Robot servo system self-adaptive energy-saving control method based on task power spectrum

Abstract

The invention belongs to the technical field of robot intelligent control and energy management crossing, and particularly relates to a robot servo system self-adaptive energy-saving control method based on a task power spectrum, which comprises the following steps: establishing a task power spectrum model based on multi-modal sensor data fusion, and quantifying robot task dynamic characteristics; establishing a multi-physical field coupling loss model comprising a motor loss model and a power device loss model, cooperatively optimizing the bus voltage and the PWM switching frequency based on the power spectrum model and the loss model, and determining an optimal working point parameter combination; smooth parameter switching is realized by adopting a variable parameter self-adaptive predictive control algorithm; system global energy efficiency optimization and control performance guarantee are realized through a three-layer control architecture comprising a cognitive layer, a decision layer and an execution layer and a four-domain collaborative optimization mechanism in a time domain, a frequency domain, an energy domain and a heat domain. According to the method, the prediction precision under extreme working conditions is improved by adopting multi-physical coupling modeling, the performance degradation of a traditional method is avoided, and the robust performance of the system is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the cross technical field of robot intelligent control and energy management, and specifically relates to a robot servo system adaptive energy-saving control method based on task power spectrum. BACKGROUND

[0002] The development of robot servo control technology has experienced three important stages: the early open-loop position control stage, which focuses on solving the positioning accuracy problem; the classic closed-loop control stage, which realizes high-precision trajectory tracking through PI / PID controllers; and the current intelligent control stage, which introduces advanced control algorithms to improve system robustness. However, existing intelligent control theories mainly focus on improving control accuracy and dynamic response performance, and there is a clear lack of attention to system energy efficiency optimization, especially in the energy management of variable working conditions, which lacks systematic theoretical guidance.

[0003] Existing robot servo systems generally adopt a "peak-oriented" fixed parameter design paradigm. Although this design philosophy guarantees the performance margin in extreme conditions, it forms a systematic energy efficiency trap in normal operation: the motor is long-term deviated from the optimal working point, while the drive circuit is also forced to work in the low efficiency interval, and the coupling effect of the two leads to exponential growth of the overall energy consumption, which seriously restricts the endurance and work efficiency of mobile robots.

[0004] Through in-depth analysis, it is found that traditional servo control faces the theoretical challenge of "performance-energy efficiency-robustness" impossible triangle. Specifically, it is shown that:

[0005] Theoretical lack of energy efficiency optimization: lack of mathematical model describing the relationship between task dynamic characteristics and system energy consumption, unable to establish the optimal mapping relationship from task demand to system parameters. Existing researches are mostly limited to single parameter optimization level, and no theoretical framework of multi-parameter collaborative optimization has been formed.

[0006] Parameter coupling mechanism is unknown: there is a complex nonlinear coupling relationship between bus voltage, PWM frequency and load characteristics, and traditional linear control theory is difficult to handle such multi-variable and strongly coupled optimization problems. Especially in the parameter dynamic switching process, the stability and transient characteristics of the system lack theoretical guarantee.

[0007] Multi-scale time response mismatch: task recognition, parameter adjustment and control response work on different time scales (milliseconds to seconds), and there is a lack of unified time series modeling method, which leads to poor system coordination and even control oscillation.

[0008] Current technologies primarily focus on improving efficiency in single aspects. For example, variable frequency speed control technology only focuses on motor efficiency, switching power supply technology only optimizes power electronic conversion efficiency, and control algorithm research only pursues tracking accuracy. This piecemeal approach ignores the collaborative optimization potential of the servo system as a whole and lacks deep integration with the characteristics of higher-level tasks.

[0009] Therefore, there is an urgent need in this field for a theoretical innovation that breaks through the shackles of traditional thinking. It should be able to start from the essence of the task, establish an energy efficiency optimization theoretical system covering the entire chain of perception, decision-making, and execution, and realize the fundamental transformation of robot servo systems from "passive adaptation" to "intelligent prediction". Summary of the Invention

[0010] The purpose of this invention is to provide an adaptive energy-saving control method for robot servo systems based on task power spectrum. Its core innovation lies in establishing a quantitative mapping relationship between task dynamic characteristics and system energy efficiency, and constructing a "three-layer, four-domain" collaborative optimization control architecture. First, the task power spectrum model is used to predict the instantaneous power demand and energy efficiency performance of the system under arbitrary voltage-frequency combinations. Then, a servo motor control optimization model is established using a multi-physics coupling loss model. Combined with the proposed "three-layer, four-threshold" control architecture, the optimal voltage-frequency combination control strategy under task operation mode is realized. This addresses the core technical problems of existing robot servo systems, such as the lack of theoretical guidance for energy efficiency optimization under varying operating conditions, the lack of a collaborative mechanism for parameter adjustment, and the fundamental contradiction between control performance and energy consumption.

[0011] The technical solution adopted by this invention to achieve the above objectives is: an adaptive energy-saving control method for robot servo systems based on task power spectrum, comprising the following steps:

[0012] Step S1: Establish a task power spectrum model T(ω) based on multimodal sensor data fusion to quantify the dynamic characteristics of the robot task;

[0013] Step S2: Establish a multi-physics coupling loss model that includes a motor loss model and a power device loss model. Based on the task power spectrum model and the multi-physics coupling loss model, collaboratively optimize the bus voltage V and the PWM switching frequency f to determine the optimal operating point parameter combination. ;

[0014] Step S3: Using the variable parameter adaptive predictive control algorithm, based on the optimal operating point parameter combination (V,f) and the real-time state of the system, determine the time-varying parameter vector θ(k) of the system, establish a state space model of parameter changes, and then solve for the control input vector u(k);

[0015] Step S4: Based on the output of steps S1 to S3, perform four-domain collaborative optimization through a three-layer control architecture and execute a dynamic adaptive energy-saving control loop.

[0016] Step S1 specifically includes:

[0017] Establish a three-dimensional mapping model of task-power-parameters for:

[0018] ;

[0019] Where T(ω) is the task power spectral density function, describing the power requirements of different frequency components. Let H(ω) be the power transfer function of the system under a given bus voltage V and PWM frequency f, and let H(ω) be the frequency response characteristic of the motor-load system.

[0020] The task power spectral density function T(ω) is divided into different parameterized models according to different types of robot tasks, including: periodic task model, impact task model, stochastic task model and quasi-static task power spectral model.

[0021] When a humanoid robot walks or a robotic arm carries materials, the periodic task model is as follows:

[0022] ;

[0023] in, Let be the amplitude of the i-th harmonic. It is an exponential function. For frequency, For the fundamental frequency, Let be the bandwidth of the i-th harmonic. Let be the variance of the i-th harmonic component. Based on power consumption;

[0024] When a humanoid robot jumps or a robotic arm performs machining operations, the impact of the task is as follows:

[0025] ;

[0026] in, The amplitude of the fundamental frequency. angular frequency reciprocal, This is the scaling factor of the base frequency. For frequency, Let j be the amplitude of the j-th harmonic. It is an impulse function, specifically the Dirac delta function. The harmonic frequency;

[0027] When a humanoid robot explores freely, the random task model is as follows:

[0028] ;

[0029] in, Represents the Lorentz spectrum. For low-frequency power levels, For frequency, The cutoff frequency, Noise level;

[0030] Quasi-static task power spectrum model:

[0031] ;

[0032] in, The amplitude of the first harmonic. It is an exponentially decaying function. Angular frequency, For characteristic frequencies, This refers to noise at a given frequency.

[0033] The motor loss model is established as follows:

[0034] Establish a motor loss model:

[0035] ;

[0036] in, For copper loss model, For iron loss model, Mechanical wear and tear;

[0037] a. The copper loss model is as follows:

[0038] ;

[0039] in, Indicates motor current. This refers to the winding temperature-dependent resistance in the copper loss model. Indicates the winding temperature;

[0040] b. The iron loss model adopts the Bertotti model, that is:

[0041] ;

[0042] in, Indicates the frequency of change in magnetic flux density. Indicates the maximum magnetic flux density. Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. Indicates the accessory loss coefficient. The empirical coefficients typically vary widely, ranging from 1.6 to 2.

[0043] c. Mechanical wear and tear:

[0044] ;

[0045] in, Expressed as motor speed, Expressed as load torque, The coefficient of viscous friction is... This is the wind resistance loss coefficient. The coefficient of friction is related to the load.

[0046] Establish a loss model for power electronic devices, specifically as follows:

[0047] ;

[0048] in, Indicates conduction loss. The current flowing through the device, It is the junction temperature; as the temperature increases, the conduction loss increases. This represents switching losses, which are voltage ( ),frequency( ) and junction temperature ( The function of ) This represents the gate drive loss, which is the switching frequency. The function.

[0049] Step S3 specifically includes:

[0050] Step S3-1: Couple the electrical equations and mechanical equations of the motor to define state variables and obtain the state vector. Includes electrical and mechanical subsystem variables, namely:

[0051] ;

[0052] in, , These are the q-axis and d-axis currents of the motor, respectively. The mechanical angular velocity of the motor. This refers to the motor's rotation angle;

[0053] Step S3-2: Define the control input and the control input vector. for:

[0054] ;

[0055] in, , These are the modulation voltages for the d and q axes, respectively;

[0056] Step S3-3: Based on the mechanical subsystem, define the parameter vector to obtain the time-varying parameter vector. for:

[0057] ;

[0058] in, For inductors that are voltage- and PWM frequency-dependent, This refers to the winding temperature-dependent resistance in the copper loss model. For rotational inertia, The damping coefficient;

[0059] Step S3-4: Based on steps S3-1 to S3-3, Euler forward difference discretization is used to establish a state-space model with varying parameters as follows:

[0060] ;

[0061] in, For time-varying parameter vectors, For state variables, For parameter control input, The state matrix, For the input matrix, For noise interference;

[0062] The system's time-varying parameter vector is updated according to a nonlinear function, that is:

[0063] ;

[0064] Step S3-5: Design a parameter change trajectory optimizer, and solve the optimal parameter switching trajectory through predictive control theory to achieve continuous optimization of control performance rather than step switching.

[0065] In steps S3-5, the design parameter change trajectory optimizer specifically includes:

[0066] (a) Establishing a prediction model:

[0067] ;

[0068] in, This is the system output prediction at step k. It is a state variable. It is about controlling the increment. , It is a prediction matrix;

[0069] (b) Determine the optimization objective:

[0070] ;

[0071] in, It is the reference output trajectory. It is the predicted output. It is about controlling the increment. It is a control input. , , It is a weight matrix used to weigh the costs of tracking error, control change, and control amplitude.

[0072] (c) Add constraints:

[0073] ;

[0074] ;

[0075] in, It is the control input at the i-th future time. It controls the increment of the input. It controls the length of the time domain; upper and lower limits are used to limit the output amplitude and rate of change of the controller.

[0076] (d) Online parameter updates:

[0077] Execute in each control cycle k:

[0078] Update the time-varying parameter vector based on sensor data. ;

[0079] Recalculate the state matrix A(θ(k)), input matrix B(θ(k)) and prediction matrices F and G, apply the optimal control increment to the system, and update the parameter vector. .

[0080] In step S4, the three-layer control architecture includes:

[0081] Cognitive layer: used for establishing task power spectrum models and task identification, realizing multimodal task identification and power spectrum prediction;

[0082] Decision layer: Used to perform collaborative optimization of multi-physics coupling loss models, and to achieve multi-objective optimization to solve for the optimal voltage-frequency combination (V,f);

[0083] Execution layer: Used to execute variable parameter adaptive predictive control algorithms to achieve distributed parameter synchronization and real-time control;

[0084] The four-domain collaborative optimization includes:

[0085] Time-domain optimization: Based on the prediction of future task power requirements, control parameters are pre-adjusted to adapt to the dynamic changes of the task over time.

[0086] Frequency domain optimization: Based on the power transfer function S(ω,V,f) of the system under a given bus voltage V and PWM frequency f, the harmonic characteristics and energy distribution of the system under different frequency components are optimized by dynamically adjusting the PWM switching frequency f.

[0087] Energy domain optimization: Based on the motor loss model and the power electronic device loss model, global scheduling and efficiency optimization of system energy flow are achieved;

[0088] Thermal Domain Optimization: Junction Temperature Based on Power Electronic Device Loss Model Parameters are coupled with thermal management strategies and control performance to achieve dynamic thermal load balancing and thermal runaway prevention.

[0089] Step S4 specifically includes the following steps:

[0090] Step S4.1: Based on the task power spectrum model established in step S1, predict the future task power requirements and generate corresponding robot motion mode instructions;

[0091] Step S4.2: Based on the motion mode command generated in step S4.1, call the optimal operating point parameter combination (V, f) determined in step S2, and calculate the real-time control input vector u(k) based on the variable parameter adaptive predictive control algorithm in step S3.

[0092] Step S4.3: Input the control input vector u(k) calculated in step S4.2 into the servo system to synchronously adjust the bus voltage and PWM switching frequency, thereby achieving coordinated optimization in the time domain, frequency domain, energy domain and thermal domain.

[0093] The control system of the robot servo system based on the task power spectrum adaptive energy-saving control method includes:

[0094] The intelligent energy management controller integrates a multimodal deep learning inference engine and a multi-objective optimization solver to generate a task power spectrum model T(ω) based on sensor data and determine the optimal control parameters.

[0095] The intelligent energy management controller adopts an FPGA dual-core processing architecture, wherein the first core is configured to run multimodal deep learning task recognition and power spectrum prediction algorithms, and the second core is configured to execute multi-objective optimization and variable parameter adaptive predictive control algorithms; the intelligent energy management controller also integrates a DSP coprocessor for high-frequency power spectrum analysis and real-time signal processing;

[0096] The servo driver, connected to the intelligent energy management controller via a communication bus, is used to receive control commands and drive the motor, supporting a wide range of bus voltage V input from 24V to 150V and an adjustable PWM switching frequency f from 5kHz to 25kHz.

[0097] The servo driver includes:

[0098] The current detection module has a detection accuracy of no less than 0.1% FSR and is used for real-time feedback of motor current.

[0099] Temperature feedback module is used to monitor the junction temperature of power devices and realize dynamic on-resistance compensation;

[0100] SiC MOSFET power modules are designed for efficient operation over a wide voltage range and receive frequency and voltage adjustment commands via the communication bus.

[0101] A multi-physics sensor network includes a data fusion processing unit and connected current sensors, temperature sensors, encoders, and inertial measurement units; it integrates data from the inertial measurement unit, joint encoders, and current sensors, and generates system state estimates through Kalman filtering and load estimation algorithms, which are then transmitted to the intelligent energy management controller.

[0102] The communication bus adopts the CAN-FD communication bus, which connects the intelligent energy management controller, each servo driver and sensor node. It is used for downlink transmission of control commands and parameter configurations to each servo driver, and uplink transmission of sensor data and system status to the intelligent energy management controller; to realize timing synchronization and coordinated control between each servo unit.

[0103] The present invention has the following beneficial effects and advantages:

[0104] This invention, through a deep integration of theoretical innovation and engineering implementation, can significantly improve system energy efficiency and effectively ensure system control performance. The use of multi-physics coupling modeling improves prediction accuracy under extreme conditions, avoids performance degradation of traditional methods, and greatly enhances the system's robustness. For the first time, a quantitative relationship between task characteristics and servo system energy efficiency is established, providing an important theoretical foundation for related fields. Attached Figure Description

[0105] Figure 1 This is a diagram of the overall architecture of the control system of the present invention;

[0106] Figure 2 This is a hardware architecture diagram of the control system in this embodiment;

[0107] Figure 3 This is a hardware system architecture diagram of the variable bus voltage in this embodiment;

[0108] Figure 4 This is a flowchart of the control method of the control system of the present invention;

[0109] Figure 5 This is a diagram showing the motion mode state switching of the present invention. Detailed Implementation

[0110] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0111] like Figure 4 The diagram shown is a flowchart of the control method of the control system of the present invention. The adaptive energy-saving control method for robot servo systems based on task power spectrum of the present invention includes the following steps:

[0112] Step S1: Establish a task power spectrum model T(ω) based on multimodal sensor data fusion to quantify the dynamic characteristics of the robot task;

[0113] The task power spectral density function T(ω) is divided into different parameterized models according to different types of robot tasks, including: periodic task model, impact task model, stochastic task model and quasi-static task power spectral density model;

[0114] like Figure 5 The diagram shown is a motion mode state switching diagram of the present invention. Figure 5 Each motion mode corresponds to the power spectrum model switching relationship of different types of robot tasks (periodic, impulsive, random, quasi-static) in step S1;

[0115] S1-1: The specific task power spectrum model is as follows:

[0116] a) Periodic task model (when the humanoid robot walks or the robotic arm carries materials):

[0117] ;

[0118] in, Let be the amplitude of the i-th harmonic. It is an exponential function. For frequency, For the fundamental frequency, Let be the bandwidth of the i-th harmonic. Let be the variance of the i-th harmonic component. Based on power consumption;

[0119] b) Impactful task model (when a humanoid robot jumps or a robotic arm performs processing):

[0120] ;

[0121] in, The amplitude of the fundamental frequency. angular frequency reciprocal, This is the scaling factor of the base frequency. For frequency, Let j be the amplitude of the j-th harmonic. It is an impulse function, specifically the Dirac delta function. The harmonic frequency;

[0122] c) Random task model (when the humanoid robot is freely exploring):

[0123] ;

[0124] in, Represents the Lorentz spectrum. For low-frequency power levels, For frequency, The cutoff frequency, Noise level;

[0125] d) Quasi-static task power spectrum model:

[0126] ;

[0127] in, The amplitude of the first harmonic. It is an exponentially decaying function. Angular frequency, For characteristic frequencies, This refers to noise at a given frequency.

[0128] S1-2: Specific implementation of task power spectrum modeling:

[0129] This invention proposes for the first time the concept of task power spectrum and establishes a three-dimensional mapping model of task-power-parameter. for:

[0130]

[0131] Where T(ω) is the task power spectral density function, describing the power requirements of different frequency components. Let H(ω) be the power transfer function of the system under a given bus voltage V and PWM frequency f, and let H(ω) be the frequency response characteristic of the motor-load system.

[0132] This model can predict the instantaneous power demand and energy efficiency of a system under any combination of voltage and frequency, achieving feedforward optimization of parameters rather than traditional passive feedback regulation.

[0133] S1-3: Multimodal sensor data fusion:

[0134] This invention employs a multimodal fusion network to achieve effective integration of heterogeneous sensor data;

[0135] Inertial data processing: fusing accelerometer and gyroscope data using a Kalman filter:

[0136]

[0137] Joint state sensing: estimating joint load based on encoder feedback and current sensor data.

[0138]

[0139] Step S2: Establish a multi-physics coupling loss model that includes a motor loss model and a power device loss model. Based on the power spectrum model and the multi-physics coupling loss model, optimize the bus voltage V and the PWM switching frequency f in a coordinated manner to determine the optimal operating point parameter combination (V,f).

[0140] To address the oversimplification of traditional loss analysis, this invention establishes an accurate loss model that considers the coupling of multiple physical fields including electromagnetic, thermal, and mechanical fields:

[0141] S2-1: Motor loss model:

[0142]

[0143] in, For copper loss model, For iron loss model, Mechanical wear and tear;

[0144] Copper loss model:

[0145]

[0146] in, Indicates motor current. This refers to the winding temperature-dependent resistance in the copper loss model. Indicates the winding temperature;

[0147] Iron loss model: Bertotti model is adopted;

[0148]

[0149] in, The frequency (Hz) represents the change in magnetic flux density. This represents the maximum magnetic flux density (T). Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. Indicates the accessory loss coefficient. The empirical coefficients typically vary widely, ranging from 1.6 to 2.

[0150] Mechanical loss:

[0151]

[0152] in, Expressed as motor speed, Expressed as load torque, The coefficient of viscous friction is... This is the wind resistance loss coefficient. The coefficient of friction is related to the load.

[0153] S2-2: Establishing a loss model for power electronic devices:

[0154]

[0155] In the formula: Indicates conduction loss. The current flowing through the device, It is the junction temperature; an increase in temperature will increase conduction losses. This represents switching losses, which are voltage ( ),frequency( ) and junction temperature ( The function of ) This represents the gate drive loss, which is the switching frequency. The temperature dependence coefficient in the model is determined by fitting finite element simulation and experimental data, enabling accurate prediction of electronic device losses.

[0156] S2-3: Implementation based on multiphysics coupling model optimization:

[0157] This invention employs a hierarchical optimization strategy, first optimizing global energy efficiency, and then optimizing local performance:

[0158] Main objective function:

[0159]

[0160] in, Let be the objective function. The power spectrum function, As a weighting factor, For performance penalties;

[0161] Once the objective function is established, various optimization algorithms can be used to optimize and solve it.

[0162] Step S3: Using the variable parameter adaptive predictive control algorithm, based on the optimal operating point parameter combination (V,f) and the real-time state of the system, determine the time-varying parameter vector θ(k) of the system, establish a state space model of parameter changes, and then solve for the control input vector u(k);

[0163] Implementation of Variable Parameter Adaptive Predictive Control Algorithm

[0164] (1) Establishment of the state-space model

[0165] Consider the coupling of the electrical equations and the mechanical equations of the motor:

[0166] Electrical subsystem:

[0167]

[0168]

[0169] The two equations above describe the current dynamics in the motor model, where: and These are the two axial currents of the motor (currents in the orthogonal axis directions). It is an inductance related to the motor parameters (voltage and frequency). It is a resistor. It is the electric angular velocity of the motor. and It is the voltage component of the motor.

[0170] Mechanical subsystem:

[0171]

[0172]

[0173] in, It is the moment of inertia. It is the mechanical angular velocity of the motor. It is the torque constant. It is the q-axis current. It is the damping coefficient. It is the load torque. It refers to the motor's rotation angle.

[0174] (2) Establishment of the state-space model for parameter variation

[0175] Step S3-1: Couple the electrical equations and mechanical equations of the motor to define state variables and obtain the state vector. Includes electrical and mechanical subsystem variables, namely:

[0176] ;

[0177] in, , These are the q-axis and d-axis currents of the motor, respectively. The mechanical angular velocity of the motor. This refers to the motor's rotation angle;

[0178] Step S3-2: Define the control input and the control input vector. for:

[0179] ;

[0180] in, , These are the modulation voltages for the d and q axes, respectively;

[0181] Step S3-3: Based on the mechanical subsystem, define the parameter vector to obtain the time-varying parameter vector. for:

[0182] ;

[0183] in, For inductors that are voltage- and PWM frequency-dependent, This refers to the winding temperature-dependent resistance in the copper loss model. For rotational inertia, The damping coefficient;

[0184] Step S3-4: Based on steps S3-1 to S3-3, Euler forward difference discretization is used to establish a state-space model with varying parameters as follows:

[0185] ;

[0186] in, For time-varying parameter vectors, For state variables, For parameter control input, The state matrix, For the input matrix, For noise interference;

[0187] The system's time-varying parameter vector is updated according to a nonlinear function, that is:

[0188] ;

[0189] The design of the parameter change trajectory optimizer uses predictive control theory to solve for the optimal parameter switching trajectory, thereby achieving continuous optimization of control performance rather than step switching.

[0190] (3) Design of parameter change trajectory optimizer

[0191] Step S3-5: Design a parameter change trajectory optimizer. Using predictive control theory, solve for the optimal parameter switching trajectory to achieve continuous optimization of control performance rather than abrupt switching. Specifically:

[0192] (a) Establishing a prediction model:

[0193] ;

[0194] in, This is the system output prediction at step k. It is a state variable. It is about controlling the increment. , It is a prediction matrix;

[0195] (b) Determine the optimization objective:

[0196] ;

[0197] in, It is the reference output trajectory. It is the predicted output. It is about controlling the increment. It is a control input. , , It is a weight matrix used to weigh the costs of tracking error, control change, and control amplitude.

[0198] (c) Add constraints:

[0199] ;

[0200] ;

[0201] in, It is the control input at the i-th future time. It controls the increment of the input. It controls the length of the time domain; upper and lower limits are used to limit the output amplitude and rate of change of the controller.

[0202] (d) Online parameter updates:

[0203] Execute in each control cycle k:

[0204] Update the time-varying parameter vector based on sensor data. ;

[0205] Recalculate the state matrix A(θ(k)), input matrix B(θ(k)) and prediction matrices F and G, apply the optimal control increment to the system, and update the parameter vector. .

[0206] Step S4: Based on the output of steps S1 to S3, perform four-domain collaborative optimization through a three-layer control architecture and execute a dynamic adaptive energy-saving control loop.

[0207] (1) Three-tier control architecture:

[0208] Cognitive layer: Task understanding and power spectrum prediction based on multimodal deep learning;

[0209] It integrates information from multiple sensors, including vision, IMU, and joint encoders; constructs the task power spectrum T(ω) in real time; and predicts power demand within future time windows.

[0210] Decision-making level: Parameter configuration strategy for multi-objective optimization;

[0211] With maximizing energy efficiency as the primary objective and control performance as the constraint, the optimal voltage-frequency combination (V,f) is solved, and a parameter switching timing strategy is generated.

[0212] Execution layer: Real-time parameter adjustment for distributed collaboration;

[0213] Control commands are synchronously issued via the CAN-FD bus; variable parameter adaptive predictive control is used to ensure switching stability; and millisecond-level parameter response is achieved.

[0214] (2) Four-domain collaborative optimization:

[0215] Time-domain optimization: Based on the prediction of future task power requirements, control parameters are pre-adjusted to adapt to the dynamic changes of the task over time.

[0216] Frequency domain optimization: Based on the power transfer function S(ω,V,f) of the system under a given bus voltage V and PWM frequency f, the harmonic characteristics and energy distribution of the system under different frequency components are optimized by dynamically adjusting the PWM switching frequency f.

[0217] Energy domain optimization: Based on the motor loss model and the power electronic device loss model, global scheduling and efficiency optimization of system energy flow are achieved;

[0218] Thermal Domain Optimization: Junction Temperature Based on Power Electronic Device Loss Model Parameters are coupled with thermal management strategies and control performance to achieve dynamic thermal load balancing and thermal runaway prevention.

[0219] (3) Specific steps of the dynamic adaptive energy-saving control cycle:

[0220] Step S4.1: Based on the task power spectrum model established in step S1, predict the future task power requirements and generate corresponding robot motion mode instructions;

[0221] Step S4.2: Based on the motion mode command generated in step S4.1, call the optimal operating point parameter combination (V, f) determined in step S2, and calculate the real-time control input vector u(k) based on the variable parameter adaptive predictive control algorithm in step S3.

[0222] Step S4.3: Input the control input vector u(k) calculated in step S4.2 into the servo system to synchronously adjust the bus voltage and PWM switching frequency, thereby achieving coordinated optimization in the time domain, frequency domain, energy domain and thermal domain.

[0223] like Figure 1 As shown, the control system of this invention adopts a hierarchical distributed architecture, with a core intelligent energy management controller integrating a deep learning inference engine and a multi-objective optimization solver. This controller communicates with each servo unit in real time via a high-speed CAN-FD bus, forming an intelligent network with global optimization capabilities. The control system of this invention, based on the task power spectrum-based adaptive energy-saving control method for robot servo systems, includes:

[0224] The intelligent energy management controller integrates a multimodal deep learning inference engine and a multi-objective optimization solver to generate a task power spectrum and determine the optimal control parameters based on sensor data.

[0225] The intelligent energy management controller adopts an FPGA dual-core processing architecture, where the first core is configured to run multimodal deep learning task recognition and power spectrum prediction algorithms, and the second core is configured to execute multi-objective optimization and variable parameter predictive control algorithms; the intelligent energy management controller also integrates a DSP coprocessor for high-frequency power spectrum analysis and real-time signal processing;

[0226] The servo driver, connected to the intelligent energy management controller via a communication bus, receives control commands and drives the motor, supporting a wide-range bus voltage input of 24V-150V and an adjustable PWM switching frequency of 5kHz-25kHz. ;

[0227] The servo driver includes:

[0228] The current detection module has a detection accuracy of no less than 0.1% FSR and is used for real-time feedback of motor current.

[0229] Temperature feedback module is used to monitor the junction temperature of power devices and realize dynamic on-resistance compensation;

[0230] SiC MOSFET power modules are designed for efficient operation over a wide voltage range and receive frequency and voltage adjustment commands via the communication bus.

[0231] A multi-physics sensor network includes: a data fusion processing unit and connected current sensors, temperature sensors, encoders, and inertial measurement units; used to integrate data from the inertial measurement unit, joint encoders, and current sensors, and generate system state estimates through Kalman filtering and load estimation algorithms, which are then transmitted to the intelligent energy management controller.

[0232] The communication bus adopts the CAN-FD communication bus, which connects the intelligent energy management controller, each servo driver and sensor node. It is used for downlink transmission of control commands and parameter configurations to each servo driver, and uplink transmission of sensor data and system status to the intelligent energy management controller; to realize timing synchronization and coordinated control between each servo unit.

[0233] Example:

[0234] This embodiment uses a humanoid robot servo system as an example to specifically implement the control method of the present invention:

[0235] 1. Implementation of the task power spectrum:

[0236] Data is collected through a multimodal sensor network, and a deep learning model is used to identify the current task type (such as walking, jumping, free exploration, etc.) and call the corresponding task power spectrum model.

[0237] Dynamic power demand prediction is achieved by updating power spectrum parameters based on real-time sensor data.

[0238] 2. Implementation of multiphysics coupling model optimization:

[0239] Based on the motor loss model and the power device loss model, the total system loss under different (V,f) combinations is calculated.

[0240] Use multi-objective optimization algorithms (such as genetic algorithms or gradient descent) to solve for the optimal (V,f) combination, ensuring that energy efficiency is maximized while performance constraints are met.

[0241] 3. Implementation of variable parameter adaptive predictive control algorithm:

[0242] Establish a state-space model of the motor and design a predictive controller.

[0243] The parameter vector is updated online, and the control parameters are smoothly switched through the trajectory optimizer to ensure system stability and response speed.

[0244] 4. Architecture selected in this embodiment

[0245] like Figure 2 The diagram shows the selection of components in the architecture of this embodiment. The above algorithm is deployed on an FPGA+DSP hardware platform and connected to the servo driver via a CAN-FD bus.

[0246] (1) Hardware architecture of intelligent energy management controller, including:

[0247] Main control chip: adopts FPGA dual-core parallel processing; among which, core 1 runs the deep learning inference engine and task recognition algorithm; core 2 executes multi-objective optimization and real-time control algorithm;

[0248] Coprocessor: Integrates a DSP chip for processing high-frequency signals and power spectrum analysis;

[0249] Communication interfaces, including:

[0250] CAN-FD interface: 4 channels, supporting a maximum data rate of 8Mbps;

[0251] Ethernet interface: Gigabit Ethernet, used for parameter updates and remote monitoring;

[0252] Wireless interface: WiFi 6;

[0253] (2) Wide voltage servo driver design, such as Figure 3 The diagram shown is a hardware system architecture diagram of the variable bus voltage in this embodiment.

[0254] Power stage design: Employs SiC MOSFETs, supporting a wide input voltage range of 24V-150V;

[0255] On-resistance: <10mΩ@25°C; Switching frequency: 5kHz-25kHz adjustable; Current detection accuracy: 0.1%FSR.

[0256] The energy-saving effect and control performance were verified through testing on a real robot platform. Experimental results show that the method significantly reduces system energy consumption and extends robot endurance while ensuring control performance.

[0257] In summary, this invention provides an adaptive energy-saving control method and system for robot servo systems based on task power spectrum. Its core lies in breaking through the limitations of traditional servo control that only focuses on accuracy and response, and pioneering a new paradigm for energy efficiency optimization starting from the "essence of the task." Specifically, the contributions of this invention are mainly reflected in three aspects:

[0258] Theoretical innovation: The concept of "task power spectrum" was proposed for the first time, a quantitative mapping model between task dynamic characteristics and system energy consumption was established, and combined with the multi-physics coupling loss model, a solid theoretical foundation was laid for global energy efficiency optimization.

[0259] Architectural innovation: A three-layer control architecture of "cognition-decision-execution" and a four-domain collaborative optimization mechanism of "time domain-frequency domain-energy domain-thermal domain" were constructed, realizing a closed loop of the entire link from task identification and parameter optimization to real-time control, ensuring the overall optimization of the system under complex and variable working conditions.

[0260] Algorithm innovation: A "variable parameter adaptive predictive control algorithm" was designed. Through a state-space model of parameter changes and a trajectory optimizer, the control parameters are smoothly and quickly switched between different optimal operating points, which fundamentally solves the contradiction between energy efficiency improvement and control performance degradation.

[0261] Through the aforementioned innovations, this invention successfully transforms the servo system from a "passively adapting to load" energy consumption mode to an "actively predicting task" energy efficiency mode. Practice has proven that this method can significantly reduce the overall energy consumption of the robot servo system, effectively extend the mobile robot's endurance, and improve its thermal safety and reliability, while strictly ensuring the system's dynamic performance and stability. It possesses significant theoretical value and broad engineering application prospects in the fields of robot intelligent control and green energy management.

[0262] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0263] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. An adaptive energy-saving control method for robot servo systems based on task power spectrum, characterized in that, Includes the following steps: Step S1: Establish a task power spectrum model T(ω) based on multimodal sensor data fusion to quantify the dynamic characteristics of the robot task; Step S2: Establish a multi-physics coupling loss model that includes a motor loss model and a power device loss model. Based on the task power spectrum model and the multi-physics coupling loss model, collaboratively optimize the bus voltage V and the PWM switching frequency f to determine the optimal operating point parameter combination. ; Step S3: Using the variable parameter adaptive predictive control algorithm, based on the optimal operating point parameter combination (V,f) and the real-time state of the system, determine the time-varying parameter vector θ(k) of the system, establish a state space model of parameter changes, and then solve for the control input vector u(k); Step S4: Based on the output of steps S1 to S3, perform four-domain collaborative optimization through a three-layer control architecture and execute a dynamic adaptive energy-saving control loop.

2. The adaptive energy-saving control method for robot servo systems based on task power spectrum according to claim 1, characterized in that, Step S1 specifically includes: Establish a three-dimensional mapping model of task-power-parameters for: ； Where T(ω) is the task power spectral density function, describing the power requirements of different frequency components. Let H(ω) be the power transfer function of the system under a given bus voltage V and PWM frequency f, and let H(ω) be the frequency response characteristic of the motor-load system.

3. The adaptive energy-saving control method for robot servo systems based on task power spectrum according to claim 1, characterized in that, The task power spectral density function T(ω) is divided into different parameterized models according to different types of robot tasks, including: periodic task model, impact task model, stochastic task model and quasi-static task power spectral model. When a humanoid robot walks or a robotic arm carries materials, the periodic task model is as follows: ； in, Let be the amplitude of the i-th harmonic. It is an exponential function. For frequency, For the fundamental frequency, Let be the bandwidth of the i-th harmonic. Let be the variance of the i-th harmonic component. Based on power consumption; When a humanoid robot jumps or a robotic arm performs machining operations, the impact of the task is as follows: ； in, The amplitude of the fundamental frequency. angular frequency reciprocal, This is the scaling factor of the base frequency. For frequency, Let j be the amplitude of the j-th harmonic. It is an impulse function, specifically the Dirac delta function. Harmonic frequencies; When a humanoid robot explores freely, the random task model is as follows: ； in, Represents the Lorentz spectrum. For low-frequency power levels, For frequency, The cutoff frequency, Noise level; Quasi-static task power spectrum model: ； in, The amplitude of the first harmonic. It is an exponentially decaying function. Angular frequency, For characteristic frequencies, This refers to noise at a given frequency.

4. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 1, characterized in that, The motor loss model is established as follows: Establish a motor loss model: ； in, For copper loss model, For iron loss model, Mechanical wear and tear; a. The copper loss model is as follows: ； in, Indicates motor current. This refers to the winding temperature-dependent resistance in the copper loss model. Indicates the winding temperature; b. The iron loss model adopts the Bertotti model, that is: ； in, Indicates the frequency of change in magnetic flux density. Indicates the maximum magnetic flux density. Indicates the hysteresis loss coefficient. This represents the eddy current loss coefficient. Indicates the accessory loss coefficient. The empirical coefficients typically vary widely, ranging from 1.6 to 2. c. Mechanical wear and tear: ； in, Expressed as motor speed, Expressed as load torque, The coefficient of viscous friction is... This is the wind resistance loss coefficient. The coefficient of friction is related to the load.

5. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 1, characterized in that, Establish a loss model for power electronic devices, specifically as follows: ； in, Indicates conduction loss. The current flowing through the device, It is the junction temperature; as the temperature increases, the conduction loss increases. This represents switching losses, which are voltage losses. ,frequency and junction temperature The function, This represents the gate drive loss, which is the switching frequency. The function.

6. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 1, characterized in that, Step S3 specifically includes: Step S3-1: Couple the electrical equations and mechanical equations of the motor to define state variables and obtain the state vector. Includes electrical and mechanical subsystem variables, namely: ； in, , These are the q-axis and d-axis currents of the motor, respectively. The mechanical angular velocity of the motor. This refers to the motor's rotation angle; Step S3-2: Define the control input and the control input vector. for: ； in, , These are the modulation voltages for the d and q axes, respectively; Step S3-3: Based on the mechanical subsystem, define the parameter vector to obtain the time-varying parameter vector. for: ； in, For inductors that are voltage- and PWM frequency-dependent, This refers to the winding temperature-dependent resistance in the copper loss model. For rotational inertia, The damping coefficient; Step S3-4: Based on steps S3-1 to S3-3, Euler forward difference discretization is used to establish a state-space model with varying parameters as follows: ； in, For time-varying parameter vectors, For state variables, For parameter control input, The state matrix, For the input matrix, For noise interference; The system's time-varying parameter vector is updated according to a nonlinear function, that is: ； Step S3-5: Design a parameter change trajectory optimizer, and solve for the optimal control input vector u(k) through predictive control theory to achieve continuous optimization of control performance rather than step switching.

7. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 6, characterized in that, In steps S3-5, the design parameter change trajectory optimizer specifically includes: (a) Establishing a prediction model: ； in, This is the system output prediction at step k. It is a state variable. It is about controlling the increment. , It is a prediction matrix; (b) Determine the optimization objective: ； in, It is the reference output trajectory. It is the predicted output. It is about controlling the increment. It is a control input. , , It is a weight matrix used to weigh the costs of tracking error, control change, and control amplitude. (c) Add constraints: ； ； in, It is the control input at the i-th future time. It controls the increment of the input. It controls the length of the time domain; upper and lower limits are used to limit the output amplitude and rate of change of the controller. (d) Online parameter updates: Execute in each control cycle k: Update the time-varying parameter vector based on sensor data. ; Recalculate the state matrix A(θ(k)), input matrix B(θ(k)) and prediction matrices F and G, solve for the optimization objective, apply the optimal control increment to the system, and update the parameter vector. .

8. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 1, characterized in that, In step S4, the three-layer control architecture includes: Cognitive layer: used for establishing task power spectrum models and task identification, realizing multimodal task identification and power spectrum prediction; Decision layer: Used to perform collaborative optimization of multi-physics coupling loss models, and to achieve multi-objective optimization to solve for the optimal voltage-frequency combination (V,f); Execution layer: Used to execute variable parameter adaptive predictive control algorithms to achieve distributed parameter synchronization and real-time control; The four-domain collaborative optimization includes: Time-domain optimization: Based on the prediction of future task power requirements, control parameters are pre-adjusted to adapt to the dynamic changes of the task over time. Frequency domain optimization: Based on the power transfer function S(ω,V,f) of the system under a given bus voltage V and PWM frequency f, the harmonic characteristics and energy distribution of the system under different frequency components are optimized by dynamically adjusting the PWM switching frequency f. Energy domain optimization: Based on the motor loss model and the power electronic device loss model, global scheduling and efficiency optimization of system energy flow are achieved; Thermal Domain Optimization: Junction Temperature Based on Power Electronic Device Loss Model Parameters are coupled with thermal management strategies and control performance to achieve dynamic thermal load balancing and thermal runaway prevention.

9. The adaptive energy-saving control method for a robot servo system based on task power spectrum according to claim 1, characterized in that, Step S4 specifically includes the following steps: Step S4.1: Based on the task power spectrum model established in step S1, predict the future task power requirements and generate corresponding robot motion mode instructions; Step S4.2: Based on the motion mode command generated in step S4.1, call the optimal operating point parameter combination (V, f) determined in step S2, and calculate the real-time control input vector u(k) based on the variable parameter adaptive predictive control algorithm in step S3. Step S4.3: Input the control input vector u(k) calculated in step S4.2 into the servo system to synchronously adjust the bus voltage and PWM switching frequency, thereby achieving coordinated optimization in the time domain, frequency domain, energy domain and thermal domain.

10. The control system of the adaptive energy-saving control method for robot servo systems based on task power spectrum according to claims 1 to 9, characterized in that, include: The intelligent energy management controller integrates a multimodal deep learning inference engine and a multi-objective optimization solver to generate a task power spectrum model T(ω) based on sensor data and determine the optimal control parameters. The intelligent energy management controller adopts an FPGA dual-core processing architecture, wherein the first core is configured to run multimodal deep learning task recognition and power spectrum prediction algorithms, and the second core is configured to execute multi-objective optimization and variable parameter adaptive predictive control algorithms; the intelligent energy management controller also integrates a DSP coprocessor for high-frequency power spectrum analysis and real-time signal processing; The servo driver, connected to the intelligent energy management controller via a communication bus, is used to receive control commands and drive the motor, supporting a wide range of bus voltage V input from 24V to 150V and an adjustable PWM switching frequency f from 5kHz to 25kHz. The servo driver includes: The current detection module has a detection accuracy of no less than 0.1% FSR and is used for real-time feedback of motor current. Temperature feedback module is used to monitor the junction temperature of power devices and realize dynamic on-resistance compensation; SiC MOSFET power modules are designed for efficient operation over a wide voltage range and receive frequency and voltage adjustment commands via the communication bus. A multi-physics sensor network includes a data fusion processing unit and connected current sensors, temperature sensors, encoders, and inertial measurement units; it integrates data from the inertial measurement unit, joint encoders, and current sensors, and generates system state estimates through Kalman filtering and load estimation algorithms, which are then transmitted to the intelligent energy management controller. The communication bus adopts the CAN-FD communication bus, which connects the intelligent energy management controller, each servo driver and sensor node. It is used for downlink transmission of control commands and parameter configurations to each servo driver, and uplink transmission of sensor data and system status to the intelligent energy management controller; to realize timing synchronization and coordinated control between each servo unit.

Images