Power supply control method and device of robot, robot and medium

Abstract

The invention relates to the technical field of power supply control, in particular to a power supply control method and device of a robot, the robot and a medium, and is used for solving the problem that the peak redundancy capability and the transient response capability cannot be improved by power supply control of the robot in the related technology. The current operation state of the computing power module is determined, the target PWM frequency in the current sampling period is determined according to the current operation state, the computing power module is used for processing artificial intelligence (AI) computing power work and is powered by multiple power supply units connected in parallel, each power supply unit comprises a driver and a metal oxide semiconductor field effect transistor (MOSFET), and the working load value comprises a current instantaneous value or a power instantaneous value; the PWM signal with the target PWM frequency is output to the multi-path power supply unit, so that the current output to the computing power module in the current sampling period is adjusted, and the peak redundancy capability and the transient response capability of the power supply system are remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of power control technology, and in particular to a power control method, device, robot and medium for a robot. Background Art

[0002] In robots with embodied intelligence, high-computing graphics processing units (GPUs) or tensor processing units (TPUs) are usually integrated to act as computing centers to realize the reasoning and learning capabilities of machine intelligence. These GPUs or TPUs that support artificial intelligence (AI) computing power are generally manufactured using advanced semiconductor processes. The voltage of the core computing unit is around 1V or lower, and it requires more than 100A of current when working, or even more than 500A of current. Therefore, the computing power module consumes a lot of current in the robot system. In the actual operation of the computing power module, according to the distribution characteristics of the computing load of different tasks, the high current state of the work is not maintained at a stable level, but presents huge and rapid fluctuations. This requires that the current supply unit must have sufficient peak redundancy and transient response capabilities in addition to the ability to output large currents.

[0003] For peak redundancy, the related technologies usually provide a larger current by increasing the number of phases (paths) of parallel power supply units, which can also ensure voltage stability and maintain the thermal balance of each power component. However, the more phases of multi-phase power supply units in parallel, the more components they contain, which not only increases the total failure rate of the power system, but also increases the PCB area and cost, and increases the difficulty of miniaturizing the power system. At the same time, it also increases the switching loss of the power component switch at low load, reducing the efficiency of the overall power conversion system. As for transient response capability, there is no corresponding solution in the related technologies. Summary of the invention

[0004] The embodiments of the present application provide a power control method, device, robot and medium for a robot, so as to improve the current transient response capability and peak redundancy capability of a load and reduce battery power consumption.

[0005] The specific technical solutions provided by the embodiments of this application are as follows:

[0006] In a first aspect, an embodiment of the present application provides a power control method for a robot, comprising:

[0007] Continuously sample to obtain the workload value of the preset computing power module;

[0008] Determine the current operating state of the computing power module based on the obtained workload values;

[0009] Determine a target pulse width modulation (PWM) frequency within a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload;

[0010] Outputting a PWM signal with the target PWM frequency to the multi-channel power supply unit to adjust the current output to the computing power module in the current sampling period;

[0011] Among them, the computing power module is used to process artificial intelligence AI computing power work, and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes the instantaneous value of the current passing through the computing power module, or the instantaneous value of the power consumed by the computing power module.

[0012] In some embodiments of the present application, by continuously sampling the instantaneous current value / instantaneous power value of the computing power module and analyzing it, the actual operating state of the computing power module, which is a module in the robot used to process AI computing power work, can be determined, thereby determining a target PWM frequency adapted to the corresponding actual operating state based on the actual operating state of the computing power module, so as to dynamically adjust the driving switching frequency of the multi-channel power supply unit that supplies power to the computing power module, wherein the target PWM frequency is positively correlated with the workload (i.e., power consumption), thereby reducing switching losses and improving power supply life in a low-power operating state, and significantly improving the peak redundancy capability and transient response capability of the power supply system in a high-power operating state.

[0013] In a possible implementation, determining the current operating state of the computing power module based on the obtained workload values ​​includes:

[0014] Comparing the obtained workload values ​​with a first threshold, a second threshold, and a third threshold in sequence, respectively, wherein the second threshold is greater than the first threshold and less than the third threshold;

[0015] If it is determined that each workload value is less than or equal to the first threshold, determining that the current operating state of the computing power module is a dormant state;

[0016] If it is determined that each workload value is greater than the first threshold and less than or equal to the second threshold, determining that the current operating state of the computing power module is the first operating state;

[0017] If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, the current operating state of the computing power module is determined to be the second operating state.

[0018] The above method determines the current operating state of the computing power module by sampling the workload value, so as to facilitate the subsequent use of different PWM frequency adjustments for different operating states, thereby significantly improving the peak redundancy capability and transient response capability of the power supply system.

[0019] In a possible implementation, determining a target pulse width modulation (PWM) frequency in a current sampling period according to the current operating state includes:

[0020] If the current operating state is the sleep state, shutting down at least one of the multiple power supply units, and determining the first fixed PWM frequency as the target PWM frequency;

[0021] If the current operating state is the first working state, at least one of the multiple power supply units that is turned off is turned on, and a second fixed PWM frequency is determined as the target PWM frequency, wherein the first fixed PWM frequency is less than the second fixed PWM frequency;

[0022] If the current operating state is the second operating state, the target PWM frequency within the current sampling period is determined according to the change trend of the workload values ​​and the expected PWM frequency corresponding to the next sampling moment, wherein the expected PWM frequency is determined based on the waveform obtained by fitting the workload values ​​and the next sampling moment.

[0023] In the above method, different methods of determining the target PWM frequency are adopted for different operating states of the computing power module, so as to reduce switching losses and improve power supply life in low-power operating states, and significantly improve the peak redundancy capability and transient response capability of the power supply system in high-power operating states.

[0024] In a possible implementation, determining the target PWM frequency in the current sampling period according to the change trend of the workload values ​​and the expected PWM frequency corresponding to the next sampling moment includes:

[0025] Based on the workload values, fitting the workload waveform of the computing power module;

[0026] The expected PWM frequency is predicted according to the workload waveform and the next sampling time, wherein if the expected PWM frequency is greater than an upper limit value of the PWM frequency, the expected PWM frequency is the upper limit value of the PWM frequency;

[0027] Based on the workload values, determining the difference between two adjacent workload values ​​in sequence;

[0028] The target PWM frequency in the current sampling period is determined according to the determined variation trend of each difference and the expected PWM frequency.

[0029] In the above method, the target PWM frequency in the current sampling period is predicted according to the variation trend of each workload value obtained by sampling, and an overshoot operation can be performed in advance, thereby improving the current output capacity and ensuring the power consumption requirement of the load under intensive computing tasks.

[0030] In a possible implementation, determining the target PWM frequency in the current sampling period according to the determined change trends of the differences and the expected PWM frequency includes:

[0031] If the variation trend of the differences is an increasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a first adjustment amplitude, wherein the first adjustment amplitude is greater than a preset value;

[0032] If the variation trend of the differences is a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a second adjustment amplitude, wherein the second adjustment amplitude is less than the preset value;

[0033] If the change trend of the differences changes from an increasing trend to a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a third adjustment amplitude, wherein the third adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value;

[0034] If the changing trend of the differences changes from a decreasing trend to an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a fourth adjustment amplitude, wherein the fourth adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value.

[0035] In the above method, specific adjustment methods are made according to different change trends to achieve pre-overshoot operation, thereby improving the current output capacity and ensuring the power consumption requirements of the load under intensive computing tasks.

[0036] In a possible implementation, if there are at least two turned-on power supply units in the sleep state or the first working state, and the target PWM frequency includes a plurality of sub-PWM frequencies arranged in time sequence, each sub-PWM frequency is obtained in the following manner:

[0037] Determine the fixed PWM frequency as the first sub-PWM frequency among the multiple sub-PWM frequencies;

[0038] For each non-first sub-PWM frequency, perform the following operations:

[0039] For any sub-PWM frequency other than the first one, generate a random number;

[0040] Determine the sum of the PWM frequency obtained by mapping the random number and the fixed PWM frequency as any non-first sub-PWM frequency;

[0041] If the current operating state is the sleep state, the fixed PWM frequency is the first fixed PWM frequency; if the current operating state is the first working state, the fixed PWM frequency is the second fixed PWM frequency.

[0042] In the above method, the fixed PWM frequency adjustment method is used in the sleep state or the first working state, which may cause output energy superposition and may cause interference to surrounding electronic equipment and systems. In the embodiment of the present application, the PWM scrambling working mode is adopted to effectively reduce the energy pulse peaks in EMI, thereby weakening or eliminating EMI.

[0043] In a possible implementation, the step of outputting a PWM signal with the target PWM frequency to the multi-power supply unit includes:

[0044] Determining whether to perform grouping adjustment on the multiple power supply units;

[0045] If yes, the multiple power supply units are divided into multiple groups, and the PWM signals with the target PWM frequency are output to the power supply units in the multiple groups in sequence;

[0046] If not, the PWM signals with the target PWM frequency are output to the multiple power supply units in sequence.

[0047] In the above method, the configuration of the group adjustment mode can make the computing power module have a larger PWM frequency bandwidth when the computing load fluctuation frequency is high, which can effectively separate the output current ripple and reduce its noise, and also reduce the harmonic components in the output waveform, further reducing the noise level. Furthermore, it can also make the high-bandwidth high-frequency DC control usually have a faster dynamic response, and can adjust the output voltage more quickly to adapt to the computing load changes of the computing power module, thereby reducing the noise caused by transient response.

[0048] In a second aspect, an embodiment of the present application provides a power control device for a robot, comprising:

[0049] A sampling module is used to continuously sample and obtain the workload value of a preset computing power module;

[0050] A first determination module, configured to determine a current operating state of the computing power module based on each obtained workload value;

[0051] A second determination module is used to determine a target pulse width modulation (PWM) frequency within a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload;

[0052] An output module, used to output a PWM signal with the target PWM frequency to the multi-channel power supply unit, so as to adjust the current output to the computing power module in the current sampling period;

[0053] Among them, the computing power module is used to process artificial intelligence AI computing power work, and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes the instantaneous value of the current passing through the computing power module, or the instantaneous value of the power consumed by the computing power module.

[0054] In a possible implementation manner, the first determining module is specifically configured to:

[0055] Comparing the obtained workload values ​​with a first threshold, a second threshold, and a third threshold in sequence, respectively, wherein the second threshold is greater than the first threshold and less than the third threshold;

[0056] If it is determined that each workload value is less than or equal to the first threshold, determining that the current operating state of the computing power module is a dormant state;

[0057] If it is determined that each workload value is greater than the first threshold and less than or equal to the second threshold, determining that the current operating state of the computing power module is the first operating state;

[0058] If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, the current operating state of the computing power module is determined to be the second operating state.

[0059] In a possible implementation manner, the second determining module is specifically configured to:

[0060] If the current operating state is the sleep state, shutting down at least one of the multiple power supply units, and determining the first fixed PWM frequency as the target PWM frequency;

[0061] If the current operating state is the first working state, at least one of the multiple power supply units that is turned off is turned on, and a second fixed PWM frequency is determined as the target PWM frequency, wherein the first fixed PWM frequency is less than the second fixed PWM frequency;

[0062] If the current operating state is the second operating state, the target PWM frequency within the current sampling period is determined according to the change trend of the workload values ​​and the expected PWM frequency corresponding to the next sampling moment, wherein the expected PWM frequency is determined based on the waveform obtained by fitting the workload values ​​and the next sampling moment.

[0063] In a possible implementation manner, the second determining module is specifically configured to:

[0064] Based on the workload values, fitting the workload waveform of the computing power module;

[0065] The expected PWM frequency is predicted according to the workload waveform and the next sampling time, wherein if the expected PWM frequency is greater than an upper limit value of the PWM frequency, the expected PWM frequency is the upper limit value of the PWM frequency;

[0066] Based on the workload values, determining the difference between two adjacent workload values ​​in sequence;

[0067] The target PWM frequency in the current sampling period is determined according to the determined variation trends of the differences and the expected PWM frequency.

[0068] In a possible implementation manner, the second determining module is specifically configured to:

[0069] If the variation trend of the differences is an increasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a first adjustment amplitude, wherein the first adjustment amplitude is greater than a preset value;

[0070] If the variation trend of the differences is a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a second adjustment amplitude, wherein the second adjustment amplitude is less than the preset value;

[0071] If the change trend of the differences changes from an increasing trend to a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a third adjustment amplitude, wherein the third adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value;

[0072] If the changing trend of the differences changes from a decreasing trend to an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a fourth adjustment amplitude, wherein the fourth adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value.

[0073] In a possible implementation, if there are at least two turned-on power supply units in the sleep state or the first working state, and the target PWM frequency includes a plurality of sub-PWM frequencies arranged in time sequence, each sub-PWM frequency is obtained in the following manner:

[0074] Determine the fixed PWM frequency as the first sub-PWM frequency among the multiple sub-PWM frequencies;

[0075] For each non-first sub-PWM frequency, perform the following operations:

[0076] For any sub-PWM frequency other than the first one, generate a random number;

[0077] Determine the sum of the PWM frequency obtained by mapping the random number and the fixed PWM frequency as any non-first sub-PWM frequency;

[0078] If the current operating state is the sleep state, the fixed PWM frequency is the first fixed PWM frequency; if the current operating state is the first working state, the fixed PWM frequency is the second fixed PWM frequency.

[0079] In a possible implementation, the output module is specifically used to:

[0080] Determining whether to perform grouping adjustment on the multiple power supply units;

[0081] If yes, the multiple power supply units are divided into multiple groups, and the PWM signals with the target PWM frequency are output to the power supply units in the multiple groups in sequence;

[0082] If not, the PWM signals with the target PWM frequency are output to the multiple power supply units in sequence.

[0083] In a third aspect, an embodiment of the present application provides a robot, comprising:

[0084] Computing module, used to process artificial intelligence (AI) computing work;

[0085] Multiple power supply units connected in parallel are used to supply power to the computing power module, each power supply unit comprising a driver and a MOSFET;

[0086] Memory, used to store computer programs or instructions;

[0087] A processor is used to execute the computer program or instructions in the memory so that any method in the first aspect described above is executed.

[0088] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which, when instructions in the storage medium are executed by a processor, enables the processor to execute any of the methods described in the first aspect above.

[0089] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes: a computer program code, when the computer program code is run on a computer, enables the computer to execute the method as described in any one of the first aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 A schematic diagram of current peak value / power peak value and waveform in a different operating state in an embodiment of the present application;

[0091] Figure 2 A schematic flow chart of a power control method for a robot in an embodiment of the present application;

[0092] Figure 3 A schematic diagram of a system architecture of power control of a robot in an embodiment of the present application;

[0093] Figure 4 A flowchart of a method for determining the current operating state of a computing power module in an embodiment of the present application;

[0094] Figure 5 A schematic diagram of a flow chart of a PWM frequency scrambling method in an embodiment of the present application;

[0095] Figure 6 A flow chart of a method for determining a target PWM frequency according to a change trend in an embodiment of the present application;

[0096] Figure 7 A waveform diagram of an envelope of current sampled at different times in an embodiment of the present application;

[0097] Figure 8A schematic diagram of 4-way PWM frequency conversion adjustment of the optional DC controller predicted output in an embodiment of the present application;

[0098] Fig. 9 A flowchart of a method for determining whether to perform group adjustment in an embodiment of the present application;

[0099] Fig.10 This is a schematic diagram of 4-way PWM frequency conversion adjustment of the output of an optional grouped DC controller in an embodiment of the present application;

[0100] Fig.11 A schematic diagram of the logic architecture of a power control device of a robot in an embodiment of the present application;

[0101] Fig.12 Schematic diagram of the physical structure of the robot in the embodiment of the present application. DETAILED DESCRIPTION

[0102] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0103] Below, some terms in the embodiments of the present application are explained to facilitate understanding by those skilled in the art.

[0104] (1) The term “MOSFET” in the embodiments of the present application is a metal-oxide semiconductor field-effect transistor, also known as a metal-oxide-semiconductor field-effect transistor (MOSFET, abbreviated as MOS).

[0105] (2) The terms "first", "second", "third", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that the terms used in this way are interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in sequences other than those illustrated or described herein.

[0106] The following is a brief introduction to the design concept of the embodiments of the present application.

[0107] The present application relates to the field of power control technology, and mainly to a power control method, device, robot and medium of a robot.

[0108] In robots with embodied intelligence, high-computing GPUs or TPUs are usually integrated to serve as computing centers to realize the reasoning and learning capabilities of machine intelligence. Since these computing modules that support AI computing capabilities are generally manufactured using advanced semiconductor processes, the voltage of the core computing unit is around 1V or lower, and a current of more than 100A or even more than 500A is required when working. In the actual operation of the computing module, according to the distribution characteristics of the computing load of different tasks, the high current state of work is not maintained at a stable level, but presents huge and rapid fluctuations. This requires that the current supply unit of the computing module must have sufficient peak redundancy and transient response capabilities in addition to having high current output capabilities.

[0109] In the related art, the implementation method of multi-phase power supply units in parallel is usually adopted to provide sufficient peak redundancy capacity for the computing power module, that is, each phase power supply unit provides a fraction of the total current to the computing power module, so that each phase also shares the thermal power. The implementation method of using multi-phase power supply units in parallel includes but is not limited to the following advantages: First, the current of each phase can be small, and multi-phase parallel connection can provide a larger current; second, by utilizing phase shift operation, the core voltage signal that can be obtained by multi-phase power supply is also more stable than that of single-phase; third, the temperature of the power supply circuit can be reduced, because the total current performs multi-way shunting, and the heat generation of each device is naturally reduced, thereby reducing the temperature of the power supply circuit. Furthermore, the multi-phase power supply unit can also very accurately balance the current output by each phase power supply unit to maintain the thermal balance of each power component.

[0110] However, the multi-phase parallel implementation also has the following obvious disadvantages: First, the total failure rate of the system is relatively large. The reason for the high total failure rate is that each component has a "failure rate". The more phases designed in the multi-phase parallel implementation, the more components are used, and the greater the total failure rate of the system; second, more components cause a higher printed circuit board (PCB) area and cost, which makes it difficult to implement the miniaturization of the power system design; third, when the computing power module is in low computing load, no computing load, or sleep state, due to the switching loss of the DC controller driving the power MOSFET switch, the increase in the number of power devices such as MOSFET caused by too many phases will significantly reduce the efficiency of the overall power conversion system.

[0111] Regarding transient response capability, the above-mentioned multi-phase parallel implementation method cannot improve the current transient response capability. Furthermore, when the number of phases is fixed, there is also a method in the related art to reduce switching losses by opening / closing certain phase circuits, but because there is inductive energy storage and capacitor filtering at the output end, opening / closing with too long an interval will greatly reduce the current transient response, and therefore cannot improve the current transient response capability.

[0112] In view of this, in order to solve the problem that the power control of the robot in the related art cannot improve the peak redundancy capability and transient response capability, an embodiment of the present application provides a power control method for a robot. In the embodiment of the present application, continuous sampling is performed to obtain a workload value of a preset computing power module; wherein the computing power module is used to process AI computing power work and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the aforementioned workload value includes an instantaneous value of the current passing through the computing power module, or an instantaneous value of the power consumed by the computing power module; based on the obtained workload values, the current operating state of the computing power module is determined, and according to the current operating state, the target pulse width modulation (PWM) frequency within the current sampling period is determined, wherein the target PWM frequency is positively correlated with the workload; then, a PWM signal with the target PWM frequency is output to the multiple power supply units to adjust the current output to the computing power module within the current sampling period.

[0113] In the present application, by continuously sampling the instantaneous current value / instantaneous power value of the computing power module and analyzing it, the actual operating state of the computing power module, which is a module in the robot used to process AI computing power work, can be determined, thereby determining the target PWM frequency adapted to the corresponding actual operating state according to the actual operating state of the computing power module, so as to dynamically adjust the driving switching frequency of the multi-way power supply unit powered by the computing power module, wherein the target PWM frequency is positively correlated with the workload (i.e., power consumption), thereby reducing switching losses and improving power supply life in a low-power operating state, and significantly improving the peak redundancy capability and transient response capability of the power supply system in a high-power operating state.

[0114] A robot power control method provided in an embodiment of the present application is applicable to robots, such as mobile robots, intelligent agents, multi-legged robots, bionic humans, bipedal robots, humanoid / type robots, etc.

[0115] The preferred implementation modes of the present application are further described in detail below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application, and the embodiments of the present application and the features in the embodiments may be combined with each other if there is no conflict.

[0116] As mentioned above, in actual operation, the computing power module that supports AI computing power will show huge and rapid fluctuations in the high current state during operation due to the distribution characteristics of the computing load, i.e., irregularity. For example, take GPU or XPU as an example. Figure 1 As shown, a schematic diagram of the current peak / power peak and waveform of the GPU or xPU in different operating states.

[0117] See also Figure 1 As shown, I0 represents the current consumption level when the computing power module is in a dormant state, and I1 represents the current consumption level when the computing power module is in a working state but without computing power tasks loaded. At this time, the overall current fluctuation is small, and the power output of the robot system can be regarded as a light load state, a low power consumption state, a small load, and a computing load-free working state; I2 and I3 represent different computing power tasks. Because a large number of internal Tensor core calculations are called and enabled, they can be regarded as high power consumption states and computing load working states. At this time, as the computing tasks are loaded, the current shows a very steep increase, and as the computing load changes, the current drops very steeply. Figure 1 The current consumption of the computing power module in different situations shown clearly reflects that the power supply unit of the computing power module needs to have very strong current / power peak redundancy capability and transient response capability.

[0118] Based on this, see Figure 2 As shown, the embodiment of the present application provides a power control method for a robot, and the specific process of the method includes but is not limited to the following:

[0119] Step 200: Continuously sample to obtain a workload value of a preset computing power module, wherein the computing power module is used to process AI computing power work and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes an instantaneous value of the current passing through the computing power module, or an instantaneous value of the power consumed by the computing power module.

[0120] In an embodiment of the present application, a sampling circuit can be set for a computing power module used to process AI computing power in an existing power supply system to achieve real-time sampling of the workload value of the computing power module, wherein the computing power module can be any one of a GPU, a TPU, and an XPU; the workload value includes an instantaneous value of the current passing through the computing power module, or an instantaneous value of the power consumed by the computing power module.

[0121] In some embodiments of the present application, a sampling circuit for a computing power module can be set for a DC controller of the robot. Through the sampling circuit, the DC controller can sample the workload value of the computing power module. Then, the DC controller sends the sampled workload value to the microcontroller unit (MCU) / central processing unit (CPU) of the robot through a bus such as PWBus, and the MCU / CPU executes subsequent steps 210 to 220 based on the sampled workload value to obtain a target PWM frequency. Then, when executing step 230, the MCU / CPU sends the target PWM frequency to the DC controller, and the DC controller outputs a PWN signal with the target PWM frequency to the multi-channel power supply unit to adjust the current output to the computing power module in the current sampling period.

[0122] In some other embodiments of the present application, a sampling circuit for the computing power module can be set for the MCU / CPU of the robot. In this case, the MCU / CPU directly obtains the workload value of the computing power module through continuous sampling based on the sampling circuit, and adjusts the current output to the computing power module in the current sampling period by executing subsequent steps 210 to 230.

[0123] It should be noted that the present application does not limit the form of the sampling circuit specifically for the computing power module.

[0124] The following only takes the sampling of the workload value of the aforementioned computing power module through the DC controller, and the workload value is the instantaneous value of the current passing through the computing power module as an example to explain this solution in detail.

[0125] For example, Figure 3 As shown in FIG. 1 , the computing power module is powered by four power supply units connected in parallel. For each power supply unit, the DC controller outputs a PWM signal to the driver and MOSFET included in the power supply unit, and uses a synchronous rectification method to adjust the output current and voltage of the power supply unit. The four power supply units are connected in parallel to obtain an I out Output to the computing power module to provide power supply for the computing power module.

[0126] In some embodiments of the present application, real-time sampling can be performed Figure 3 The R of the computing module side is shown ref The voltage is calculated in real time to obtain the instantaneous value of the current I passing through the computing power module. sample , which is the workload value.

[0127] like Figure 3 As shown, each power supply unit is also equipped with a phase current sampling, namely Figure 3The resistors set at each power supply unit are shown as R1, R2, R3, and R4, respectively. Based on the resistors, corresponding CS signals can be sampled in real time, respectively as CS1 signal, CS2 signal, CS3 signal, and CS4 signal, which can reflect the phase current of the corresponding power supply unit. Based on this, it can be determined which power supply unit has failed to assist in adjusting the PWM frequency. For example, if a power supply unit fails, the power supply unit can be locked at this time. When adjusting the PWM frequency, adjust the power supply units of other roads to ensure the effectiveness of power supply control.

[0128] It should be noted that Figure 3 The circuit shown is only an example, and the present application is not limited to Figure 3 The sampling circuit shown implements the power control method of the present solution.

[0129] Step 210: Based on the obtained workload values, determine the current operating status of the computing power module.

[0130] In some embodiments of the present application, when executing step 210, refer to Figure 4 This can be achieved by executing the following steps:

[0131] Step 2101: Compare the obtained workload values ​​with the first threshold, the second threshold and the third threshold in sequence, wherein the second threshold is greater than the first threshold and less than the third threshold.

[0132] based on Figure 1 The waveform of the current of the computing power module in the running state is shown. In some embodiments of the present application, three thresholds can be preset, wherein the first threshold can be Figure 1 The second threshold value can be Figure 1 The third threshold value can be Figure 2 I2±Δ2 is shown, where Δ0, Δ1, and Δ2 are deviation values ​​respectively.

[0133] It should be noted that if the sampled workload value is an instantaneous power value, since the voltage of each power supply unit can be obtained based on the circuit structure, the current passing through the computing power module can be calculated through the relationship between power and current. Based on the above content, the current operating status of the computing power module can be determined, and the PWM frequency of each power supply unit of the computing power module can be dynamically adjusted based on the current operating status.

[0134] Of course, in practical applications, it can also be based on Figure 1 The current waveform shown, the voltage through the computing power module, is calculated based on Figure 1The power thresholds corresponding to I0, I1, I2, and I3 are shown, so that the current operating status of the computing power module is determined based on the power thresholds and the instantaneous power values ​​obtained by sampling, which will not be repeated here.

[0135] Step 2102: If it is determined that each workload value is less than or equal to the first threshold, it is determined that the current operating state of the computing power module is a sleep state.

[0136] In some embodiments of the present application, by executing step 2101, each workload value is compared with the first threshold, the second threshold and the third threshold in turn. If it is determined that each workload value is less than or equal to the first threshold, step 2102 is executed.

[0137] Step 2103: If it is determined that each workload value is greater than the first threshold and less than or equal to the second threshold, the current operating state of the computing power module is determined to be the first operating state.

[0138] In some embodiments of the present application, by executing step 2101, each workload value is compared with the first threshold, the second threshold and the third threshold in turn. If it is determined that each workload value is greater than the first threshold, the relationship between each workload value and the second threshold is further determined. If it is determined that each workload value is less than or equal to the second threshold, step 2103 is executed.

[0139] Step 2104: If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, the current operating state of the computing power module is determined to be the second operating state.

[0140] In some embodiments of the present application, by executing step 2101, each workload value is compared with the first threshold, the second threshold, and the third threshold in sequence. If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, step 2103 is executed, wherein the power consumption of the first working state is less than the power consumption of the second working state. For example, the first working state is the aforementioned light load state, a small load, no calculation load working state; the second working state is the aforementioned high power consumption state, and the current will show a very steep increase with the loading of the calculation task, and the current will show a very steep decrease with the change of the calculation load.

[0141] Step 220: Determine a target PWM frequency within a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload.

[0142] In some embodiments of the present application, different methods of determining the target PWM frequency in the current sampling period are adopted according to different current operating states. Specifically, according to different operating states, the following three situations are included but not limited to:

[0143] In case 1, if the current operating state is the sleep state, at least one power supply unit among the multiple power supply units is turned off, and the first fixed PWM frequency is determined as the target PWM frequency.

[0144] In some embodiments of the present application, since the workload of the computing power module is very small in the sleep state, the switching loss of the MOSFET accounts for a high proportion of the power consumption loss of the whole machine. At this time, shutting down a part of the power supply units can reduce the power consumption loss of the whole machine, and at the same time determining the target PWM frequency as the first fixed PWM frequency, wherein the first fixed PWM frequency is a lower switching frequency, which can effectively reduce useless power consumption and reduce heat generation.

[0145] Optionally, due to the limitations of the driver and MOSFET, the minimum on-time that can be controlled by the digital DC controller and the PWM frequency set by the MCU / CPU are also limited. Therefore, in some embodiments of the present application, based on the physical parameters of the switch MOSFET, DC controller, driver and inductor, and based on the measured empirical data, the output PWM frequency conversion adjustment range is set to 200KHz-2MHz, where 200KHz is the lower limit of the PWM frequency and 2MHz is the upper limit of the PWM frequency. It should be noted that in actual applications, specific settings can be made based on actual needs, actual circuit structures, etc., and this application does not make specific limitations.

[0146] Optionally, in some embodiments of the present application, the first fixed PWM frequency can be selected as a lower limit of the PWM frequency, such as 200 KHz.

[0147] In this state, a lower switching frequency and a scheduling strategy that shuts down some power supply units can be used to effectively reduce useless power consumption, save battery capacity, and thus improve battery life.

[0148] Case 2: if the current operating state is the first working state, at least one of the multiple power supply units that is turned off is turned on, and the second fixed PWM frequency is determined as the target PWM frequency, wherein the first fixed PWM frequency is less than the second fixed PWM frequency.

[0149] In some embodiments of the present application, in the first working state, that is, the computing power module is in a light load or no computing load working state, it is still a relatively low consumption state. Figure 1The waveform shown shows that the current envelope is relatively stable, which is similar to the sleep state in the aforementioned case 1. Since the computing power module has been converted from the sleep state to the first working state, there is a hysteresis effect on the transient response due to the inductive energy storage and capacitor filtering at the output end. At this time, one or some power supply units that are closed in the sleep state are turned on, and a second fixed PWM frequency that is greater than the first fixed PWM frequency is determined as the target PWM frequency. This can ensure that once a large-load computing task is started later, the load capacity can be immediately increased, thereby significantly improving the transient response capability of the current.

[0150] Optionally, in some embodiments of the present application, the second fixed PWM frequency can be selected as 500KHz.

[0151] In this state, a lower switching frequency can also reduce switching losses and effectively improve battery life.

[0152] Since excessive PWM frequency adjustment will bring an adverse effect, namely electromagnetic interference (EMI), high-frequency switching may generate more electromagnetic interference. At the same time, the PWM output of the DC controller in the sleep state and the first working state is always fixed. At the fixed output frequency, the output energy may be superimposed, which may cause interference to surrounding electronic equipment and systems. Therefore, in some embodiments of the present application, for the sleep state and the first working state, a PWM scrambling working mode is adopted to effectively reduce the energy pulse peaks in EMI.

[0153] Optionally, in some embodiments of the present application, if there are at least two turned-on power supply units in the sleep state or the first working state, and the target PWM frequency includes a plurality of sub-PWM frequencies arranged in a time sequence, that is, multiple adjustment methods are used in one sampling period for control smoothness, then in the aforementioned case 1 and case 2, refer to Figure 5 As shown, the frequency of each sub-PWM can be obtained as follows:

[0154] Step 500: determining the fixed PWM frequency as the first sub-PWM frequency among the multiple sub-PWM frequencies;

[0155] Step 510: Perform the following operations for each non-first sub-PWM frequency: generate a random number for any non-first sub-PWM frequency; determine the sum of the PWM frequency mapped from the random number and the fixed PWM frequency as the non-first sub-PWM frequency; wherein, if the current operating state is the sleep state, the fixed PWM frequency is the first fixed PWM frequency, and if the current operating state is the first working state, the fixed PWM frequency is the second fixed PWM frequency.

[0156] Case three, if the current operating state is the second working state, the target PWM frequency within the current sampling period is determined according to the changing trend of each workload value and the expected PWM frequency corresponding to the next sampling time, wherein the expected PWM frequency is determined based on the waveform obtained by fitting each workload value and the next sampling time.

[0157] In some embodiments of the present application, Figure 1 The different peak values ​​of the current sampling thresholds I2 and I3 shown generally indicate different requirements for current / power consumption under different computing load tasks, but the waveforms of their basic current envelopes are similar.

[0158] In this state, let's assume that the DC controller samples the workload value of the computing power module in real time. In case three, the DC controller samples the load current / power (i.e., the aforementioned workload value) of the computing power module in real time, and the MCU / CPU obtains these data from the DC controller through the PMBus bus, calculates and analyzes the change trend, and determines the target PWM frequency based on the change trend, so as to output a PWM signal with the target PWM frequency to each power supply unit of the computing power module, control the driver and MOSFET to accurately control the power output, and achieve both effective current transient response capability and reduced switching energy consumption.

[0159] In some embodiments of the present application, in case three, refer to Figure 6 As shown, the corresponding functions can be implemented by executing the following steps including but not limited to:

[0160] Step 600: Based on each workload value, fit the workload waveform of the computing power module.

[0161] In some embodiments of the present application, the workload waveform of the computing power module can be fitted by executing step 600. Assuming that the fitted waveform is as follows: Figure 7 As shown, Figure 7 A schematic diagram of the waveform of the envelope curve drawn by the current sampled at different times is shown.

[0162] See also Figure 7 As shown, the instantaneous current values ​​sampled at sampling times such as ts00-ts05, ts10-ts12, and ts20-ts22 in the second working state are respectively shown, which are Is00-Is05, Is12-Is10, and Is20-Is22, respectively, and the waveforms shown are obtained by fitting based on the instantaneous current values. Figure 7 As shown in the figure, the waveform has two trends: increasing (rising) and decreasing (falling), and will reach local maximum and minimum values ​​at the same time, such as Figure 7 It should be noted that, Figure 7The sampling frequency shown is fixed. Of course, in actual control, the sampling frequency can also be dynamically adjusted following the current change. This application only illustrates the sampling frequency as a fixed mode, but it should be noted that this example does not make a specific limitation on this.

[0163] Optionally, when the sampling frequency is fixed, based on empirical data, the sampling frequency can be selected to be 5 to 6 times the upper limit of the PWM frequency. For example, if the upper limit of the PWM frequency is 2 MHz, the sampling frequency can be 10 MHz.

[0164] Step 610: predicting an expected PWM frequency according to the workload waveform and the next sampling time, wherein if the expected PWM frequency is greater than the upper limit value of the PWM frequency, the expected PWM frequency is the upper limit value of the PWM frequency.

[0165] In some embodiments of the present application, after executing step 600 to obtain the waveform, when executing step 610, the expected PWN frequency corresponding to the next sampling moment is predicted based on the obtained workload waveform.

[0166] In some embodiments, the robot may store Figure 1 The PWM frequency associated with the waveform shown, then, when executing step 610, the PWM frequency at the next sampling moment, that is, the aforementioned expected PWM frequency, can be predicted based on the workload waveform and the pre-stored PWM frequency.

[0167] Step 620: Based on each workload value, determine the difference between two adjacent workload values ​​in sequence.

[0168] In some embodiments of the present application, when executing step 620, the difference between two adjacent workloads is calculated in sequence based on each workload value.

[0169] For example, see Figure 7 As shown, the sampled current is the instantaneous value. It is assumed that the instantaneous current values ​​obtained by sampling at the six sampling moments ts00-ts05 are recorded as Is00-Is05 respectively.

[0170] Then, the difference between two adjacent instantaneous current values ​​among the above six instantaneous current values ​​is determined, which is recorded as ΔI00=Is01-Is00, ΔI01=Is02-Is01, ΔI02=Is03-Is02, ΔI03=Is04-Is03, ΔI04=Is05-Is04 respectively.

[0171] Step 630: Determine the target PWM frequency in the current sampling period according to the determined change trends of the differences and the expected PWM frequency.

[0172] In some embodiments of the present application, when executing step 630, different methods are used to determine the target PWM frequency according to different change trends, including but not limited to the following methods:

[0173] Method 1: If the change trend of each difference is an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the first adjustment amplitude, wherein the first adjustment amplitude is greater than the preset value.

[0174] In some embodiments of the present application, in method 1, the target PWM frequency is determined based on the change trend of each difference determined in step 620 and the first adjustment amplitude.

[0175] Still Figure 7 For example, if ΔI00, ΔI01, ΔI02, ΔI03, and ΔI04 are all positive values, it is determined that the change trend of each difference is an increasing trend. Then, based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the first adjustment amplitude, the target PWM frequency in the current sampling period is determined, where the preset value can be 0.

[0176] Optionally, the aforementioned target PWM frequency may include multiple sub-PWM frequencies, that is, the PWM frequency can be adjusted multiple times within a sampling period. Then, the multiple sub-PWM frequencies can be determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the first adjustment amplitude.

[0177] Optionally, in some embodiments, after determining that the changing trend is an increasing trend, the size relationship between ΔI04, ΔI03, ΔI02, and ΔI00 can be further determined. Based on the size relationship, it is determined whether the increasing trend is a fast increasing trend or a slow increasing trend. Accordingly, the first adjustment amplitude includes a first sub-adjustment amplitude and a second sub-adjustment amplitude, and the first sub-adjustment amplitude is greater than the second sub-adjustment amplitude.

[0178] In some embodiments, if ΔI04>ΔI03>ΔI02>ΔI00, it is determined that the change trend is a rapid increasing trend, that is, the current of the computing power module is rising rapidly and needs to be accelerated and adjusted in advance. Then, the DC controller is configured to quickly step the output to increase the PWM frequency, that is, the first sub-adjustment amplitude is used to determine the target PWM frequency to effectively improve the peak redundancy capability. If the target PWM frequency includes 5 sub-PWM frequencies, the target PWM frequency in the previous sampling period is used as a reference, and the first sub-adjustment amplitude is used to determine the 5 sub-PWM frequencies, wherein the value of the first sub-PWM frequency is the sum of the current PWM frequency and the first sub-adjustment amplitude, and the value of the second sub-PWM frequency is the sum of the first sub-PWM frequency and the first sub-adjustment amplitude, and so on. The value of the fifth sub-PWM frequency is the expected PWM frequency, wherein the current PWM frequency is the target PWM frequency last adjusted in the previous sampling period.

[0179] In other embodiments, if ΔI0i (i=0, 1, ...4) starts to decrease, it is determined that the change trend is a slowly increasing trend, that is, the current of the computing power module is slowly rising. At this time, the second sub-adjustment amplitude can be used to determine the target PWM frequency. The value determination method can refer to the above description and will not be repeated here.

[0180] Method 2: If the change trend of each difference is a decreasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the second adjustment amplitude, wherein the second adjustment amplitude is less than the preset value.

[0181] In some embodiments of the present application, in the second method, the target PWM frequency is determined based on the change trend of each difference determined in step 620 and the second adjustment amplitude.

[0182] For example, see Figure 7 As shown, the sampling is the instantaneous current value, assuming that the instantaneous current values ​​obtained by sampling at the three sampling moments ts10-ts12 are recorded as Is10-Is12. Then, the difference between two adjacent instantaneous current values ​​among the above three instantaneous current values ​​is determined, which is recorded as ΔI10=Is11-Is10, ΔI11=Is12-Is11.

[0183] In some embodiments, if ΔI10 and ΔI11 are both negative values, the change trend is determined to be a decreasing trend, and the target PWM frequency is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the second adjustment amplitude, wherein the preset value can be 0.

[0184] Since the decreasing trend also includes a fast decreasing trend and a slow decreasing trend, after determining that the changing trend is a decreasing trend, it is further determined whether it is a fast decreasing trend or a slow decreasing trend. Optionally, the second adjustment range includes a third sub-adjustment range and a fourth sub-adjustment range, and the third sub-adjustment range is greater than the fourth sub-adjustment range.

[0185] In some embodiments, if |ΔI11|>|ΔI10|, it is determined that the change trend is a rapidly decreasing trend, that is, the current of the computing power module is rapidly decreasing and needs to be decelerated and adjusted in advance. Then, the DC controller can be configured to quickly step the output to reduce the PWM frequency, that is, the fourth sub-adjustment amplitude is used to determine the target PWM frequency. The value-taking method can refer to the above description and will not be repeated here.

[0186] In other embodiments, if |ΔI11|<|ΔI10|, it is determined that the change trend is a slowly decreasing trend, that is, the current of the computing power module is slowly decreasing, and the third sub-adjustment amplitude can be used to determine the target PWM frequency. The value determination method can refer to the above description and will not be repeated here.

[0187] Method three, if the changing trend of each difference changes from an increasing trend to a decreasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the third adjustment amplitude, wherein the third adjustment amplitude is smaller than the first adjustment amplitude and greater than the preset value.

[0188] For example, see Figure 7 As shown, the sampling is the instantaneous current value. Assume that the instantaneous current values ​​sampled at the three sampling moments ts20-ts22 are Is20-Is22. Then, the difference between two adjacent instantaneous current values ​​among the above three instantaneous current values ​​is determined, which is recorded as ΔI20=Is21-Is20, ΔI21=Is22-Is21.

[0189] In some embodiments, if ΔI20 is a positive value and ΔI21 is a negative value, it is determined that the change trend is from an increasing trend to a decreasing trend, that is, there should be a maximum value around the current Is21 sampling point of the computing power module. At this time, the current value decreases, but it is still very large overall. Then, the DC controller can be configured to use a fast step output to increase the PWM frequency to the maximum value, and it can be maintained for a period of time, where the preset value can be 0.

[0190] Method four, if the changing trend of each difference changes from a decreasing trend to an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and the fourth adjustment amplitude, wherein the fourth adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value.

[0191] In the embodiment of the present application, if ΔI20 is a negative value and ΔI21 is a positive value, it is determined that the change trend is from a decreasing trend to an increasing trend, that is, there should be a minimum value around the current Is21 sampling point of the computing power module, and the DC controller can be configured to quickly step the output to enter the accelerated adjustment to ensure that once a large load computing task is started, the load capacity can be immediately increased, thereby showing the transient response capability of the increased current.

[0192] Through the processing of the above four methods, the PWM frequency can be dynamically adjusted based on the workload value of the computing power module sampled in real time. In this way, according to the change trend, overshoot operations can be made in advance, thereby improving the current output capacity and ensuring the power consumption requirements of the load under intensive computing tasks. It should be noted that the specific values ​​of the first adjustment amplitude, the second adjustment amplitude, the third adjustment amplitude and the fourth adjustment amplitude can be set according to actual needs, and this application does not make specific limitations.

[0193] like Figure 8 As shown, a schematic diagram of a 4-way PWM frequency conversion adjustment of an optional DC controller predictive output is shown. Figure 8 As shown, in low power consumption such as sleep state or the first working state, the DC controller outputs PWM signals to the four power supply units in sequence, and each PWM signal is sequentially expanded in timing. During acceleration adjustment such as the increasing trend in the second working state, the time interval of the PWM signal output by the DC controller to each power supply unit is gradually reduced, so as to achieve fast step output to increase the PWM frequency until the Imax current is output to the computing power module and maintained; during deceleration adjustment such as the increasing trend in the second working state, the time interval of the PWM signal output by the DC controller to each power supply unit is gradually increased, so as to achieve fast step output to reduce the PWM frequency; with the completion of the computing task, the PWM signal returns to low power consumption such as sleep state or the first working state.

[0194] In actual applications, the multi-channel power supply unit of the computing power module may fail. Figure 3 The phase current sampling circuit of each power supply unit shown can receive the phase current sampling value of each power supply unit in real time. If it is determined that a fault exists in any power supply unit based on the phase current sampling value of the power supply unit, the target PWM frequency of the remaining power supply units in the multiple power supply units is determined based on the change trend of each difference and the expected PWM frequency, so as to adjust the PWM frequency of the remaining power supply units. In this way, the specific PWM frequency adjustment strategy is determined in combination with the sampling of each phase current, so as to facilitate better adjustment of the current output to the computing power module to ensure the current peak redundancy capability and transient response capability.

[0195] Step 230: Output a PWM signal with a target PWM frequency to the multi-channel power supply unit to adjust the current output to the computing power module during the current sampling period.

[0196] In some embodiments of the present application, when executing step 230, refer to Fig. 9 As shown, the specific steps that can be performed include but are not limited to the following:

[0197] Step 2301: Determine whether to perform grouping adjustment on the multiple power supply units, if yes, execute step 2302, if not, execute step 2303;

[0198] Step 2302: Divide the multiple power supply units into multiple groups, and output PWM signals with target PWM frequencies to the power supply units in the multiple groups in sequence;

[0199] Step 2303: Outputting PWM signals with target PWM frequencies to multiple power supply units in sequence.

[0200] In some embodiments of the present application, since the computing power module is powered by multiple power supply units connected in parallel, at least two of the multiple power supply units can be merged into one to perform group PWM frequency adjustment, or each power supply unit can be used as a separate one to perform PWM frequency adjustment.

[0201] For example, Figure 3 The 4-way power supply unit shown can combine the 1st and 2nd roads into one road, and combine the 3rd and 4th roads into one road for PWM frequency adjustment. In this way, when the computing load fluctuation frequency of the computing power module is high, the bandwidth of the PWM frequency can be made larger. Although the accuracy of some adjustments is lost, the two groups can effectively separate the output current ripple, and the noise is also lower. Further, the harmonic components in the output waveform are also reduced, and the noise level is reduced. Fig.10 As shown, a schematic diagram of 4-way PWM frequency conversion adjustment of the grouped DC controller predicted output is shown.

[0202] Furthermore, the above-mentioned group adjustment method makes the high-bandwidth high-frequency DC control generally have a faster dynamic response, and can adjust the output voltage more quickly to adapt to the computing load changes of the computing power module, thereby reducing the noise caused by transient response.

[0203] In other embodiments, when determining the target PWM frequency, it is also necessary to comprehensively consider the performance requirements, energy-saving requirements, cost constraints, and EMI requirements of the DC power supply system to select a better PWM frequency.

[0204] Based on the same inventive concept, see Fig.11 As shown, in an embodiment of the present application, a power control device for a robot is provided, comprising:

[0205] The sampling module 1110 is used for continuously sampling to obtain the workload value of the preset computing power module;

[0206] A first determination module 1120, configured to determine a current operating state of the computing power module based on each obtained workload value;

[0207] A second determination module 1130 is used to determine a target pulse width modulation (PWM) frequency in a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload;

[0208] An output module 1140, configured to output a PWM signal with the target PWM frequency to the multi-channel power supply unit, so as to adjust the current output to the computing power module in the current sampling period;

[0209] Among them, the computing power module is used to process artificial intelligence AI computing power work, and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes the instantaneous value of the current passing through the computing power module, or the instantaneous value of the power consumed by the computing power module.

[0210] In a possible implementation, the first determining module 1120 is specifically configured to:

[0211] Comparing the obtained workload values ​​with a first threshold, a second threshold, and a third threshold in sequence, respectively, wherein the second threshold is greater than the first threshold and less than the third threshold;

[0212] If it is determined that each workload value is less than or equal to the first threshold, determining that the current operating state of the computing power module is a dormant state;

[0213] If it is determined that each workload value is greater than the first threshold and less than or equal to the second threshold, determining that the current operating state of the computing power module is the first operating state;

[0214] If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, the current operating state of the computing power module is determined to be the second operating state.

[0215] In a possible implementation, the second determining module 1130 is specifically configured to:

[0216] If the current operating state is the sleep state, shutting down at least one of the multiple power supply units, and determining the first fixed PWM frequency as the target PWM frequency;

[0217] If the current operating state is the first working state, at least one of the multiple power supply units that is turned off is turned on, and a second fixed PWM frequency is determined as the target PWM frequency, wherein the first fixed PWM frequency is less than the second fixed PWM frequency;

[0218] If the current operating state is the second operating state, the target PWM frequency within the current sampling period is determined according to the change trend of the workload values ​​and the expected PWM frequency corresponding to the next sampling moment, wherein the expected PWM frequency is determined based on the waveform obtained by fitting the workload values ​​and the next sampling moment.

[0219] In a possible implementation, the second determining module 1130 is specifically configured to:

[0220] Based on the workload values, fitting the workload waveform of the computing power module;

[0221] The expected PWM frequency is predicted according to the workload waveform and the next sampling time, wherein if the expected PWM frequency is greater than an upper limit value of the PWM frequency, the expected PWM frequency is the upper limit value of the PWM frequency;

[0222] Based on the workload values, determining the difference between two adjacent workload values ​​in sequence;

[0223] The target PWM frequency in the current sampling period is determined according to the determined variation trends of the differences and the expected PWM frequency.

[0224] In a possible implementation, the second determining module 1130 is specifically configured to:

[0225] If the variation trend of the differences is an increasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a first adjustment amplitude, wherein the first adjustment amplitude is greater than a preset value;

[0226] If the variation trend of the differences is a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a second adjustment amplitude, wherein the second adjustment amplitude is less than the preset value;

[0227] If the change trend of the differences changes from an increasing trend to a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a third adjustment amplitude, wherein the third adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value;

[0228] If the changing trend of the differences changes from a decreasing trend to an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a fourth adjustment amplitude, wherein the fourth adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value.

[0229] In a possible implementation, if there are at least two turned-on power supply units in the sleep state or the first working state, and the target PWM frequency includes a plurality of sub-PWM frequencies arranged in time sequence, each sub-PWM frequency is obtained in the following manner:

[0230] Determine the fixed PWM frequency as the first sub-PWM frequency among the multiple sub-PWM frequencies;

[0231] For each non-first sub-PWM frequency, perform the following operations:

[0232] For any sub-PWM frequency other than the first one, generate a random number;

[0233] Determine the sum of the PWM frequency obtained by mapping the random number and the fixed PWM frequency as any non-first sub-PWM frequency;

[0234] If the current operating state is the sleep state, the fixed PWM frequency is the first fixed PWM frequency; if the current operating state is the first working state, the fixed PWM frequency is the second fixed PWM frequency.

[0235] In a possible implementation, the output module 1140 is specifically configured to:

[0236] Determining whether to perform grouping adjustment on the multiple power supply units;

[0237] If yes, the multiple power supply units are divided into multiple groups, and the PWM signals with the target PWM frequency are output to the power supply units in the multiple groups in sequence;

[0238] If not, the PWM signals with the target PWM frequency are output to the multiple power supply units in sequence.

[0239] Based on the same inventive concept, see Fig.12As shown, an embodiment of the present application provides a robot, comprising:

[0240] The computing module 121 is used to process artificial intelligence AI computing work;

[0241] Multiple power supply units 122 connected in parallel, used to supply power to the computing power module, each power supply unit including a driver and a MOSFET;

[0242] Memory 123, used to store computer programs or instructions;

[0243] The processor 124 is used to execute the computer program or instructions in the memory 123 so that any one of the methods in the above embodiments is executed.

[0244] The processor 124 may include one or more central processing units (CPUs) or digital processing units, etc., for executing computer programs or instructions in the memory 123, so that any one of the methods in the above embodiments is executed;

[0245] It should be noted that the specific connection medium between the memory 123 and the processor 124 is not limited in the embodiment of the present application. Fig.12 In the embodiment, the memory 123 and the processor 124 are connected via a bus 125, and the connection between other components is only for schematic illustration and is not intended to be limiting. The bus 125 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.12 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0246] Based on the same inventive concept, the embodiment of the present application provides a computer-readable storage medium. When the instructions in the storage medium are executed by a processor, the processor is enabled to execute any one of the methods in the above embodiments. Since the principle of solving the problem by the above computer-readable storage medium is similar to a power control method for a robot, the implementation of the above computer-readable storage medium can refer to the implementation of the method, and the repeated parts will not be repeated.

[0247] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0248] The present application is described with reference to the flowchart and / or block diagram of the method, device (system), and computer program product according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the process and / or box in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one process or multiple processes in the flowchart and / or one box or multiple boxes in the block diagram.

[0249] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0250] These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0251] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A power control method for a robot, characterized in that: include: Continuously sample to obtain the workload value of the preset computing power module; Determine the current operating state of the computing power module based on the obtained workload values; Determine a target pulse width modulation (PWM) frequency within a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload; Outputting a PWM signal with the target PWM frequency to the multi-channel power supply unit to adjust the current output to the computing power module in the current sampling period; Among them, the computing power module is used to process artificial intelligence AI computing power work, and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes the instantaneous value of the current passing through the computing power module, or the instantaneous value of the power consumed by the computing power module.

2. The method according to claim 1, characterized in that The determining the current operating state of the computing power module based on each obtained workload value includes: Comparing the obtained workload values ​​with a first threshold, a second threshold, and a third threshold in sequence, respectively, wherein the second threshold is greater than the first threshold and less than the third threshold; If it is determined that each workload value is less than or equal to the first threshold, determining that the current operating state of the computing power module is a dormant state; If it is determined that each workload value is greater than the first threshold and less than or equal to the second threshold, determining that the current operating state of the computing power module is the first operating state; If it is determined that each workload value is greater than the second threshold and less than or equal to the third threshold, the current operating state of the computing power module is determined to be the second operating state.

3. The method according to claim 2, characterized in that Determining a target pulse width modulation (PWM) frequency within a current sampling period according to the current operating state includes: If the current operating state is the sleep state, shutting down at least one of the multiple power supply units, and determining the first fixed PWM frequency as the target PWM frequency; If the current operating state is the first working state, at least one of the multiple power supply units that is turned off is turned on, and a second fixed PWM frequency is determined as the target PWM frequency, wherein the first fixed PWM frequency is less than the second fixed PWM frequency; If the current operating state is the second operating state, the target PWM frequency within the current sampling period is determined according to the change trend of the workload values ​​and the expected PWM frequency corresponding to the next sampling moment, wherein the expected PWM frequency is determined based on the waveform obtained by fitting the workload values ​​and the next sampling moment.

4. The method according to claim 3, characterized in that The determining the target PWM frequency in the current sampling period according to the change trend of each workload value and the expected PWM frequency corresponding to the next sampling moment includes: Based on the workload values, fitting the workload waveform of the computing power module; The expected PWM frequency is predicted according to the workload waveform and the next sampling time, wherein if the expected PWM frequency is greater than an upper limit value of the PWM frequency, the expected PWM frequency is the upper limit value of the PWM frequency; Based on the workload values, determining the difference between two adjacent workload values ​​in sequence; The target PWM frequency in the current sampling period is determined according to the determined variation trend of each difference and the expected PWM frequency.

5. The method according to claim 4, characterized in that Determining the target PWM frequency in the current sampling period according to the determined change trends of the differences and the expected PWM frequency includes: If the variation trend of the differences is an increasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a first adjustment amplitude, wherein the first adjustment amplitude is greater than a preset value; If the variation trend of the differences is a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a second adjustment amplitude, wherein the second adjustment amplitude is less than the preset value; If the change trend of the differences changes from an increasing trend to a decreasing trend, determining the target PWM frequency in the current sampling period based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a third adjustment amplitude, wherein the third adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value; If the changing trend of the differences changes from a decreasing trend to an increasing trend, the target PWM frequency in the current sampling period is determined based on the expected PWM frequency, the target PWM frequency in the previous sampling period, and a fourth adjustment amplitude, wherein the fourth adjustment amplitude is smaller than the first adjustment amplitude and larger than the preset value.

6. The method according to claim 3, characterized in that If there are at least two turned-on power supply units in the sleep state or the first working state, and the target PWM frequency includes a plurality of sub-PWM frequencies arranged in time sequence, each sub-PWM frequency is obtained in the following manner: Determine the fixed PWM frequency as the first sub-PWM frequency among the multiple sub-PWM frequencies; For each non-first sub-PWM frequency, perform the following operations: For any sub-PWM frequency other than the first one, generate a random number; Determine the sum of the PWM frequency obtained by mapping the random number and the fixed PWM frequency as any non-first sub-PWM frequency; If the current operating state is the sleep state, the fixed PWM frequency is the first fixed PWM frequency; if the current operating state is the first working state, the fixed PWM frequency is the second fixed PWM frequency.

7. The method according to any one of claims 1 to 5, characterized in that: The step of outputting the PWM signal with the target PWM frequency to the multi-channel power supply unit comprises: Determining whether to perform grouping adjustment on the multiple power supply units; If yes, the multiple power supply units are divided into multiple groups, and the PWM signals with the target PWM frequency are output to the power supply units in the multiple groups in sequence; If not, the PWM signals with the target PWM frequency are output to the multiple power supply units in sequence.

8. A power control device for a robot, characterized in that: include: A sampling module is used to continuously sample and obtain the workload value of a preset computing power module; A first determination module, configured to determine a current operating state of the computing power module based on each obtained workload value; A second determination module is used to determine a target pulse width modulation (PWM) frequency within a current sampling period according to the current operating state, wherein the target PWM frequency is positively correlated with the workload; An output module, used to output a PWM signal with the target PWM frequency to the multi-channel power supply unit, so as to adjust the current output to the computing power module in the current sampling period; Among them, the computing power module is used to process artificial intelligence AI computing power work, and is powered by multiple power supply units connected in parallel, each power supply unit includes a driver and a MOSFET, and the workload value includes the instantaneous value of the current passing through the computing power module, or the instantaneous value of the power consumed by the computing power module.

9. A robot, characterized in that: include: Computing module, used to process artificial intelligence (AI) computing work; Multiple power supply units connected in parallel are used to supply power to the computing power module, each power supply unit comprising a driver and a MOSFET; Memory, used to store computer programs or instructions; A processor is used to execute the computer program or instructions in the memory so that the method according to any one of claims 1 to 7 is executed.

10. A computer-readable storage medium, characterized in that: When the instructions in the storage medium are executed by a processor, the processor is enabled to execute the method according to any one of claims 1 to 7.

Images