Adaptive knowledge-driven optimization method and system for intelligent inspection of quadruped robot

Abstract

The application discloses a quadruped robot intelligent inspection adaptive knowledge driving optimization method and system, and particularly relates to the technical field of driving optimization, and comprises the following steps: acquiring inspection task data of a quadruped robot, inputting the inspection task data into a pre-constructed power consumption prediction model to obtain a minimum power consumption value; acquiring a decision training data set of the quadruped robot; training, based on the decision training data set, a power consumption replacement decision model for deciding whether to replace the power consumption of the inspection task when detecting power consumption abnormality of the quadruped robot; judging whether the quadruped robot has power consumption abnormality based on the minimum power consumption value predicted by the power consumption prediction model; and the application decides whether to enable replacement of the power consumption of the inspection task according to real-time data, thereby minimizing energy waste and prolonging single inspection endurance of the robot; in long-term large-scale inspection operation, the frequency of frequent charging is reduced, not only the energy cost is reduced, but also the inspection continuity is ensured, and the overall efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of drive optimization technology, and more specifically, to an adaptive knowledge-driven optimization method and system for intelligent inspection of quadruped robots. Background Technology

[0002] With the development of industrial automation and intelligence, quadruped robots are widely used in inspection tasks in complex environments due to their excellent terrain adaptability and flexible movement performance. However, quadruped robots face many challenges in actual operation, among which power consumption optimization is a key issue. Robots need to operate for extended periods in complex environments, and battery life and energy efficiency directly determine task completion capability and economic benefits. However, uncertainties in complex environments, such as terrain complexity, obstacle distribution, and changes in task priorities, often lead to low power efficiency of traditional methods, or even task failure.

[0003] Existing methods, such as the Chinese patent application CN118656734A, disclose a petrochemical area inspection system based on an inspection robot, including: real-time collection of sensor data between inspection points, analysis of data between multiple inspection points, integration of node dependency and data flow indicators, and calculation of the interaction strength of each node through a neural network model. While the above technical solution optimizes inspection path configuration and improves inspection efficiency and safety by dynamically integrating and analyzing sensor data to calculate the interaction strength and dependency between nodes, research and application of the technical solution and existing technologies have revealed at least the following shortcomings:

[0004] The robot failed to dynamically adjust its power consumption according to task requirements, resulting in insufficient battery life or wasted resources.

[0005] To this end, the present invention provides an adaptive knowledge-driven optimization method and system for intelligent inspection of quadruped robots. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art, this invention provides an adaptive knowledge-driven optimization method and system for intelligent inspection of quadruped robots, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] In a first aspect, this invention provides an adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots, including:

[0009] Step 1: Obtain the inspection task data of the quadruped robot and input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value;

[0010] Step 2: Obtain the decision training dataset of the quadruped robot; based on the decision training dataset, train a power replacement decision model to determine whether to use the power replacement function to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected.

[0011] Step 3: Based on the minimum power consumption value predicted by the power consumption prediction model, determine whether there is a power consumption anomaly in the quadruped robot; if there is a power consumption anomaly, generate a power consumption anomaly command and proceed to step 4; if there is no power consumption anomaly, return to step 1.

[0012] Step 4: Receive the power consumption anomaly command, obtain the power consumption of the replacement inspection task, and based on the power consumption of the replacement inspection task, the running power consumption, and the power consumption replacement decision model, output a decision on whether to use the power consumption of the replacement inspection task.

[0013] Furthermore, the inspection task data includes motion power consumption score and task power consumption score;

[0014] The motion power consumption score is obtained by combining data on the robot's environment, motion speed, acceleration, and motion type.

[0015] Furthermore, the method for obtaining the task power consumption score includes:

[0016] Step a11: Obtain the task data of the quadruped robot. The task data includes the detection task type and the task execution time. The detection task type includes equipment appearance inspection task, equipment operating status inspection task, and equipment internal parameter measurement task.

[0017] Step a12: Analyze the task data to obtain the task power consumption score.

[0018] Furthermore, the training method for the power consumption prediction model is as follows:

[0019] The historical inspection task training data is divided into a task training set and a task test set to construct a regression network model. The historical inspection task training data includes inspection task data and the minimum power consumption value corresponding to the inspection task data.

[0020] Using inspection task data from the task training set as input and the corresponding minimum power consumption value as output, both are fed into a regression network model for training to construct an initial regression network model. The initial regression network model is a neural network model, and the training process focuses on minimizing the sum of prediction accuracies. After the initial training is completed, the initial regression network model is evaluated using a task test set. If the sum of prediction accuracies achieved by the initial regression network model in the task test set is lower than a preset threshold, then the initial regression network model is established as a power consumption prediction model. Conversely, if the sum of prediction accuracies reaches or exceeds the preset threshold, the task training set is re-inputted, and the initial regression network model is continuously iterated and trained until the sum of prediction accuracies obtained in the testing phase meets the preset standard.

[0021] Furthermore, the method for obtaining the minimum power consumption value corresponding to the inspection task data in the historical inspection task training data is as follows:

[0022] Step b1: Obtain the robot's mass and total running time;

[0023] Step b2: Formulate the inspection task data, robot quality, and total running time of the m-th industrial equipment to obtain the power consumption evaluation score of the m-th industrial equipment; m = 1, 2, ..., M; M is the total number of industrial equipment.

[0024] Step b3: Let m = m + 1, repeat the above steps until m = M, then end the loop and mark the minimum power consumption evaluation score corresponding to the industrial equipment as the minimum power consumption value.

[0025] Furthermore, methods for obtaining the decision training dataset include:

[0026] Step c1: Obtain replacement decision data for the quadruped robot, including the power consumption value of the m-th industrial device, the task completion index, the power consumption of the replacement inspection task, the robot quality, and the task power consumption replacement decision.

[0027] Step c2: Generate a decision training data set based on the replacement decision data;

[0028] The methods for obtaining the task completion index include:

[0029] Obtain the task completion feature data of the mth industrial equipment, wherein the task completion feature data includes the inspection task completion rate and the data transmission completion rate;

[0030] The inspection task completion rate is the ratio between the number of inspection tasks completed by the quadruped robot for the m-th industrial equipment and the preset number of inspection tasks completed; the data transmission completion rate is the ratio between the data transmission of the quadruped robot for the m-th industrial equipment after completing the inspection task and the preset data transmission.

[0031] The task completion status is determined based on the task completion index. The task completion status includes inspection task not completed and inspection task completed.

[0032] Methods for determining task completion status based on a task completion index include:

[0033] A preset task index threshold is set, and the task completion index is compared with the task index threshold.

[0034] When the task completion index is less than or equal to the task index threshold, the task completion status will be generated as "incomplete inspection task".

[0035] When the task completion index is greater than the task index threshold, the task completion status will be generated as inspection task completed.

[0036] The power consumption of the replacement inspection task is the power consumption of stopping the inspection task and returning to the starting point;

[0037] The task power consumption replacement decision is that each time the quadruped robot faces a power consumption state during an inspection task, it chooses or does not choose to use one of the replacement inspection task power consumption options.

[0038] Furthermore, the decision training data set refers to generating a set of decision training data corresponding to each set of replacement decision data; and generating a decision training data set based on all replacement decision data.

[0039] The decision training data includes the current state, selected action, reward value, and next state corresponding to each set of replacement decision data;

[0040] The method for generating a decision training dataset based on replacement decision data is as follows:

[0041] The current state is defined by the robot's battery level, the power consumption value of each set of replacement decision data, and the power consumption of the replacement inspection task.

[0042] The action to be selected is to replace the task power consumption in the decision data with the action of selection.

[0043] Calculate the reward value Q for each set of replacement decision data after selecting an action;

[0044] The power consumption of the next set of replacement decision data is the next state.

[0045] Furthermore, the method for training a power substitution decision model to determine whether to use a power substitution model to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected includes:

[0046] The current state and reward value in the decision training dataset are used as inputs to the power consumption replacement decision model. The power consumption replacement decision model randomly selects multiple sets of decision training data from the decision training dataset for training, and learns whether to choose to replace the power consumption of the inspection task under different power consumption states in order to obtain the maximum reward value. The power consumption replacement decision model is a deep learning model.

[0047] Furthermore, methods for determining whether a quadruped robot exhibits abnormal power consumption include:

[0048] Obtain the remaining power consumption value of the quadruped robot, calculate the difference between the minimum power consumption value predicted by the power consumption prediction model and the remaining power consumption value, and obtain the power consumption difference value.

[0049] The power consumption difference is compared with a preset power consumption difference threshold;

[0050] If the power consumption difference is less than the preset power consumption difference threshold, a power consumption abnormality command will be generated.

[0051] If the power consumption difference is greater than or equal to the preset power consumption difference threshold, no power consumption abnormality instruction will be generated.

[0052] The methods for deciding whether to use the power consumption of the replacement inspection task include:

[0053] The decision model will output whether to use the power consumption of the replacement inspection task, based on the lowest power consumption value, the power consumption of the generated replacement inspection task, and the robot mass input power consumption of the quadruped robot to be controlled.

[0054] Secondly, the present invention provides an adaptive knowledge-driven optimization system for intelligent inspection of quadruped robots; the adaptive knowledge-driven optimization method for implementing the above-mentioned intelligent inspection of quadruped robots includes:

[0055] The data acquisition module is used to acquire the inspection task data of the quadruped robot and input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value.

[0056] The decision model training module is used to acquire the decision training data set of the quadruped robot; based on the decision training data set, a power replacement decision model is trained to determine whether to use the power replacement function to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected.

[0057] The judgment module determines whether the quadruped robot has power consumption anomalies based on the minimum power consumption value predicted by the power consumption prediction model. If a power consumption anomaly exists, a power consumption anomaly command is generated and transferred to the decision module. If no power consumption anomaly exists, the data acquisition module is returned.

[0058] The decision module is used to receive power consumption anomaly commands, obtain the power consumption of the replacement inspection task, and output a decision on whether to use the power consumption of the replacement inspection task based on the power consumption of the replacement inspection task, the running power consumption, and the power consumption replacement decision model.

[0059] The technical effects and advantages of this invention are as follows:

[0060] 1. This invention utilizes a precise power consumption prediction model to calculate the minimum power consumption value in advance, providing a basis for inspection task planning and avoiding unnecessary high-power paths. When abnormal power consumption is detected, the power consumption replacement decision model quickly intervenes, determining whether to activate a power replacement for the inspection task based on real-time data, minimizing energy waste and extending the robot's single inspection endurance. In long-term, large-scale inspection operations, this reduces the frequency of charging, not only reducing energy costs but also ensuring inspection continuity and improving overall efficiency.

[0061] 2. This invention relies on detailed task power consumption scoring and motion power consumption scoring methods to quantify power consumption in accordance with actual scenarios. During task execution, progress is dynamically tracked through a task completion index to accurately determine the completion status of the inspection task. Any power consumption anomalies are promptly adjusted to ensure no critical inspection steps are missed. A deep learning-based power replacement decision model continuously learns and optimizes, enabling the robot to make scientific power consumption decisions even under complex working conditions. This significantly reduces inspection interruptions caused by power consumption issues and comprehensively enhances the accuracy and completeness of inspection data. Attached Figure Description

[0062] Figure 1 The flowchart of the adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots in Example 1 is shown below.

[0063] Figure 2 This is a flowchart of the method for obtaining the minimum power consumption value corresponding to the inspection task data in the historical inspection task training data of Example 1;

[0064] Figure 3 Here is a flowchart of the method for determining whether a quadruped robot has abnormal power consumption in Example 1;

[0065] Figure 4 This is a schematic diagram of the adaptive knowledge-driven optimization system for intelligent inspection of a quadruped robot in Example 2. Detailed Implementation

[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0067] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0068] It should be understood that although terms such as "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and a similar second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated items listed.

[0069] Example 1

[0070] Please see Figure 1 As shown, this embodiment discloses an adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots, applied to quadruped robots. The method includes:

[0071] Step 1: Obtain the inspection task data of the quadruped robot and input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value;

[0072] It should be noted that the quadruped robot can inspect industrial equipment or power equipment, etc. In this embodiment, the inspection task of the quadruped robot on M industrial equipment in the target factory is used as an example. Each industrial equipment inspection task includes a movement task and a detection task. The movement task includes walking, jumping and running; the detection task includes equipment appearance inspection task, equipment operating status inspection task and equipment internal parameter measurement task.

[0073] The inspection task data includes motion power consumption score and task power consumption score.

[0074] The methods for obtaining the motion power consumption score include:

[0075] Step a01: Obtain the basic motion parameters of the quadruped robot, including motion type assignment, robot mass, motion speed, and acceleration;

[0076] It should be noted that: the motion type is preset, with different values ​​assigned based on the quadruped robot's motion state. A quadruped robot in a walking state is assigned a motion type value of 1, one in a jumping state is assigned a motion type value of 2, and one in a running state is assigned a motion type value of 3. The motion speed is the average forward speed of the quadruped robot within a preset time period, calculated as the ratio of displacement to the duration of the preset time period. The acceleration reflects the change in the quadruped robot's speed. Both motion speed and acceleration are acquired using a speed sensor.

[0077] Step a02: Formulate the basic motion parameters and the standby power consumption of the quadruped robot to obtain the motion power consumption score. The calculation formula is as follows:

[0078]

[0079] In the formula, E yd ZL represents the motion power consumption score, Hy represents the robot mass, BT represents the motion type assignment, SD represents the motion speed, DX represents the acceleration, and P0 represents the standby power consumption of the quadruped robot.

[0080] It should be noted that the environmental impact data is obtained by the quadruped robot using devices such as lidar and ultrasonic sensors to scan the terrain around the robot. The degree of environmental influence on the robot's movement is determined by analyzing parameters such as terrain undulation and roughness. This data is pre-set by those skilled in the art based on the geographical environment. For example, if the quadruped robot is moving on flat ground, the terrain sensors detect relatively small terrain changes, and the environmental impact data is set to a small value, such as 0.2; however, when the quadruped robot is in a rugged mountainous environment, the terrain undulation is large, the terrain parameters detected by the sensors change drastically, and the environmental impact data increases to 0.8.

[0081] The larger the robot's mass and the heavier the load, the greater the gravitational potential energy the robot needs to overcome, and the higher the energy consumption for movement will be. This is to demonstrate the extent of its impact on energy consumption.

[0082] The methods for obtaining the task power consumption score include:

[0083] Step a11: Obtain the task data of the quadruped robot, the task data including the power consumption of the detected task type and the task execution time;

[0084] It should be noted that the power consumption of the detection task type includes the power consumption P1 for appearance detection and the power consumption P2 for internal parameter measurement.

[0085] The power consumption P1 for appearance inspection is obtained by recording power data in real time using a power sensor equipped on the robot during the appearance inspection task, and calculating the total energy consumption P1 through time integration. The calculation formula is as follows: t1 and t2 represent the start and end times of the visual inspection task. This indicates the real-time power measured by the energy consumption sensor during the appearance inspection process. The monitoring method for the power consumption P2 of the internal parameter measurement is the same as that for the power consumption P1 of the appearance inspection, so it will not be repeated here.

[0086] Step a12: Analyze the task data to obtain the task power consumption score, the fitting formula of which is:

[0087]

[0088] In the formula, E rw T represents the task power consumption score. r This represents the task execution time, and φ1 and φ2 are the corresponding weight data, where φ1 + φ2 = 1.

[0089] In implementation, the training method for the power consumption prediction model is as follows:

[0090] The historical inspection task training data is divided into a task training set and a task test set to construct a regression network model. The historical inspection task training data includes inspection task data and the minimum power consumption value corresponding to the inspection task data. The inspection task training data is obtained by analyzing the quadruped robot in different inspection tasks.

[0091] Please see Figure 2 As shown, the method for obtaining the minimum power consumption value corresponding to the inspection task data in the historical inspection task training data is as follows:

[0092] Step b1: Obtain the robot's mass and total running time;

[0093] It should be noted that: the robot's mass is obtained in advance by measuring the gravity sensor; the total running time is calculated by the device's built-in timer, which starts accumulating the time from the moment the battery begins to power the quadruped robot until the battery is depleted and the quadruped robot stops running. The data accumulated by the timer at the end is the usage time of the battery's complete power supply process.

[0094] Step b2: Formulate the inspection task data, robot quality, and total operating time of the m-th industrial equipment to obtain the power consumption evaluation score of the m-th industrial equipment; m = 1, 2, ..., M; the calculation formula is:

[0095] EP m =(E yd +E rw)×α1+ZL×α2+DCZ×α3;

[0096] In the formula, EP m α represents the power consumption assessment score of the m-th industrial device, DCZ represents the total operating time, and α1, α2, and α3 are all weighted data, α1+α2+α3=1.

[0097] Step b3: Let m = m + 1, repeat the above steps until m = M, then end the loop and mark the minimum power consumption evaluation score corresponding to the industrial equipment as the minimum power consumption value.

[0098] Using inspection task data from the task training set as input and the corresponding minimum power consumption value as output, both are fed into a regression network model for training, constructing an initial regression network model. This initial regression network model is a neural network model, and the training process focuses on minimizing the sum of prediction accuracies. After initial training, the initial regression network model is evaluated using a task test set. If the sum of prediction accuracies achieved by the initial regression network model in the task test set is lower than a preset threshold, then the initial regression network model is established as a power consumption prediction model. Conversely, if the sum of prediction accuracies reaches or exceeds the preset threshold, the task training set is re-inputted, and the initial regression network model is continuously iterated and trained until the sum of prediction accuracies obtained in the testing phase meets the preset standard.

[0099] In this step, the quadruped robot predicts the power consumption of inspecting M industrial devices based on inspection task data, which informs the next step of adjusting the order of inspection tasks. Simple, low-power visual inspection tasks are performed when the battery is lower, while high-power tasks such as deep scanning of complex equipment and multi-sensor collaborative detection are scheduled when the battery is full, making the quadruped robot's inspection task execution rhythm more rational.

[0100] Step 2: Obtain the decision training dataset of the quadruped robot; based on the decision training dataset, train a power replacement decision model to determine whether to use the power replacement function to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected.

[0101] In practice, methods for obtaining the decision training dataset include:

[0102] Step c1: Obtain replacement decision data for the quadruped robot, including the power consumption value of the m-th industrial device, the task completion index, the power consumption of the replacement inspection task, the robot quality, and the task power consumption replacement decision.

[0103] The method for obtaining the task completion index includes:

[0104] Obtain the task completion feature data of the mth industrial equipment, wherein the task completion feature data includes the inspection task completion rate and the data transmission completion rate;

[0105] It should be noted that: the inspection task completion rate is the ratio between the number of inspection tasks completed by the quadruped robot for the m-th industrial equipment and the preset number of inspection tasks completed; the data transmission completion rate is the ratio between the data transmission of the quadruped robot for the m-th industrial equipment and the preset data transmission.

[0106]

[0107] In the formula, RW represents the task completion index, and P... sx P represents the completion rate of the inspection equipment. sc This indicates the data transmission completion rate.

[0108] It should be noted that: in the molecular part, when P sx ×P sc As P approaches 1, the value of the logarithmic function rises rapidly, more acutely reflecting the advantage of a high completion rate; when P... sx ×P sc As P approaches zero, the growth is slow; in the denominator, when P... sx ×P sc The closer the fraction is to 1, the closer the denominator is to 0, and the larger the value of the whole fraction, which means the higher the task completion index. Conversely, when both completion rates are low, the denominator is close to 1, which has a limited effect on improving the overall value, thus highlighting the disadvantage when the task completion is poor.

[0109] The task completion status is determined based on the task completion index. The task completion status includes inspection task not completed and inspection task completed.

[0110] In practice, methods for determining task completion status based on a task completion index include:

[0111] A preset task index threshold is set. The task completion index is compared with the task index threshold. When the task completion index is less than or equal to the task index threshold, it means that the inspection of the m-th industrial equipment has not been completed, and the task completion status is generated as "incomplete inspection task". When the task completion index is greater than the task index threshold, it means that the inspection of the m-th industrial equipment has been completed, and the task completion status is generated as "complete inspection task".

[0112] A higher task completion index indicates a higher completion rate for the corresponding m-th industrial equipment inspection task, meaning the quadruped robot does not use the replacement inspection task power consumption. This is because the quadruped robot has already undergone deep optimization and adaptation for the existing inspection task process. Even if the new replacement inspection task has lower power consumption, it may have compatibility issues with the existing system in areas such as instruction issuance, action execution, and data feedback, leading to instability and increased error probability in the quadruped robot's operation. Therefore, for reliability considerations, the original high-power, high-completion-rate task is still used.

[0113] The power consumption of the replacement inspection task is the power consumption when the inspection task is stopped and returned to the starting point.

[0114] The method for obtaining the power consumption of the replacement inspection task includes:

[0115] Step d1: Obtain the path back to the starting point:

[0116]

[0117] In the formula, d represents the path back to the starting point, and x c With y c The x-axis represents the current coordinates of the quadruped robot. s With y s Indicates the starting coordinates of the quadruped robot;

[0118] Step d2: Obtain operation switching power consumption:

[0119] E qh =P qh ×T qh ;

[0120] In the formula, E qh P represents the switching power consumption. qh Indicates the switching power, T qh Indicates the operation switching time.

[0121] It should be noted that the switching power is obtained through a power sensor.

[0122] Step d3: Obtain the return speed, and comprehensively analyze the return speed, the path back to the starting point, the motion power consumption score, and the switching power consumption to obtain the power consumption of the replacement inspection task. The calculation formula is as follows:

[0123]

[0124] In the formula, E gh V represents the power consumption of the replacement inspection task. fh This indicates the return speed.

[0125] The task power consumption replacement decision is that each time the quadruped robot is in the process of an inspection task, when faced with a power consumption state, it chooses to either replace the inspection task power consumption or not. If the power consumption is abnormal, the inspection task power consumption is replaced; if the power consumption is normal, the inspection task power consumption is not replaced.

[0126] Step c2: Generate a set of decision training data based on the replacement decision data.

[0127] It should be noted that: the decision training data set refers to generating a set of decision training data corresponding to each set of replacement decision data; the decision training data set is generated based on all replacement decision data.

[0128] The decision training data includes the current state, selected action, reward value, and next state corresponding to each set of replacement decision data;

[0129] In practice, the method for obtaining the decision training data set based on the replacement decision data is as follows:

[0130] The current state is defined by the robot's battery level, the power consumption value of each set of replacement decision data, and the power consumption of the replacement inspection task.

[0131] The action to be selected is to replace the task power consumption in the decision data with the action of selection.

[0132] Calculate the reward value for each set of replacement decision data after selecting an action; wherein the reward value is calculated as follows:

[0133] Let K be the power consumption value of the m-th industrial device, ZL be the robot mass, and XS be the power consumption of the replacement inspection task; then the formula for calculating the reward value Q is:

[0134] Q=h×[γ×(K-XS)-θ×ZL+μ×XS+RW]+C×(1-h);

[0135] Where h = 0 or h = 1, for example, when the task power consumption replacement decision is not to select the replacement inspection task power consumption, then h = 0; when the task power consumption replacement decision is to select the replacement inspection task power consumption, then h = 1. In the formula, γ represents the preset proportional score, θ represents the weight data of robot quality, μ represents the weight data of replacement inspection task power consumption, and C is a constant correction score, which is corrected by the staff based on experience.

[0136] It is understandable that when K-XS is smaller, that is, when the difference between the power consumption value and the power consumption of the replacement inspection task is smaller, when the power consumption of the quadruped robot is normal, it tends to h=0, that is, it tends not to select the power consumption of the replacement inspection task; when RW is smaller, it means that the task completion index is smaller, and when the power consumption of the quadruped robot is abnormal, it tends to make h=1, that is, select to use the power consumption of the replacement inspection task.

[0137] The power consumption of the next set of replacement decision data is the next state.

[0138] In practice, the methods trained to determine whether to use a power replacement decision model to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected include:

[0139] The current state and reward value in the decision training dataset are used as inputs to the power consumption replacement decision model. The power consumption replacement decision model randomly selects multiple sets of decision training data from the decision training dataset for training, and learns whether to choose to replace the power consumption of the inspection task under different power consumption states in order to obtain the maximum reward value. The power consumption replacement decision model is a deep learning model.

[0140] Step 3: Based on the minimum power consumption value predicted by the power consumption prediction model, determine whether there is a power consumption anomaly in the quadruped robot; if there is a power consumption anomaly, generate a power consumption anomaly instruction and proceed to step 4; if there is no power consumption anomaly, return to step 1.

[0141] Please see Figure 3 As shown, in practice, methods for determining whether a quadruped robot has abnormal power consumption include:

[0142] Obtain the remaining power consumption value of the quadruped robot, calculate the difference between the minimum power consumption value predicted by the power consumption prediction model and the remaining power consumption value, and obtain the power consumption difference value.

[0143] It should be noted that the remaining power consumption value includes the sum of the safe remaining amount and the remaining power consumption; the safe remaining amount is the minimum power consumption required for the quadruped robot to safely return to its initial position or charging station.

[0144] The power consumption difference is compared with a preset power consumption difference threshold;

[0145] If the power consumption difference is less than the preset power consumption difference threshold, a power consumption abnormality instruction is generated, indicating that the quadruped robot cannot complete the inspection task of the m-th industrial equipment.

[0146] If the power consumption difference is greater than or equal to the preset power consumption difference threshold, no power consumption abnormality instruction will be generated, indicating that the quadruped robot can complete the inspection task of the m-th industrial equipment.

[0147] Step 4: Receive the power consumption anomaly command, obtain the power consumption of the replacement inspection task, and output a decision on whether to use the power consumption of the replacement inspection task based on the power consumption of the replacement inspection task, the running power consumption, and the power consumption replacement decision model.

[0148] In implementation, methods for determining whether to use the replacement inspection task power consumption include:

[0149] The decision model for whether to use the power consumption of the replacement inspection task is obtained from the minimum power consumption value, the power consumption of the generated replacement inspection task, and the power consumption of the robot mass input of the quadruped robot to be controlled.

[0150] This embodiment utilizes a precise power consumption prediction model to calculate the minimum power consumption value in advance, providing a basis for inspection task planning and avoiding unnecessary high-power paths. When abnormal power consumption is detected, the power consumption replacement decision model quickly intervenes, determining whether to activate a power replacement for the inspection task based on real-time data, minimizing energy waste and extending the robot's single inspection endurance. In long-term, large-scale inspection operations, this reduces the frequency of charging, not only reducing energy costs but also ensuring inspection continuity and improving overall efficiency.

[0151] This embodiment relies on detailed task power consumption scoring and motion power consumption scoring methods to quantify power consumption in accordance with actual scenarios. During task execution, progress is dynamically tracked through a task completion index to accurately determine the completion status of the inspection task. Any power consumption anomalies are promptly adjusted to ensure no critical inspection steps are missed. A deep learning-based power replacement decision model continuously learns and optimizes, enabling the robot to make scientific power consumption decisions even under complex working conditions. This significantly reduces inspection interruptions caused by power consumption issues and comprehensively enhances the accuracy and completeness of inspection data.

[0152] Example 2

[0153] Please see Figure 4 As shown, this embodiment provides an adaptive knowledge-driven optimization system for intelligent inspection of quadruped robots. The system includes: a data acquisition module, a decision model training module, a judgment module, and a decision module. The modules are connected via wired and / or wireless means to realize data transmission between the modules.

[0154] The data acquisition module is used to acquire the inspection task data of the quadruped robot and input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value.

[0155] The decision model training module is used to acquire the decision training data set of the quadruped robot; based on the decision training data set, a power replacement decision model is trained to determine whether to use the power replacement function to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected.

[0156] The judgment module determines whether the quadruped robot has power consumption anomalies based on the minimum power consumption value predicted by the power consumption prediction model. If a power consumption anomaly exists, a power consumption anomaly command is generated and transferred to the decision module. If no power consumption anomaly exists, the data acquisition module is returned.

[0157] The decision module is used to receive power consumption anomaly commands, obtain the power consumption of the replacement inspection task, and output a decision on whether to use the power consumption of the replacement inspection task based on the power consumption of the replacement inspection task, the running power consumption, and the power consumption replacement decision model.

[0158] All the formulas mentioned above are calculated using only numerical values ​​after removing dimensions. These formulas were derived through the collection of massive amounts of data and simulation using software, and are the closest to reality. The weight data in the formulas, as well as the various preset thresholds in the analysis process, were all set by professionals in the field based on actual conditions, or derived through simulation and derivation of a large amount of data.

[0159] The main purpose of weighted data is to quantify various parameters to obtain specific numerical values, facilitating subsequent comparative analysis. The value of the weighted data depends on the size of the sample data and the pre-set processing score for each set of sample data by the technical staff. When setting weighted data, it is sufficient to ensure that it does not affect the proportional relationship between the parameter and the quantified value.

[0160] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0161] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. 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.

Claims

1. An adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots, characterized in that, include: Step 1: Obtain the inspection task data of the quadruped robot, which includes motion power consumption score and task power consumption score; The motion power consumption score is obtained by combining data on the robot's environment, motion speed, acceleration, and motion type; the task power consumption score is obtained by combining data on the robot's environment, motion speed, acceleration, and motion type. Step a11: Obtain the task data of the quadruped robot. The task data includes the detection task type and the task execution time. The detection task type includes equipment appearance inspection task, equipment operating status inspection task, and equipment internal parameter measurement task. Step a12: Analyze the task data to obtain the task power consumption score; input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value; Step 2: Obtain the decision training dataset for the quadruped robot; methods for obtaining the decision training dataset include: Step c1: Obtain replacement decision data for the quadruped robot, including the power consumption value of the m-th industrial device, the task completion index, the power consumption of the replacement inspection task, the robot mass, and the task power consumption replacement decision; Step c2: Generate a decision training dataset based on the replacement decision data; the method for obtaining the task completion index includes: Obtain task completion feature data for the m-th industrial device. This data includes the inspection task completion rate and the data transmission completion rate. The inspection task completion rate is the ratio of the number of completed inspection tasks for the m-th industrial device by the quadruped robot to the preset number of completed inspection tasks. The data transmission completion rate is the ratio of the data transmission of the quadruped robot for the completed inspection task for the m-th industrial device to the preset data transmission. The replacement inspection task power consumption is the power consumption when stopping the inspection task and returning to the starting point. Based on the decision training dataset, train a power replacement decision model to determine whether to use the replacement inspection task power consumption when an abnormal power consumption of the quadruped robot is detected. Step 3: Based on the lowest power consumption value predicted by the power consumption prediction model, determine whether the quadruped robot has power consumption anomalies; methods for determining whether the quadruped robot has power consumption anomalies include: Obtain the remaining power consumption value of the quadruped robot, which includes the safe remaining amount and the sum of the remaining power consumption; the safe remaining amount is the minimum power consumption required for the quadruped robot to safely return to its initial position or charging station; calculate the difference between the minimum power consumption value predicted by the power consumption prediction model and the remaining power consumption value to obtain the power consumption difference. The power consumption difference is compared with a preset power consumption difference threshold. If the power consumption difference is less than the preset power consumption difference threshold, a power consumption abnormality command will be generated. If the power consumption difference is greater than or equal to the preset power consumption difference threshold, no power consumption abnormality instruction is generated; if a power consumption abnormality exists, a power consumption abnormality instruction is generated and the process proceeds to step 4; if no power consumption abnormality exists, the process returns to step 1. Step 4: Receive the power consumption anomaly command, obtain the power consumption of the replacement inspection task, replace the power consumption of the replacement inspection task, the minimum power consumption value, and the robot quality input power consumption in the replacement decision model, and output the decision on whether to select to use the power consumption of the replacement inspection task.

2. The adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots according to claim 1, characterized in that, The training method for the power consumption prediction model is as follows: The historical inspection task training data is divided into a task training set and a task test set to construct a regression network model. The historical inspection task training data includes inspection task data and the minimum power consumption value corresponding to the inspection task data. Using inspection task data from the task training set as input and the corresponding minimum power consumption value as output, both are fed into a regression network model for training to construct an initial regression network model. The initial regression network model is a neural network model, and the training process focuses on minimizing the sum of prediction accuracies. After the initial training is completed, the initial regression network model is evaluated using a task test set. If the sum of prediction accuracies achieved by the initial regression network model in the task test set is lower than a preset threshold, then the initial regression network model is established as a power consumption prediction model. Conversely, if the sum of prediction accuracies reaches or exceeds the preset threshold, the task training set is re-inputted, and the initial regression network model is continuously iterated and trained until the sum of prediction accuracies obtained in the testing phase meets the preset standard.

3. The adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots according to claim 2, characterized in that, The method for obtaining the minimum power consumption value corresponding to the inspection task data in the historical inspection task training data is as follows: Step b1: Obtain the robot's mass and total running time; Step b2: Formulate the inspection task data, robot quality, and total running time of the m-th industrial equipment to obtain the power consumption evaluation score of the m-th industrial equipment; m=1,2,...,M; M represents the total number of industrial equipment; Step b3: Let m = m + 1, repeat the above steps until m = M, then end the loop and mark the minimum power consumption evaluation score corresponding to the industrial equipment as the minimum power consumption value.

4. The adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots according to claim 3, characterized in that: The task completion status is determined based on the task completion index. The task completion status includes inspection task not completed and inspection task completed. Methods for determining task completion status based on a task completion index include: A preset task index threshold is set, and the task completion index is compared with the task index threshold. When the task completion index is less than or equal to the task index threshold, the task completion status will be generated as "incomplete inspection task". When the task completion index is greater than the task index threshold, the task completion status will be generated as inspection task completed. The task power consumption replacement decision is that each time the quadruped robot faces a power consumption state during an inspection task, it chooses or does not choose to use one of the replacement inspection task power consumption options.

5. The adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots according to claim 4, characterized in that, The decision training data set refers to a set of decision training data corresponding to each set of replacement decision data; a decision training data set is generated based on all replacement decision data. The decision training data includes the current state, selected action, reward value, and next state corresponding to each set of replacement decision data; The method for generating a decision training dataset based on replacement decision data is as follows: The current state is defined by the robot's battery level, the power consumption value of each set of replacement decision data, and the power consumption of the replacement inspection task. The action to be selected is to replace the task power consumption in the decision data with the action of selection. Calculate the reward value after selecting an action for each set of replacement decision data. ; The power consumption of the next set of replacement decision data is the next state.

6. The adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots according to claim 5, characterized in that, Methods for training a power substitution decision model that determines whether to use a power replacement model to replace the power consumption of an inspection task when an abnormal power consumption of a quadruped robot is detected include: The current state and reward value in the decision training dataset are used as inputs to the power consumption replacement decision model. The power consumption replacement decision model randomly selects multiple sets of decision training data from the decision training dataset for training, and learns whether to choose to replace the power consumption of the inspection task under different power consumption states in order to obtain the maximum reward value. The power consumption replacement decision model is a deep learning model.

7. An adaptive knowledge-driven optimization system for intelligent inspection of quadruped robots, used to implement the adaptive knowledge-driven optimization method for intelligent inspection of quadruped robots as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire the inspection task data of the quadruped robot and input the inspection task data into the pre-built power consumption prediction model to obtain the lowest power consumption value. The decision model training module is used to acquire the decision training data set of the quadruped robot; based on the decision training data set, a power replacement decision model is trained to determine whether to use the power replacement function to replace the power consumption of the inspection task when an abnormal power consumption of the quadruped robot is detected. The judgment module determines whether the quadruped robot has abnormal power consumption based on the lowest power consumption value predicted by the power consumption prediction model. If a power consumption anomaly is detected, a power consumption anomaly command is generated and transferred to the decision-making module; if no power consumption anomaly is detected, the process returns to the data acquisition module. The decision module is used to receive power consumption anomaly commands, obtain the power consumption of the replacement inspection task, and output a decision on whether to use the power consumption of the replacement inspection task based on the power consumption of the replacement inspection task, the running power consumption, and the power consumption replacement decision model.

Images