connected to the source of the switch DH, and the source of the switch DL is connected to the terminal t4.

[0036] The secondary winding of the transformer T1 is connected between a node N3 on the path P3 between the switch CH and the switch CL, and a node N4 on the path P4 between the switch DH and the switch DL.

[0037] 1 shows the parasitic capacitance of each switch, and each parasitic capacitance is connected in parallel to the corresponding switch in the equivalent circuit. Also, FIG. 1 shows the body diode of each switch, and each body diode is connected in parallel to the corresponding switch in the equivalent circuit. Specifically, the anode of each body diode is connected to the source of the corresponding switch in the equivalent circuit, and the cathode is connected to the drain of the corresponding switch.

[0038] The secondary circuit of the power conversion device 1 is not limited to a full-bridge circuit configured with switches CH, CL, DH, and DL as long as it has a rectifying function. For example, the secondary circuit of the power conversion device 1 may be a center-tap circuit, may include diodes instead of transistors, or may have multiple switches or diodes connected in parallel.

[0039] The control unit 10 controls the switching of the switches AH, AL, BH, and BL. The control unit 10 also controls the switching of the switches CH, CL, DH, and DL. For example, the control unit 10 performs synchronous rectification control by controlling the switching of the switches CH, CL, DH, and DL. For example, the control unit 10 controls the switching of the switches AH, AL, BH, BL, CH, CL, DH, and DL by controlling a primary-side gate drive circuit 20 connected to the gates of the switches AH, AL, BH, and BL, and a secondary-side gate drive circuit (not shown) connected to the gates of the switches CH, CL, DH, and DL via a PWM generator (not shown) or the like.

[0040] The control unit 10 is realized by, for example, a computer including a processor (microprocessor) and a memory. The memory may be a read-only memory (ROM) or a random access memory (RAM), and can store programs executed by the processor. For example, the control unit 10 is realized by a microcontroller.

[0041] The primary-side gate drive circuit 20 drives the gates of the switches AH, AL, BH, and BL using output parameters output from the control unit 10. For example, as shown in FIG. 1 , when the control unit 10 outputs the switching frequencies of the switches AH, AL, BH, and BL and the phase difference between the switching of the switches AH and AL and the switching of the switches BH and BL as output parameters, the primary-side gate drive circuit 20 drives the gates of the switches AH, AL, BH, and BL so that the switches AH, AL, BH, and BL operate in accordance with these output parameters. Note that while FIG. 1 shows gate drive signals for the switches AH and BL, it does not show gate drive signals for the switches BH and AL. The switch AL is in an off state when the switch AH is in an on state and in an on state when the switch AH is in an off state. The switch BH is in an off state when the switch BL is in an on state and in an on state when the switch AL is in an off state. A dead time may also be provided.

[0042] The control unit 10 controls the switching of the switches AH, AL, BH, and BL using a trained model that outputs an output parameter when an input parameter is input, or table data having the function of the trained model. The trained model is a machine learning model trained by machine learning. The trained model or the table data is incorporated into the power conversion device 1 and stored in a memory included in the control unit 10, a buffer included in the power conversion device 1, or the like.

[0043] Output parameters for controlling the switching of the switches AH, AL, BH, and BL when a desired gain is achieved and efficiency is improved are prepared in advance, and machine learning is performed using these as training data, thereby generating a trained model that outputs output parameters that can improve efficiency while achieving the desired gain. Since the trained model has the function of outputting output parameters in response to input parameters, table data showing the correspondence between input parameters input to the trained model and output parameters output in response to the input parameters can be generated to generate table data having the function of the trained model. Therefore, by controlling the switching of the switches AH, AL, BH, and BL using such a trained model or table data having the function of such a trained model, it is possible to improve efficiency while achieving the desired gain.

[0044] For example, the trained model is a neural network in which the efficiency of the power conversion device 1 is set as an objective function. That is, the trained model is a model that outputs output parameters that make the efficiency equal to or greater than a predetermined value when input parameters are input. For example, the input parameters include parameters detected in a circuit that constitutes the power conversion device 1, and the control unit 10 controls the switching of the switches AH, AL, BH, and BL using two or more output parameters obtained by inputting two or more input parameters into the trained model or table data.

[0045] For example, the two or more types of input parameters include two or more types of parameters of the input voltage, output voltage, input current, and output current of the power conversion device 1. For example, the two or more types of input parameters include either the input voltage and output voltage of the power conversion device 1, or the input current and output current of the power conversion device 1. There is a correlation between the input current or the output current and the output resistance, and the output resistance corresponds to the input current or the output current. Below, an example will be described in which the two or more types of input parameters include the input voltage, the output voltage, and the output current (output resistance) as shown in FIG. 1.

[0046] 1, the two or more types of output parameters include the switching frequencies of the switches AH, AL, BH, and BL, and the phase difference between the switching of the switches AH and AL and the switching of the switches BH and BL. By using the switching frequencies and the phase difference for gate control of each switch, not only can the efficiency be improved but also a wider range of the power conversion device 1 can be achieved (i.e., it can handle a wide range of input and output voltages).

[0047] Here, a method for generating a trained model will be described with reference to FIGS. 2 to 4.

[0048] FIG. 2 is a diagram illustrating an example of training data used when generating a trained model.

[0049] The trained model is generated by training with supervised data. Specifically, a trained model is generated using pre-prepared parameters, which, when input parameters are input, output parameters for controlling the switching of the switches AH, AL, BH, and BL such that the efficiency of the power conversion device 1 is equal to or greater than a predetermined value. The predetermined value is not particularly limited. For example, the predetermined value is set to a value such as 96% depending on the input voltage, output voltage, and output resistance of the power conversion device 1.

[0050] 2 shows an example of pre-prepared parameters serving as training data. For example, the input power, output power, output voltage, and efficiency are collected for the following conditions: the switching frequency is varied from 70 kHz to 250 kHz in 10 kHz increments; the input voltage is varied from 240 V, 300 V, 350 V, 400 V, and 450 V; the load resistance (output resistance) is varied from 80 mΩ to 200 mΩ in 20 mΩ increments; and the phase shift angle (phase difference) is varied from 0° to 180° in 15° increments. FIG. 2 shows 8,645 combinations of these values ​​as an example of training data. For example, the training data may be values ​​obtained from simulation results, evaluation results of an actual device, or a combination of simulation results and evaluation results of an actual device.

[0051] For example, using such training data (prepared parameters), first, a model (for example, a neural network) is generated to collect combinations of efficiency equal to or greater than a predetermined value and output parameters.

[0052] FIG. 3 shows an example of a neural network for collecting combinations of efficiencies and output parameters that are equal to or greater than a predetermined value.

[0053] For example, using the training data described above, a neural network is generated that outputs a phase shift angle (phase difference) and efficiency when a switching frequency, an input voltage, an output voltage, and an output resistance are input, as shown in FIG. 3 . By using this neural network, various combinations of efficiency, input voltage, output voltage, output resistance, switching frequency, and phase difference can be collected. Note that the input layer parameter group and output layer parameter group in the neural network shown in FIG. 3 are merely examples. For example, the input layer parameter group may be the switching frequency, input voltage, output resistance, and phase difference, and the output layer parameter group may be the output voltage and efficiency.

[0054] Fig. 4 is a diagram showing an example of the collected combinations. Fig. 4 shows the highest efficiency points (white dots shown in Fig. 4) for various input voltages, output voltages, output resistances, switching frequencies, and phase differences as collected combinations. For example, calculations are performed using the machine-learned neural network shown in Fig. 3, and combinations of switching frequencies and phase differences with efficiencies equal to or higher than a predetermined value (e.g., the highest efficiency) are collected from the vast number of combinations for various input voltages, output voltages, and output resistances.

[0055] Then, a trained model is generated using the collected combinations. For example, by using the combination of input voltage, output voltage, output resistance, switching frequency, and phase difference that results in the highest efficiency as training data, a trained model is generated that outputs a switching frequency and phase difference that achieves the highest efficiency when an input voltage, output voltage, and output current (output resistance) are input.

[0056] For example, the trained model outputs output parameters that are equal to or greater than a predetermined lower limit and equal to or less than a predetermined upper limit. For example, when generating the trained model, the output parameters used as training data can be set to be parameters that are equal to or greater than a predetermined lower limit and equal to or less than a predetermined upper limit, thereby making it possible to make the output parameters output from the trained model parameters that are equal to or greater than a predetermined lower limit and equal to or less than a predetermined upper limit. Alternatively, by setting a condition in the trained model that the output parameters output from the trained model be equal to or greater than a predetermined lower limit and equal to or less than a predetermined upper limit, it is possible to make the output parameters output from the trained model parameters that are equal to or greater than a predetermined lower limit and equal to or less than a predetermined upper limit.

[0057] If the output parameters do not have lower and upper limits, the number of combinations of output parameters that improve efficiency (e.g., combinations of switching frequency and phase difference) becomes enormous, resulting in a long calculation time. In contrast, by setting lower and upper limits for the output parameters, it is possible to quickly obtain output parameters that can improve efficiency while realizing a desired gain.

[0058] Next, the efficiency of a power conversion device to which the conventional technology is applied will be described with reference to FIGS. 5A and 5B, and the efficiency of the power conversion device 1 according to the embodiment will be described with reference to FIG.

[0059] Fig. 5A is a diagram illustrating the efficiency of a power conversion device to which the technology of Patent Document 1 is applied, Fig. 5B is a diagram illustrating the efficiency of a power conversion device to which the technology of Non-Patent Document 1 is applied, and Fig. 6 is a diagram illustrating the efficiency of power conversion device 1 according to an embodiment. The power conversion device to which the conventional technology is applied has the same circuit configuration as power conversion device 1 shown in Fig. 1 but differs in the method of controlling the switching of switches AH, AL, BH, and BL. Figs. 5A, 5B, and 6 show, from top to bottom, the time waveforms of the current flowing in the primary side circuit, the gate voltage of switch BL, and the gate voltage of switch AH.

[0060] In a power conversion device employing the technology of Patent Document 1, the gain is first adjusted using only the switching frequency, and if the target gain cannot be obtained even when the switching frequency reaches its upper limit, phase shift control is performed. Therefore, in a power conversion device employing the technology of Patent Document 1, it is difficult to find a switching frequency and phase difference that results in high efficiency. In a power conversion device employing the technology of Patent Document 1, as shown in FIG. 5A , both switches AH and BL are turned off while a large current is flowing, resulting in increased losses and reduced efficiency. Furthermore, because the device operates in a high switching frequency range, conduction losses increase due to the skin effect and other factors, and transformer core losses also increase.

[0061] In a power conversion device employing the technology of Non-Patent Document 1, phase shift control is performed with the switching frequency fixed. Therefore, even in a power conversion device employing the technology of Non-Patent Document 1, it is difficult to find a switching frequency and phase difference that results in high efficiency. In a power conversion device employing the technology of Non-Patent Document 1, as shown in FIG. 5B , the reactive power period increases the current when switch AH is turned off, resulting in increased losses and reduced efficiency. Furthermore, a long reactive power period increases the current RMS (Root Mean Square) value, which also increases the conduction loss of the entire system.

[0062] On the other hand, the power conversion device 1 according to the embodiment uses a trained model that outputs output parameters that can increase efficiency while realizing a desired gain, making it possible to find a switching frequency and phase difference that results in high efficiency. In the power conversion device 1 according to the embodiment, as shown in FIG. 6 , the switch AH is turned off while a large current is flowing, but the current when the switch BL is turned off is small. Furthermore, since the reactive power period can be adjusted to be as short as possible, the peak current of the primary side current can be reduced, thereby suppressing losses during turn-off. Furthermore, since the RMS value of the current can be reduced, the conduction loss of the entire system can also be suppressed. These effects result in improved efficiency.

[0063] In addition to the input voltage, output voltage, and output resistance, the two or more input parameters may include at least one of the package surface temperature or peripheral substrate temperature of at least one of the switches AH, AL, BH, and BL, the surface temperature of the core of the transformer T1, and the surface temperature of the winding of the transformer T1. Because the optimal output parameters vary depending on the temperature rise of each switch or the transformer, learning such temperature information also enables control that takes temperature dependency into account, thereby achieving even higher efficiency.

[0064] As described above, by using a trained model or table data having the functions of a trained model to control the switching of switches AH, AL, BH, and BL, it is possible to achieve a desired gain while improving efficiency.

[0065] (Other Embodiments) As described above, the embodiments have been described as examples of the technology according to the present disclosure. However, the technology according to the present disclosure is not limited to these, and can be applied to embodiments in which modifications, substitutions, additions, omissions, etc. are made as appropriate. For example, the following modifications are also included in one embodiment of the present disclosure.

[0066] For example, in the above embodiment, the output parameters are the switching frequency and the phase difference, but are not limited to this. For example, the output parameters are not particularly limited as long as they can control the switches AH, AL, BH, and BL, such as the gate on-time, the switch on-timing, or the switch off-timing.

[0067] For example, the present disclosure can be realized not only as the power conversion device 1 but also as a method for manufacturing the power conversion device 1 .

[0068] FIG. 7 is a flowchart showing an example of a method for manufacturing the power conversion device 1 according to another embodiment.

[0069] The manufacturing method is a method for manufacturing a power conversion device 1, and the power conversion device 1 includes: a first switch provided on a first path connecting a first input terminal and a second input terminal; a second switch provided on the first path and connected in series with the first switch; a third switch provided on a second path different from the first path connecting the first input terminal and the second input terminal; a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node on the first path between the first switch and the second switch and a second node on the second path between the third switch and the fourth switch; a resonant capacitor connected between the first node and the primary winding or between the second node and the primary winding; The power conversion device includes a resonant inductor and a control unit 10 that controls the switching of the first switch, the second switch, the third switch, and the fourth switch. The manufacturing method includes, as shown in FIG. 7 , a step (step S11) of generating a trained model using pre-prepared parameters to output output parameters for controlling the switching of the first switch, the second switch, the third switch, and the fourth switch such that the efficiency of the power conversion device 1 is equal to or greater than a predetermined value when input parameters are input, and a step (step S12) of incorporating the trained model into the power conversion device 1, or generating table data having the functions of the trained model from the trained model, and incorporating the table data into the power conversion device 1.

[0070] FIG. 8 is a flowchart showing a specific example of a method for manufacturing the power conversion device 1 according to another embodiment.

[0071] For example, in the generating step (step S11), a model for collecting combinations of efficiencies equal to or greater than a predetermined value and output parameters may be generated using parameters prepared in advance (step S21), and a trained model may be generated using the collected combinations (step S22). The model for collecting combinations of efficiencies equal to or greater than a predetermined value and output parameters may be, for example, the neural network shown in FIG.

[0072] When the trained model is incorporated into the power conversion device 1 (step S23) in the incorporating step (step S12), the manufacturing method may further include a step of retraining the trained model (step S24). By performing retraining at the stage of product shipping or the like, the accuracy of the trained model can be improved.

[0073] For example, in the re-learning step, the trained model may be re-trained using parameters of the same type as the pre-prepared parameters detected on the circuit constituting the power conversion device 1. There may be a discrepancy between the component parameters of the power conversion device used when training the trained model and the component parameters of the power conversion device 1 (i.e., the shipped product) in which the trained model is incorporated, which may result in the trained model not outputting optimal output parameters. Therefore, the trained model is re-trained using parameters of the same type as the parameters used when training the trained model, detected from the power conversion device 1 in which the trained model is incorporated. For example, if the input voltage, output voltage, and output current were used during training, the input voltage, output voltage, and output current detected from the power conversion device 1 in which the trained model is incorporated are also used during re-learning. This allows the discrepancy to be corrected, making it possible to provide a product that maintains high efficiency.

[0074] Furthermore, for example, in the re-learning step, the trained model may be re-trained using parameters of a type different from the pre-prepared parameters detected on the circuits constituting the power conversion device 1 (e.g., the surface temperature of the switch package or the temperature of the surrounding substrate, the surface temperature of the insulating transformer core, and the surface temperature of the insulating transformer windings, etc.). By re-training the trained model using parameters of a type different from the parameters used when training the trained model, control that takes into account the different types of parameters becomes possible.

[0075] In the above embodiment, each component included in the power conversion device 1 may be configured with dedicated hardware, or may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0076] Some or all of the functions of the power conversion device 1 according to the above embodiment are typically realized as an LSI, which is an integrated circuit. These may be individually integrated into single chips, or may be integrated into a single chip that includes some or all of the functions. Furthermore, the integrated circuit is not limited to an LSI, and may be realized using a dedicated circuit or a general-purpose processor. It is also possible to use an FPGA (Field Programmable Gate Array) that can be programmed after LSI manufacturing, or a reconfigurable processor that can reconfigure the connections and settings of circuit cells within the LSI.

[0077] Furthermore, if an integrated circuit technology that can replace LSI emerges due to advances in semiconductor technology or other derived technologies, it is natural that each component included in the power conversion device 1 can be integrated using that technology.

[0078] In addition, this disclosure also includes forms obtained by making various modifications to the embodiments that a person skilled in the art would think of, and forms realized by arbitrarily combining the components and functions in each embodiment within the scope of the present disclosure.

[0079] (Additional Notes) The above description of the embodiments discloses the following techniques.

[0080] a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node on the first path between the first switch and the second switch and a second node on the second path between the third switch and the fourth switch; a resonant capacitor and a resonant inductor connected between the first node and the primary winding or between the second node and the primary winding; and a control unit that controls switching of the first switch, the second switch, the third switch, and the fourth switch using a trained model that outputs an output parameter when an input parameter is input, or table data having a function of the trained model.

[0081] According to this, output parameters for controlling the switching of the first switch, the second switch, the third switch, and the fourth switch when a desired gain is achieved and efficiency is improved are prepared in advance, and machine learning is performed using these as training data, thereby generating a trained model that outputs output parameters that can improve efficiency while achieving the desired gain. Note that since the trained model has a function of outputting output parameters for input parameters, table data showing the correspondence between input parameters input to the trained model and output parameters output for the input parameters can be generated, thereby generating table data having the function of the trained model. Therefore, by controlling the switching of the first switch, the second switch, the third switch, and the fourth switch using such a trained model or table data having the function of such a trained model, it is possible to improve efficiency while achieving the desired gain.

[0082] (Technology 2) The power conversion device according to Technology 1, wherein the input parameters include parameters detected on a circuit constituting the power conversion device.

[0083] According to this, by inputting input parameters detected on the circuit constituting the power conversion device into a trained model or table data, it is possible to obtain output parameters that can increase efficiency while achieving the desired gain.

[0084] (Technology 3) The power conversion device according to Technology 2, wherein the control unit controls switching of the first switch, the second switch, the third switch, and the fourth switch using two or more types of output parameters obtained by inputting two or more types of the input parameters into the trained model or the table data.

[0085] According to this, by inputting two or more types of input parameters detected on the circuit that constitutes the power conversion device into a learned model or table data, two or more types of output parameters that can increase efficiency while achieving the desired gain can be obtained.

[0086] (Technology 4) The power conversion device according to Technology 3, wherein the two or more types of input parameters include two or more types of parameters from among an input voltage, an output voltage, an input current, and an output current of the power conversion device, and the two or more types of output parameters include switching frequencies of the first switch, the second switch, the third switch, and the fourth switch, and a phase difference between switching of the first switch and the second switch and switching of the third switch and the fourth switch.

[0087] According to this, by inputting two or more input parameters of the input voltage, output voltage, input current, and output current detected in the circuit constituting the power conversion device into the trained model or table data, it is possible to obtain a switching frequency and phase difference that can improve efficiency while realizing a desired gain. By using the switching frequency and phase difference for gate control of each switch, not only is efficiency improved but also a wide range of the power conversion device can be achieved (i.e., it can handle a wide range of input and output voltages).

[0088] (Technology 5) The power conversion device according to Technology 4, wherein the two or more types of input parameters include either an input voltage and an output voltage of the power conversion device, or an input current and an output current of the power conversion device.

[0089] According to this, by inputting three types of input parameters, namely the input voltage and output voltage, and the input current and output current of the power conversion device, detected on the circuit that constitutes the power conversion device into a trained model or table data, it is possible to obtain two or more types of output parameters that can increase efficiency while achieving the desired gain.

[0090] (Technology 6) The power conversion device according to any one of Techniques 3 to 5, wherein the two or more types of input parameters include at least one of a package surface temperature or a surrounding substrate temperature of at least one of the first switch, the second switch, the third switch, and the fourth switch, a surface temperature of a core of the isolation transformer, and a surface temperature of a winding of the isolation transformer.

[0091] According to this, the optimal output parameters fluctuate depending on the temperature rise of each switch or transformer, so by learning such temperature information, it becomes possible to control the system taking temperature dependency into account, thereby achieving even higher efficiency.

[0092] (Technology 7) A power conversion device according to any one of techniques 1 to 6, wherein the trained model is a neural network in which the efficiency of the power conversion device is set as an objective function.

[0093] According to this, by using a neural network in which efficiency is set as the objective function, it is possible to achieve even higher efficiency.

[0094] (Technology 8) A power conversion device according to any one of techniques 1 to 7, wherein the trained model outputs the output parameters that are equal to or greater than a predetermined lower limit value and within a predetermined upper limit value.

[0095] According to this, by setting a lower limit value and an upper limit value for the output parameter, it is possible to obtain an output parameter that can increase efficiency while realizing a desired gain in a short time during machine learning.

[0096] (Technology 9) A method for manufacturing a power conversion device, the power conversion device including: a first switch provided on a first path connecting a first input terminal and a second input terminal; a second switch provided on the first path and connected in series with the first switch; a third switch provided on a second path different from the first path connecting the first input terminal and the second input terminal; a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node on the first path between the first switch and the second switch and a second node on the second path between the third switch and the fourth switch; a resonant capacitor and a resonant inductor connected to each other, and a control unit that controls switching of the first switch, the second switch, the third switch, and the fourth switch, and the manufacturing method includes the steps of: generating, using previously prepared parameters, a trained model that outputs output parameters for controlling switching of the first switch, the second switch, the third switch, and the fourth switch when input parameters are input, so that the efficiency of the power conversion device is equal to or greater than a predetermined value; and incorporating the trained model into the power conversion device, or generating table data from the trained model that has the functions of the trained model, and incorporating the table data into the power conversion device.

[0097] This makes it possible to manufacture a power conversion device that can achieve a desired gain while increasing efficiency.

[0098] (Technology 10) A manufacturing method described in Technology 9, wherein in the generating step, a model is generated for collecting combinations of the efficiency and the output parameters that are equal to or greater than the predetermined value using the pre-prepared parameters, and the trained model is generated using the collected combinations.

[0099] In this way, a model may be generated to collect combinations of efficiency and output parameters that are equal to or greater than a predetermined value, and then a trained model may be generated using this model.

[0100] (Technology 11) A manufacturing method described in Technology 9 or 10, wherein, if the trained model is incorporated into the power conversion device in the incorporating step, the manufacturing method further includes a step of re-training the trained model.

[0101] This allows for re-learning at the product shipping stage, etc., thereby improving the accuracy of the trained model.

[0102] (Technology 12) A manufacturing method described in Technology 11, wherein in the re-learning step, the trained model is re-trained using parameters of the same type as the previously prepared parameters detected on a circuit constituting the power conversion device.

[0103] There may be a discrepancy between the component parameters of the power conversion device used when training the trained model and the component parameters of the power conversion device (i.e., the product to be shipped) into which the trained model is incorporated, and the trained model may not output optimal output parameters. Therefore, the trained model is retrained using parameters of the same type as those used when training the trained model, detected from the power conversion device into which the trained model is incorporated. For example, if the input voltage, output voltage, and output current were used during training, the input voltage, output voltage, and output current detected from the power conversion device into which the trained model is incorporated are also used during retraining. This allows the discrepancy to be corrected, making it possible to provide a product that maintains high efficiency.

[0104] (Technology 13) A manufacturing method described in Technology 11, wherein in the re-learning step, the trained model is re-trained using parameters of a type different from the pre-prepared parameters detected on a circuit constituting the power conversion device.

[0105] This allows the trained model to be retrained using parameters of different types than those used when training the trained model, making it possible to perform control that takes into account these different types of parameters.

[0106] The present disclosure can be applied to LLC converters that can accommodate a wide range of input and output voltages.

[0107] REFERENCE SIGNS LIST 1 Power conversion device 10 Control unit 20 Primary side gate drive circuit AH, AL, BH, BL, CH, CL, DH, DL Switch Cr Capacitor Lm, Lr Inductor N1, N2, N3, N4 Node P1, P2, P3, P4 Path T1 Transformer t1, t2, t3, t4 Terminal

Claims

1. A power conversion device comprising: a first switch provided on a first path connecting a first input terminal and a second input terminal; a second switch provided on the first path and connected in series with the first switch; a third switch provided on a second path different from the first path and connecting the first input terminal and the second input terminal; a fourth switch provided on the second path and connected in series with the third switch; an insulating transformer having a primary winding connected between a first node between the first switch and the second switch on the first path and a second node between the third switch and the fourth switch on the second path; a resonant capacitor and a resonant inductor connected between the first node and the primary winding or between the second node and the primary winding; and a control unit configured to control switching of the first switch, the second switch, the third switch, and the fourth switch using a learned model that outputs an output parameter when an input parameter is input, or table data having a function of the learned model.

2. The power conversion device according to claim 1, wherein the input parameter includes a parameter detected on a circuit constituting the power conversion device.

3. The power conversion device according to claim 2, wherein the control unit controls switching of the first switch, the second switch, the third switch, and the fourth switch using two or more output parameters obtained by inputting two or more of the input parameters to the learned model or the table data.

4. The power conversion device according to claim 3, wherein the two or more input parameters include two or more parameters among an input voltage, an output voltage, an input current, and an output current of the power conversion device, and the two or more output parameters include switching frequencies of the first switch, the second switch, the third switch, and the fourth switch, and a phase difference between switching of the first switch and the second switch and switching of the third switch and the fourth switch.

5. The power conversion device according to claim 4, wherein the two or more input parameters include any one of an input voltage and an output voltage of the power conversion device and any one of an input current and an output current of the power conversion device.

6. The power conversion device according to claim 3, wherein at least two of the input parameters include at least one temperature among the package surface temperature or the peripheral substrate temperature of at least one of the first switch, the second switch, the third switch, and the fourth switch, the surface temperature of the core of the insulation transformer, and the surface temperature of the winding of the insulation transformer.

7. The power conversion device according to any one of claims 1 to 6, wherein the learned model is a neural network in which the efficiency of the power conversion device is set as an objective function.

8. The power conversion device according to any one of claims 1 to 6, wherein the learned model outputs the output parameter within a predetermined upper limit value and not less than a predetermined lower limit value.

9. A method for manufacturing a power conversion device, wherein the power conversion device includes: a first switch provided on a first path connecting a first input terminal and a second input terminal; a second switch provided on the first path and connected in series with the first switch; a third switch provided on a second path different from the first path and connecting the first input terminal and the second input terminal; a fourth switch provided on the second path and connected in series with the third switch; an isolation transformer having a primary winding connected between a first node between the first switch and the second switch on the first path and a second node between the third switch and the fourth switch on the second path; a resonance capacitor and a resonance inductor connected between the first node and the primary winding or between the second node and the primary winding; and a control unit configured to control switching of the first switch, the second switch, the third switch, and the fourth switch. The manufacturing method includes: generating a learned model that outputs output parameters for controlling switching of the first switch, the second switch, the third switch, and the fourth switch such that the efficiency of the power conversion device is equal to or higher than a predetermined value when input parameters are input, using the parameters prepared in advance; and incorporating the learned model into the power conversion device, or generating table data having the functions of the learned model from the learned model and incorporating the table data into the power conversion device.

10. The manufacturing method according to claim 9, wherein in the generating step, a model for collecting combinations of the efficiency equal to or higher than the predetermined value and the output parameters is generated using the parameters prepared in advance, and the learned model is generated using the collected combinations.

11. The manufacturing method according to claim 9 or 10, wherein when the learned model is incorporated into the power conversion device in the incorporating step, the manufacturing method further includes a step of re-learning the learned model.

12. The manufacturing method according to claim 11, wherein in the step of relearning, the learned model is relearned using parameters of the same type as the parameters prepared in advance, which are detected on the circuit constituting the power conversion device.

13. The manufacturing method according to claim 11, wherein in the step of relearning, the learned model is relearned using parameters of a type different from the parameters prepared in advance, which are detected on the circuit constituting the power conversion device.

Images