An industrial robot energy consumption modeling, prediction and optimization method based on mechanism prior

Abstract

This application discloses a mechanism-based prior knowledge-based method for modeling, predicting, and optimizing energy consumption of industrial robots. The method includes: collecting the total power and operational parameters (including joint angles, speeds, accelerations, and torques) of the industrial robot under different loads and motion parameters to construct an operational parameter-joint torque and total power dataset; integrating mechanism-based prior knowledge to construct an industrial robot energy consumption model consisting of a dynamics model and a power loss model; training the model using the constructed dataset to obtain the optimal energy consumption model; and using the optimal energy consumption model as the objective equation, employing an adaptive optimization algorithm to find the optimal motion parameters under given conditions to obtain the trajectory with minimum energy consumption. This invention solves the problems in related technologies, such as the complexity of constructing and identifying dynamics and energy consumption models, the difficulty in operation and implementation, and the poor interpretability, versatility, and accuracy of purely data-driven modeling.

Description

Technical Field

[0001] This application relates to the field of industrial robot technology, and in particular to a method for modeling, predicting and optimizing energy consumption of industrial robots based on mechanistic priors. Background Technology

[0002] As a mainstay of intelligent manufacturing, industrial robots play an increasingly important role in improving production efficiency and reducing manufacturing costs. They are widely used in automobile production, 3C product manufacturing, and other fields, driving the transformation and upgrading of the manufacturing industry. The widespread application of industrial robots inevitably leads to huge energy consumption. Improving the energy efficiency of industrial robots has become an urgent problem to solve. High-precision energy consumption modeling of industrial robots is the prerequisite and foundation for energy consumption optimization. However, industrial robots have numerous components, and the energy consumption characteristics of each component vary greatly and are complex, making energy consumption modeling difficult.

[0003] Currently, energy consumption modeling for industrial robots is mainly divided into direct modeling methods based on mechanistic models and indirect modeling methods based on data-driven approaches. Since industrial robot manufacturers generally do not disclose dynamic parameters, the core of direct modeling methods lies in dynamic derivation and parameter identification. However, dynamic modeling and identification processes are complex and require specialized knowledge; a single identification result is often insufficient to meet the calculation requirements of different operating conditions. Furthermore, in addition to the mechanical power handled by dynamics, there is also highly nonlinear power loss, which is difficult to accurately describe using mechanistic models. Therefore, direct modeling methods based on mechanistic models are complex and struggle to obtain high-precision energy consumption models, failing to meet the modeling requirements under complex operating conditions.

[0004] Intelligent algorithms, represented by deep learning, have been used in recent years for indirect energy consumption modeling of industrial robots due to their powerful nonlinear fitting capabilities. This leverages data-driven techniques to address the inaccuracy of mechanistic models. Current methods for modeling indirect energy consumption in industrial robots are typically based on "black box" models, employing purely end-to-end data-driven techniques to uncover the mapping relationship between operational information and energy consumption. However, existing research focuses on single load conditions, and the resulting models cannot adapt to the modeling needs of multiple load conditions. Furthermore, "black box" models have poor interpretability, failing to guarantee model accuracy and generalization performance, leading to poor model performance in scenarios outside the training dataset.

[0005] Currently, most industrial robots rely on offline programming software for trajectory simulation and planning. However, most industrial robot simulation software lacks accurate energy consumption assessment and optimization capabilities, failing to meet the current needs of green and low-carbon manufacturing. Therefore, it is necessary to construct an accurate energy consumption model to assess the energy consumption of the trajectory output by industrial robot simulation software. Based on this model, motion parameters can be optimized to find an energy-optimal trajectory, enabling energy consumption simulation, assessment, and optimization of industrial robots during the process planning stage. Summary of the Invention

[0006] The purpose of this application is to provide a mechanism-based prior modeling, prediction and optimization method for energy consumption of industrial robots, in order to solve the problems in related technologies, such as the complexity of energy consumption model construction and identification process, the difficulty of operation and implementation, and the poor interpretability, versatility and low accuracy of pure end-to-end data-driven modeling.

[0007] To achieve the above objectives, this application adopts the following technical solution:

[0008] A method for modeling, predicting, and optimizing the energy consumption of industrial robots based on prior knowledge of mechanisms includes the following steps:

[0009] Based on different loads and motion parameters, the total power and operating parameters of the industrial robot during operation are collected to construct an operating parameter-joint torque and total power dataset.

[0010] By integrating prior knowledge of the mechanism, an industrial robot energy consumption model consisting of a dynamic model and a loss power model is constructed. The industrial robot energy consumption model is then trained end-to-end based on the pre-constructed dataset to obtain the industrial robot energy consumption model with optimal parameters.

[0011] The optimal energy consumption model of an industrial robot is used for energy consumption prediction and optimization of industrial robots: taking the optimal energy consumption model of an industrial robot as the objective equation, an optimization algorithm is used to find the optimal motion parameters in a given working condition and obtain the running trajectory with the minimum energy consumption.

[0012] Preferably, the operating parameters include:

[0013] The design of trajectory clusters based on different loads and motion parameters should cover the entire workspace of the industrial robot as much as possible, and include the motion of each joint, covering the position, velocity and acceleration space of each joint.

[0014] The operating parameters need to include the joint position θ and velocity. acceleration The joint torque τ, the end load mass M, and the end load center of gravity position r.

[0015] More preferably, the operating parameters also include motor speed v m acceleration a m Current I m information.

[0016] Preferably, the specific process of constructing the operating parameters - joint torque and total power dataset is as follows: take the operating parameters other than the joint torque τ as samples x, and the joint torque and total power as labels y; combine each sample and the corresponding label to form a sample (x, y), and construct an industrial robot operating parameters - joint torque and total power dataset D containing multiple samples.

[0017] Preferably, the energy consumption model of the industrial robot, constructed by integrating prior knowledge of the fusion mechanism and consisting of a dynamic model and a power loss model, includes:

[0018] The prior knowledge of the mechanism includes that the total power of an industrial robot during operation consists of mechanical power and loss power; the mechanical power is obtained by multiplying the predicted joint torque value output by the dynamic model by the joint speed; and the energy consumption of the industrial robot is obtained by integrating the total power over time.

[0019] By integrating prior knowledge of the mechanism, an energy consumption model for industrial robots is constructed, consisting of a dynamic model and a power loss model. The dynamic model and the power loss model are constructed by intelligent algorithms.

[0020] Preferably, the step of performing end-to-end training of the industrial robot energy consumption model based on the pre-constructed dataset includes:

[0021] The total loss of the industrial robot energy consumption model consists of the total power loss and the joint torque loss; the joint torque loss of the industrial robot consists of the torque loss of each joint, and the total loss of the industrial robot energy consumption model is obtained by weighted summation of the torque loss of each joint and the total power loss.

[0022] The modeling and prediction results of joint torque and total power are expressed using the root mean square error percentage (RMSE%) and the coefficient of determination (R²). 2 To assess the accuracy of energy consumption modeling, relative error is used.

[0023] The running parameters—joint torque and total power—are divided into a training set and a test set according to a preset ratio. The training set data is used for model training, and the test set is used to test the model during the training process. The parameters with the smallest test loss are selected as the final model parameters.

[0024] Preferably, the energy consumption model of the industrial robot with optimal parameters is used for energy consumption prediction and optimization of the industrial robot, specifically including:

[0025] Given motion parameters, the running trajectory of the industrial robot is simulated and planned to obtain time-series trajectory information including joint angles, speed and acceleration. Then, the energy consumption of the running trajectory is predicted and evaluated using the industrial robot energy consumption model with optimal parameters.

[0026] Given the range of parameters, determine the appropriate optimization algorithm, take the energy consumption model of the industrial robot with the optimal parameters as the objective equation, and find the motion parameters with the minimum energy consumption and the corresponding running trajectory.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] (1) This invention uses intelligent algorithms such as deep learning to construct the energy consumption model of industrial robots. It can achieve high-precision power and energy consumption modeling and prediction without the need for complex mechanism modeling and parameter identification processes.

[0029] (2) This invention incorporates the prior mechanism of power composition of industrial robots into the energy consumption model, realizes the modeling of mechanical power and loss power, and improves the interpretability and generalization performance of the model.

[0030] (3) The present invention can realize the construction of energy consumption models for industrial robots with different load conditions and different degrees of freedom, and has general energy consumption modeling capabilities. Attached Figure Description

[0031] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0032] Figure 1 This is a flowchart illustrating a mechanism-based prior art method for modeling, predicting, and optimizing energy consumption of industrial robots, according to an embodiment of this application.

[0033] Figure 2 This is a schematic diagram of an energy consumption model for an industrial robot according to an embodiment of this application;

[0034] Figure 3 This is a flowchart illustrating a mechanism-based prior art method for modeling, predicting, and optimizing energy consumption of industrial robots according to an embodiment of this application. Detailed Implementation

[0035] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0036] Specifically, Figure 1This is a flowchart illustrating a mechanism-based prior art method for modeling, predicting, and optimizing energy consumption of industrial robots, as provided in an embodiment of this application.

[0037] like Figure 1 As shown, the method for modeling, predicting, and optimizing the energy consumption of industrial robots based on prior knowledge of mechanisms includes the following steps:

[0038] In step S101, based on different working conditions, the total power of the industrial robot during operation and the operating parameters, including the angles, speeds, accelerations and torques of each joint, are collected to construct an operating parameter-joint torque and total power dataset.

[0039] In practical implementation, embodiments of this application can design trajectory clusters based on different loads and motion parameter conditions, send them to the industrial robot for execution, and collect the total power and operating information during its motion. The aforementioned trajectory clusters need to cover the entire workspace of the industrial robot as much as possible, and include the motion of each joint, covering the position, velocity, and acceleration space of each joint. Each trajectory can be a point-to-point straight line trajectory in the workspace, or a Fourier series trajectory including the joint space.

[0040] Total power P r Data can be collected by power sensors or oscilloscopes, or obtained through the industrial robot's own data acquisition interface. Operating parameters must include at least the joint positions θ and speeds. acceleration Joint information such as joint torque τ, and end-load operating condition information such as end-load mass M and end-load center of gravity position r, may also include motor speed v. m acceleration a m Current I m Other information, such as operating parameters, can be obtained from the industrial robot's own data acquisition interface.

[0041] Some industrial robots support acquiring motor current but not joint torque information. In this case, the motor current I of each joint can be used to acquire joint torque information. m With the torque constant K of the motor m and transmission ratio G m Multiplying these together yields the joint torque. Some industrial robots do not support direct acquisition of joint acceleration information; instead, the joint acceleration can be calculated from the joint velocity using a center difference algorithm.

[0042] The acquisition of operational information from industrial robots is susceptible to noise, leading to a decline in data quality. Noise significantly impacts data such as total power, joint torque, and motor current, causing substantial fluctuations. These data can be processed using a zero-phase low-pass filter.

[0043] After obtaining sufficient data, construct a dataset of operating parameters—joint torque and total power. Use the operating parameters as samples x, and joint torque and total power as labels y. Then, divide the operating parameters by joint position θ and velocity... acceleration Optional information other than the end load mass M and the end load center of gravity position r is denoted as x′, i.e.:

[0044] x′={v m ,a m ,I m ,…}

[0045]

[0046] y={P r ,τ}

[0047] Each sample and its corresponding label are combined to form a sample (x, y), constructing a dataset D containing m samples of industrial robot operating parameters - joint torques and total power:

[0048] D = {(x1,y1),(x2,y2),…,(x m ,y m )}

[0049] In step S102, prior knowledge of the mechanism is integrated to construct an industrial robot energy consumption model consisting of a dynamic model and a loss power model. The industrial robot energy consumption model is then trained based on the pre-constructed dataset to obtain an industrial robot energy consumption model with optimal parameters.

[0050] The total power of an industrial robot consists of three parts: drive system power, control system power, and auxiliary system power. The power of the control system and auxiliary systems is relatively constant. The drive system power is a significant component of the industrial robot's energy consumption, comprising mechanical power, motor electrical losses, and drive system electrical circuit losses. For industrial robots, mechanical power is a crucial part of the total power and a key component in energy consumption modeling, which can be calculated using a dynamic model. Other power components can be considered as different forms of power losses, exhibiting high nonlinearity and making accurate direct modeling difficult.

[0051] like Figure 2 As shown, based on the power composition of industrial robots, the energy consumption model of industrial robots is designed as a dynamic model f dm and power loss model f plm constitute.

[0052] A typical dynamic model of an industrial robot can be expressed by the Lagrange equations as follows:

[0053]

[0054] Where H(θ) represents the inertia matrix; G(θ) represents the Coriolis force and centrifugal force terms; G(θ) represents the gravity term; τ f τ represents the joint friction term; e This represents the change in generalized force at the joint caused by the end effector force.

[0055] The Coulomb viscous friction model is a commonly used model for friction forces in industrial robot joints, and it is as follows:

[0056]

[0057] Among them, f c f is the Coulomb friction coefficient; v The coefficient of viscous friction; sign(·) is the sign function used to return the joint velocity. The positive and negative signs.

[0058] When the contact between the end effector of an industrial robot and the environment is not considered, the forces and torques at the end effector can be regarded as being caused by the end effector load. The end effector load of an industrial robot can be considered as part of the end effector linkage, τ e Based on the fact that the dynamic equations can be rewritten into a linear expression:

[0059]

[0060] Where, Φ e A set of dynamic parameters representing the end load; Represents the observation matrix.

[0061] For different end loads, the dynamic parameter set Φ of the end load e for:

[0062] Φ e =f Φ (M,r)

[0063] Where M represents the end-effector mass of the industrial robot; r = [x c ,y c ,z c ] represents the location of the center of gravity of the end load; f Φ (·) represents the independent variable M,r and the dependent variable Φ. e The functional relationship between them.

[0064] To meet the modeling requirements under different load conditions, the end-effector load information is also input into the dynamic model. Therefore, the inputs to the dynamic model are the joint position θ and velocity of the industrial robot. acceleration The output, given the end-load mass M and the end-load center of gravity r, is the predicted torque values ​​for each joint. Predicted joint torque values With joint velocity Multiplying them together yields the predicted value of the mechanical power. Right now:

[0065]

[0066]

[0067] For the power loss model, the input consists of various operational information related to power loss. Based on the above analysis, the power loss of an industrial robot includes the power of the control system, auxiliary system, motor electrical losses, and drive system electrical circuit losses. The power of the control system and auxiliary system is less affected by the robot's operating conditions and parameters and can be approximated as constants. Motor electrical losses include copper losses and iron losses, which are generally related to the motor's materials, structure, speed, and current. Many factors influence this loss, and it is difficult to measure or calculate directly; therefore, it can be approximated as being related to the joint position, speed, and acceleration. The power loss of the drive system electrical circuit includes the losses of electrical components such as rectifiers and inverters, which are generally considered to be related to the circuit current, and can be approximated as being related to the motor current. The motor current can be approximated by the joint torque. Therefore, the input to the power loss model should include the joint position θ and speed of the industrial robot. acceleration Information such as joint torque τ is used. Since joint torque cannot be directly obtained during the simulation prediction stage, the predicted joint torque values ​​output by the dynamic model are used here. Right now:

[0068]

[0069] in, It is a predicted value of power loss.

[0070] So, the predicted total power of the industrial robot for:

[0071]

[0072] Where x′ represents the joint position θ and velocity. acceleration The end load mass M, the end load center of gravity position r, and the joint torque Other operating parameters besides these are not required.

[0073] After obtaining the predicted total power of the industrial robot, the predicted energy consumption of the industrial robot can be obtained by integrating over time.

[0074]

[0075] Where t1 represents the time when the industrial robot starts moving; t2 represents the time when the industrial robot stops moving.

[0076] The dynamic model and power loss model used to construct the energy consumption model of industrial robots can employ intelligent algorithm models such as neural networks, convolutional neural networks, and recurrent neural networks. Optionally, the dynamic model can also adopt a mechanistic model, and then its parameters can be identified using a data-driven method.

[0077] The dynamic model uses mean square error (MSE) as the loss function.

[0078]

[0079] Among them, y i Represents the true value; represents the predicted value of the dynamic model; m represents the number of samples.

[0080] To ensure the accurate construction of the dynamic model, in addition to the total power loss, joint torque loss must also be introduced. The joint torque loss of an industrial robot consists of the joint torque loss of each joint. Since the scale range of the torque of each joint is not exactly the same, it is necessary to balance the torque loss of each joint.

[0081]

[0082] Where loss_t represents the joint torque loss of the industrial robot; λ represents the MSE loss of the torque at the i-th joint. i The balance coefficient represents the torque loss of the i-th joint.

[0083] Because the scale range between total power and joint torque is quite large, the total power loss is also multiplied by a balance factor to eliminate the influence of scale factors.

[0084]

[0085] Where loss represents the total loss of the industrial robot energy consumption model; loss_p represents the total power loss; L p λ represents the MSE loss of the total power; λ0 represents the balance coefficient of the total power.

[0086] After normalizing the sample data in the running parameters - joint torque, total power dataset, the dataset D is divided into training set D according to a certain ratio. train and test set D test The training set data is input into the industrial robot energy consumption model and trained using the backpropagation algorithm until the cumulative loss function value of the entire training set reaches its minimum. After training is complete, the model is tested on the test set D. test The energy consumption model of the industrial robot during the training process was tested, and the parameters with the minimum test loss were selected as the final model parameters.

[0087] The modeling and prediction results of joint torque and total power are expressed using the root mean square error percentage (RMSE%) and the coefficient of determination (R²). 2 To evaluate this, RMSE% is used to represent the percentage of power modeling error; a lower value indicates higher modeling accuracy. R... 2 This value reflects the effectiveness of the power curve fitting; the closer it is to 1, the better the curve fitting.

[0088]

[0089]

[0090] The modeling accuracy of energy consumption is evaluated using the relative error δ; the lower the value, the higher the modeling accuracy.

[0091]

[0092] Among them, E r Represents the actual energy consumption value. This represents the predicted energy consumption value.

[0093] In step S103, the energy consumption model of the industrial robot with the above-mentioned optimal parameters is used as the objective equation, and a suitable optimization algorithm is used to find the optimal motion parameters in the given working condition to obtain the running trajectory with the minimum energy consumption.

[0094] Industrial robots typically offer a variety of motion parameters to choose from when performing specific tasks, with varying operating times and energy consumption depending on the parameters. For a given maximum speed V... k Acceleration A k After obtaining the motion parameters, simulation software can be used to simulate and plan the movement trajectory of the industrial robot, obtaining time-series trajectory information including joint angles, velocity, and acceleration. After obtaining the trajectory information, the energy consumption of the trajectory is evaluated using the industrial robot energy consumption model with the above-mentioned optimal parameters.

[0095] After obtaining the energy consumption of the trajectory, energy consumption optimization can be performed on the industrial robot. For coarse-grained optimization, the range of parameters to be optimized is relatively small, and an exhaustive search method can be used to find the trajectory with the minimum energy consumption. For fine-grained optimization, the range of parameters to be searched is larger, and intelligent optimization algorithms such as genetic algorithms can be used, with the above energy consumption model as the objective equation, to find the motion parameters (V) with the minimum energy consumption. opt A opt (,...) and their corresponding trajectories

[0096] The following is combined with Figure 2 and Figure 3 As shown, a specific embodiment of the energy consumption modeling, prediction and optimization method for industrial robots based on mechanism priors in this application is described in detail.

[0097] like Figure 3 As shown, the embodiments of this application include the following steps:

[0098] Step S301: Collect the total power and operating parameters of the industrial robot during operation under different working conditions.

[0099] Trajectory clusters based on different loads and motion parameter conditions are sent to the industrial robot for execution, and the total power and operating parameters during its motion are collected. The trajectory clusters need to cover the entire workspace of the industrial robot as much as possible, and include the motion of each joint, covering the position, velocity, and acceleration space of each joint. Each trajectory can be a point-to-point straight line trajectory in the workspace, or a Fourier series trajectory that includes the joint space.

[0100] Total power P r Data can be collected by power sensors or oscilloscopes, or obtained through the industrial robot's own data acquisition interface. Operating parameters must include at least the joint positions θ and speeds. acceleration Joint information such as joint torque τ, and end-load operating condition information such as end-load mass M and end-load center of gravity position r, may also include motor speed v. m acceleration a m Current I m Other information, such as operating parameters, can be obtained from the industrial robot's own data acquisition interface.

[0101] Some industrial robots support acquiring motor current but not joint torque information. In this case, the motor current I of each joint can be used to acquire joint torque information. m With the torque constant K of the motor m and transmission ratio G mMultiplying these together yields the joint torque. Some industrial robots do not support direct acquisition of joint acceleration information; instead, the joint acceleration can be calculated from the joint velocity using a center difference algorithm.

[0102] Step S302: Construct the runtime parameter - joint torque, total power dataset.

[0103] The acquisition of operational information from industrial robots is susceptible to noise, leading to a decline in data quality. Noise significantly impacts data such as total power, joint torque, and motor current, causing substantial fluctuations. These data can be processed using a zero-phase low-pass filter.

[0104] After obtaining sufficient data, construct a dataset of operating parameters—joint torque and total power. Use the operating parameters as samples x, and joint torque and total power as labels y. Then, divide the operating parameters by joint position θ and velocity... acceleration Optional information other than the end load mass M and the end load center of gravity position r is denoted as x′, i.e.:

[0105] x′={v m ,a m ,I m ,…}

[0106]

[0107] y={P r ,τ}

[0108] Each sample and its corresponding label are combined to form a sample (x, y), constructing a dataset D containing m samples of industrial robot operating parameters - joint torques and total power:

[0109] D = {(x1,y1),(x2,y2),…,(x m ,y m )}

[0110] Step S303: Construct an industrial robot dynamics model based on intelligent algorithms.

[0111] Industrial robot dynamics models can be constructed end-to-end using intelligent algorithms such as neural networks, convolutional neural networks, and recurrent neural networks. These algorithms directly extract the dynamic characteristics of industrial robots from data, eliminating the need for complex mechanistic modeling and parameter identification, thus enabling the construction of high-precision dynamic models. Industrial robot dynamics models built based on intelligent algorithms require input parameters related to industrial robot dynamics and output joint torques.

[0112] The input to the dynamic model is determined as follows:

[0113] A typical dynamic model of an industrial robot can be expressed by the Lagrange equations as follows:

[0114]

[0115] Where H(θ) represents the inertia matrix; G(θ) represents the Coriolis force and centrifugal force terms; G(θ) represents the gravity term; τ f τ represents the joint friction term; e This represents the change in generalized force at the joint caused by the end effector force.

[0116] The Coulomb viscous friction model is a commonly used model for friction forces in industrial robot joints, and it is as follows:

[0117]

[0118] Among them, f c f is the Coulomb friction coefficient; v The coefficient of viscous friction; sign(·) is the sign function used to return the joint velocity. The positive and negative signs.

[0119] When the contact between the end effector of an industrial robot and the environment is not considered, the forces and torques at the end effector can be regarded as being caused by the end effector load. The end effector load of an industrial robot can be considered as part of the end effector linkage, τ e Based on the fact that the dynamic equations can be rewritten into a linear expression:

[0120]

[0121] Where, Φ e A set of dynamic parameters representing the end load; Represents the observation matrix.

[0122] For different end loads, the dynamic parameter set Φ of the end load e for:

[0123] Φ e =f Φ (M,r)

[0124] Where M represents the end-effector mass of the industrial robot; r = [x c ,y c ,z c ] represents the location of the center of gravity of the end load; f Φ (·) represents the independent variable M,r and the dependent variable Φ. e The functional relationship between them.

[0125] To meet the modeling requirements under different load conditions, the end-effector load information is also input into the dynamic model. Therefore, the inputs to the dynamic model are the joint position θ and velocity of the industrial robot. acceleration The output, given the end-load mass M and the end-load center of gravity r, is the predicted torque values ​​for each joint. Right now:

[0126]

[0127] Among them, f dm (·) represents an industrial robot dynamics model built based on intelligent algorithms.

[0128] Step S304: Construct an industrial robot power loss model based on intelligent algorithms.

[0129] The power loss of industrial robots is highly nonlinear, making direct modeling difficult. Therefore, the powerful fitting capabilities of intelligent algorithms are utilized to model the power loss. For a power loss model based on intelligent algorithms, the input consists of various operational information related to power loss, and the output is the power loss itself.

[0130] The input to the power loss model is determined as follows:

[0131] The power losses of industrial robots include the power of the control system, auxiliary system, motor electrical losses, and drive system electrical circuit losses. The power of the control system and auxiliary system is less affected by the robot's operating conditions and parameters and can be approximated as constants. Motor electrical losses include copper losses and iron losses, which are generally related to the motor's materials, structure, speed, and current. Many factors influence these losses, and they are difficult to measure or calculate directly; however, they can be approximated as being related to joint position, speed, and acceleration. The power losses of the drive system electrical circuits include the losses of electrical components such as rectifiers and inverters, which are generally considered to be related to the circuit current, and can be approximated as being related to the motor current. The motor current can be approximately calculated from the joint torque.

[0132] Therefore, the inputs to the power loss model should include the joint position θ and velocity of the industrial robot. acceleration Information such as joint torque τ is included. Since joint torque cannot be directly obtained during the simulation prediction stage, the predicted joint torque values ​​output from the dynamic model are used here. Right now:

[0133]

[0134] Among them, f plm (·) represents an industrial robot loss model built based on intelligent algorithms.

[0135] Step S305: Construct an industrial robot energy consumption model based on prior knowledge of the mechanism, train and test the model, and obtain the industrial robot energy consumption model with optimal parameters.

[0136] The total power of an industrial robot consists of three parts: drive system power, control system power, and auxiliary system power. The power of the control system and auxiliary systems is relatively constant. The drive system power is a significant component of the industrial robot's energy consumption, comprising mechanical power, motor electrical losses, and drive system electrical circuit losses. For industrial robots, mechanical power is a crucial component of their total power and a key part of energy consumption modeling, which can be calculated from a dynamic model. Treating all other power sources besides mechanical power as different forms of loss power, the total power of an industrial robot is composed of mechanical power and loss power. Mechanical power can be obtained by multiplying the torque output by the dynamic model by the joint velocity. After obtaining the predicted total power of the industrial robot, integrating over time yields the predicted energy consumption. Figure 2 As shown, by incorporating the above power composition, mechanical power, and energy consumption calculation mechanism into the model construction, an energy consumption model incorporating prior knowledge of the mechanism can be obtained.

[0137]

[0138]

[0139] in, A predicted value representing the total power of an industrial robot; The predicted energy consumption of the industrial robot is represented by t1; t2 represents the time when the industrial robot starts moving; t2 represents the time when the industrial robot stops moving; x′ represents the energy consumption excluding joint position θ and velocity. acceleration The end load mass M, the end load center of gravity position r, and the joint torque Other operational information besides this is not required.

[0140] The dynamic model uses mean square error (MSE) as the loss function.

[0141]

[0142] Among them, y i Represents the true value; represents the model's predicted value; m represents the number of samples.

[0143] To ensure the accurate construction of the dynamic model, in addition to the total power loss, joint torque loss must also be introduced. The joint torque loss of an industrial robot consists of the joint torque loss of each joint. Since the scale range of the torque of each joint is not exactly the same, it is necessary to balance the torque loss of each joint.

[0144]

[0145] Where loss_t represents the joint torque loss of the industrial robot; λ represents the MSE loss of the torque at the i-th joint. i The balance coefficient represents the torque loss of the i-th joint.

[0146] Because the scale range between total power and joint torque is quite large, the total power loss is also multiplied by a balance factor to eliminate the influence of scale factors.

[0147]

[0148] Where loss represents the total model loss; loss_p represents the total power loss; L p λ represents the MSE loss of the total power; λ0 represents the balance coefficient of the total power.

[0149] After normalizing the sample data in the running parameters - joint torque, total power dataset, the dataset D is divided into training set D according to a certain ratio. train and test set D test The training set data is input into the industrial robot energy consumption model and trained using the backpropagation algorithm until the cumulative loss function value of the entire training set reaches its minimum. After training is complete, the model is tested on the test set D. test The energy consumption model of the industrial robot during the training process was tested, and the parameters with the minimum test loss were selected as the final model parameters.

[0150] The modeling and prediction results of joint torque and total power are expressed using the root mean square error percentage (RMSE%) and the coefficient of determination R. 2 To evaluate this, RMSE% is used to represent the percentage of power modeling error; a lower value indicates higher modeling accuracy. R... 2 This value reflects the effectiveness of the power curve fitting; the closer it is to 1, the better the curve fitting.

[0151]

[0152]

[0153] The modeling accuracy of energy consumption is evaluated using the relative error δ; the lower the value, the higher the modeling accuracy.

[0154]

[0155] Among them, E r Represents the actual energy consumption value. This represents the predicted energy consumption value.

[0156] Step S306: Using the optimal energy consumption model of the industrial robot as the objective equation, the optimization algorithm is used to find the trajectory parameters with the minimum energy consumption, thereby achieving energy consumption optimization.

[0157] Industrial robots typically offer a variety of motion parameters to choose from when performing specific tasks, with varying operating times and energy consumption depending on the parameters. For a given maximum speed V... k Acceleration A k After obtaining the motion parameters, simulation software can be used to simulate and plan the movement trajectory of the industrial robot, obtaining time-series trajectory information including joint angles, velocity, and acceleration. After obtaining the trajectory information, the energy consumption of the trajectory is predicted and evaluated using the industrial robot energy consumption model with the above-mentioned optimal parameters.

[0158] After obtaining the energy consumption of the trajectory, energy consumption optimization can be performed on the industrial robot. For coarse-grained optimization, the range of parameters to be optimized is relatively small, and an exhaustive search method can be used to find the trajectory with the minimum energy consumption. For fine-grained optimization, the range of parameters to be searched is larger, and intelligent optimization algorithms such as genetic algorithms can be used, with the above energy consumption model as the objective equation, to find the motion parameters (V) with the minimum energy consumption. opt A opt (,...) and their corresponding trajectories

[0159] The energy consumption modeling, prediction, and optimization method for industrial robots proposed in this application integrates prior knowledge of the mechanism into the construction of the energy consumption model, generating an energy consumption model for the industrial robot composed of a dynamic model and a power loss model. The proposed method can accurately model the dynamic model and mechanical power, and also utilize intelligent algorithms to fit the nonlinear power loss component of the industrial robot, which is difficult to model mechanistically. This ensures the accuracy of energy consumption modeling and prediction, and improves the model's generalization and interpretability. Therefore, it solves the problems in related technologies, such as the complexity of mechanistic modeling and parameter identification, the difficulty in operation and implementation, and the low accuracy, poor interpretability and generalization of purely end-to-end data-driven modeling, making it difficult to apply to unfamiliar scenarios outside the dataset.

Claims

1. A method for modeling, predicting, and optimizing energy consumption of industrial robots based on mechanistic priors, characterized in that, Includes the following steps: Based on different loads and motion parameters, the total power and operating parameters of the industrial robot during operation are collected to construct an operating parameter-joint torque and total power dataset. By integrating prior knowledge of the mechanism, an industrial robot energy consumption model consisting of a dynamic model and a loss power model is constructed. The industrial robot energy consumption model is then trained end-to-end based on the pre-constructed dataset to obtain the industrial robot energy consumption model with optimal parameters. The optimal energy consumption model of an industrial robot is used for energy consumption prediction and optimization of industrial robots: taking the optimal energy consumption model of an industrial robot as the objective equation, the optimization algorithm is used to find the optimal motion parameters in a given working condition and obtain the running trajectory with the minimum energy consumption. The aforementioned fusion mechanism prior knowledge is used to construct an industrial robot energy consumption model consisting of a dynamic model and a power loss model, including: The prior knowledge of the mechanism includes that the total power of an industrial robot during operation consists of mechanical power and loss power; the mechanical power is obtained by multiplying the predicted joint torque value output by the dynamic model by the joint speed; and the energy consumption of the industrial robot is obtained by integrating the total power over time. By integrating prior knowledge of the mechanism, an energy consumption model for industrial robots is constructed, consisting of a dynamic model and a power loss model, in which the dynamic model and the power loss model are constructed by intelligent algorithms. The end-to-end training of the industrial robot energy consumption model based on the pre-constructed dataset includes: The total loss of the industrial robot energy consumption model consists of the total power loss and the joint torque loss; the joint torque loss of the industrial robot consists of the torque loss of each joint, and the total loss of the industrial robot energy consumption model is obtained by weighted summation of the torque loss of each joint and the total power loss. The modeling and prediction results of joint torque and total power are expressed using the root mean square error percentage (RMSE%) and the coefficient of determination (R²). 2 To assess the accuracy of energy consumption modeling, relative error is used. The running parameters—joint torque and total power—are divided into a training set and a test set according to a preset ratio. The training set data is used for model training, and the test set is used to test the model during the training process. The parameters with the smallest test loss are selected as the final model parameters.

2. The method according to claim 1, characterized in that, The operating parameters include: The design of trajectory clusters based on different loads and motion parameters needs to cover the entire workspace of the industrial robot and include the motion of each joint, covering the position, velocity and acceleration space of each joint. The operating parameters need to include the positions of each joint. ,speed acceleration and joint torque and end load quality and the position of the center of gravity of the end load .

3. The method according to claim 2, characterized in that, The operating parameters also include motor speed. acceleration Current information.

4. The method according to claim 1, characterized in that, The specific process for constructing the operating parameter - joint torque, total power dataset is as follows: (Except for joint torque...) Other operating parameters as samples Joint torque and total power are used as labels. Each sample and its corresponding label form a test case. Construct a dataset of industrial robot operating parameters—joint torques and total power—containing multiple examples. .

5. The method according to claim 1, characterized in that, The optimal energy consumption model for industrial robots is applied to predict and optimize industrial robot energy consumption, specifically including: Given motion parameters, the running trajectory of the industrial robot is simulated and planned to obtain time-series trajectory information including joint angles, speed and acceleration. Then, the energy consumption of the running trajectory is predicted and evaluated using the industrial robot energy consumption model with optimal parameters. Given the range of parameters, determine the appropriate optimization algorithm, take the energy consumption model of the industrial robot with the optimal parameters as the objective equation, and find the motion parameters with the minimum energy consumption and the corresponding running trajectory.

Images