Self-balancing control method and system for automatic telescopic wheel carrier of power pipeline robot

Abstract

The invention discloses a self-balancing control method and system for an automatic telescopic wheel carrier of a power pipeline robot. The method comprises the steps that multi-sensor data are collected and fused to obtain robot state information; outputting a control signal based on a PID control algorithm, and dynamically adjusting the length of each wheel carrier through the control signal to form an initial balance strategy; adjusting the wheel carrier telescoping priority based on an intelligent weighted distribution algorithm; the initial balance strategy is optimized, and rapid adaptation of the power pipeline robot to the complex pipeline environment is achieved; obtaining a balance state prediction value based on an LSTM prediction algorithm; and dynamically adjusting a robot walking path and a wheel carrier control strategy based on the balance state prediction value. The system comprises a sensor module, a data processing unit, a central control module and a wheel carrier driving module. According to the invention, the stability and adaptability of the power pipeline robot in complex, zigzag and irregular pipeline environments are improved; autonomous balance adjustment of the robot is achieved, and manual intervention is reduced.

Description

Technical Field

[0001] The present invention relates to a robot balance control method, and particularly to an automatic telescopic wheel frame self-balancing control method and system for a power pipeline robot. Background Art

[0002] With the rapid development of intelligent robot technology, the application of robots in the fields of power pipeline detection, maintenance, repair, etc. has gradually increased. The power pipeline robot needs to be able to walk stably inside complex, narrow and tortuous pipelines to complete efficient detection and repair tasks. Due to the large diameter, slope and surface irregularity of the pipelines, the movement of the robot in the pipeline is easily affected, and traditional fixed structures often have difficulty adapting to this environment, resulting in problems such as unstable walking and jamming. Therefore, the robot needs to have a certain adaptive ability to adjust its own motion posture and structure according to the real-time environment, especially in terms of the walking stability and adaptability of the robot.

[0003] Most of the existing pipeline robots rely on fixed wheel frames or traditional wheel designs. Although these designs can provide certain walking capabilities, in complex pipeline environments (such as irregular pipeline diameters, wear or corrosion on the pipeline surface, etc.), the robot is prone to tilt, jam or lose balance, resulting in the robot being unable to pass through the pipeline smoothly or complete the task. In addition, the existing pipeline robots often lack an adaptive balance control mechanism and cannot be dynamically adjusted in real time according to changes in the pipeline environment, resulting in poor walking stability of the robot in different pipeline environments.

[0004] In order to overcome the above problems, a new control method and structural design are needed to enable the power pipeline robot to adaptively perform balance adjustment in complex environments. The introduction of self-balancing control technology can, through means such as automatic telescopic wheel frames, sensor feedback, and closed-loop control, sense and adjust the posture of the robot in real time to ensure its stable operation under various complex pipeline conditions. Summary of the Invention

[0005] Object of the Invention: The object of the present invention is to provide an automatic telescopic wheel frame self-balancing control method for a power pipeline robot to solve the problem that existing robots cannot effectively maintain balance and walk stably in complex pipeline environments. Another object of the present invention is to propose an automatic telescopic wheel frame self-balancing control system for a power pipeline robot to solve the problem of how to execute the above method.

[0006] Technical Solution: An automatic telescopic wheel frame self-balancing control method for a power pipeline robot according to the present invention includes the following steps:

[0007] Collect multi-sensor data of the power pipeline robot and fuse the multi-sensor data to obtain the robot state information;

[0008] Based on the PID control algorithm, input the robot state information, output the telescopic adjustment length of each wheel frame as the control signal, and dynamically adjust the length of each wheel frame through the control signal to form the initial balance strategy;

[0009] Adjust the telescopic priority of the wheel frame based on the intelligent weighted distribution algorithm;

[0010] Utilize the reinforcement learning algorithm to optimize the initial balance strategy and enable the power pipeline robot to quickly adapt to complex pipeline environments;

[0011] Based on the LSTM prediction algorithm, input the sequence of robot state information collected in real time, and output the predicted value of the balance state of the power pipeline robot in the walking path;

[0012] Dynamically adjust the robot walking path and wheel frame control strategy based on the predicted balance state value.

[0013] The automatic telescopic wheel frame self - balancing control technology involved in the present invention enables the robot to achieve fast and autonomous balance adjustment in different pipeline environments through algorithms such as multi - sensor fusion real - time monitoring, closed - loop feedback adaptive control, intelligent weighted distribution regulation, and reinforcement learning. The sensor data is used to collect and process information such as the robot's attitude, center of gravity, and contact pressure in real time, and the telescopic and movement of the wheel frame are adjusted in real time through the closed - loop feedback control algorithm, thereby realizing the stable walking of the robot in complex environments.

[0014] In addition, the present invention also introduces a prediction algorithm based on LSTM (Long Short - Term Memory Network). Through learning historical data, the robot can predict the future balance state and optimize the walking path, improving the smoothness and efficiency of walking. Therefore, the present invention not only solves the stability problem of traditional pipeline robots in complex environments, but also improves the robot's autonomous learning ability, making it highly adaptable and robust in irregular pipelines and changing environments.

[0015] Preferably, the multi - sensor data includes synchronously collecting the robot's gyroscope, acceleration, tilt angle, and contact pressure data through multiple sensors, performing low - pass filtering on the acceleration and gyroscope data to filter out high - frequency noise; applying median filtering to the pressure and inclination data to remove outliers;

[0016] The robot state information includes tilt angle, center of gravity offset, and motion state.

[0017] Preferably, the fusion of multi - sensor data includes:

[0018] Determine the weight ω of each sensor according to the following formula i :

[0019]

[0020] Among them, ω i is the weight of sensor i; is the noise variance of sensor i, indicating data reliability;

[0021] The fused robot state estimate value x is calculated by the weighted average algorithm f , and the calculation formula is:

[0022]

[0023] Among them, x i is the output data of sensor i.

[0024] Preferably, the PID control algorithm includes:

[0025] According to the robot state information, the error e(t) between the current state and the target state is calculated in real time:

[0026] e(t) = θ current - θ target

[0027] Among them, θ current is the current tilt angle of the robot, provided by the data collected in real time by sensors (such as acceleration sensors or gyroscopes), reflecting the current attitude state of the robot; θ target is the tilt angle representing the robot target, generally a preset value, used to maintain the tilt angle of the robot in the ideal balance state; t represents the current time point, used to track the real-time state.

[0028] According to e(t), a control signal u(t) is generated using the PID formula:

[0029]

[0030] Among them, K p is the proportional gain, K i is the integral gain, K d is the derivative gain;

[0031] Proportional control: Generate a control signal directly according to the current error, quickly responding to the change of the tilt angle;

[0032] Integral control: Accumulate historical errors and correct long-term deviations;

[0033] Derivative control: Detect the change trend of the error and respond in advance to rapid tilt changes;

[0034] The control signal u(t) is restricted within the retractable range of the wheel frame.

[0035] Preferably, dynamically adjusting the length of each wheel frame through a control signal to form an initial balancing strategy includes:

[0036] According to the tilt angle and center of gravity offset fed back by the sensor, the length of each wheel frame is dynamically adjusted. The initial balancing strategy is:

[0037] Center of gravity offset adjustment: When the center of gravity offset ΔG>0, increase the length of the wheel frame in the direction opposite to the offset direction and shorten the length of the wheel frame in the offset direction;

[0038] Tilt angle adjustment: When the tilt angle θ>0, the horizontal position is restored by adjusting the length of the relative high-side and low-side wheel frames;

[0039] The telescopic length of each wheel frame of the robot is adjusted according to the following formula:

[0040] L i =L i +u i (t)

[0041] Among them, u i (t) is the telescopic length adjustment value of the wheel frame i calculated and output by the PID control algorithm; u of the low side wheel frame i (t)>0, u of the high side wheel frame i (t)<0;

[0042] Determine the tilt direction: Determine the tilt direction based on θ and ΔG;

[0043] The PID algorithm calculates u for each wheel frame separately. i (t), and update the telescopic length L of wheel frame i i .

[0044] Preferably, the adjusting the wheel frame telescopic priority based on the intelligent weighted allocation algorithm comprises:

[0045] Obtain the robot's tilt angle θ, center of gravity offset ΔG, and contact pressure P of wheel frame i i ;

[0046] Calculate the dynamic weight ω of wheel frame i i , the calculation formula is:

[0047]

[0048] in, is the effect of the tilt angle on the wheel frame, θ i is the tilt angle component corresponding to wheel carrier i; θ j is the inclination angle component corresponding to the jth wheel frame; n is the number of adjustable wheel frames.

[0049] The influence of the center-of-gravity shift on the wheel carrier i, ΔG i Represents the influence component of the center-of-gravity shift amount on the wheel carrier i, ΔG j Is the center-of-gravity shift amount corresponding to the j-th wheel carrier; n is the number of adjustable wheel carriers.

[0050] The influence of the contact pressure on the wheel carrier i, P i Represents the contact pressure of the wheel carrier i, P j Is the contact pressure of the j-th wheel carrier; n is the number of adjustable wheel carriers.

[0051] α, β, and γ are the weight coefficients of each factor, dynamically adjusted according to the task requirements;

[0052] Calculate the telescopic adjustment signal u of the wheel carrier i i , and the calculation formula is:

[0053] u i = ω i ·ΔL

[0054] Among them, ΔL is the calculated total adjustment amount; the wheel carrier with a higher weight is preferentially adjusted with a larger amplitude.

[0055] Preferably, the optimization of the initial balance strategy using the reinforcement learning algorithm includes:

[0056] Adopt Q-Learning to optimize the initial balance strategy, and the update formula of Q-Learning is:

[0057]

[0058] Among them, Q(S t ,A t ) is the Q value of executing the action A t in the state S t ;

[0059] α is the learning rate, controlling the update step size;

[0060] R t is the current reward value;

[0061] γ is the discount factor, measuring the weight of future rewards;

[0062] S t+1 is the new state after executing the action;

[0063] is the Q value of the optimal future action;

[0064] S is the current environmental state of the robot, including the tilt angle θ, the center of gravity offset ΔG, the telescopic length Li of the wheel carriage i, the contact pressure Pi of the wheel carriage i, and the pipeline environmental parameters; i , the contact pressure Pi of the wheel carriage i i and pipeline environmental parameters;

[0065] A is the controllable action that the robot can select, including adjusting the telescopic length ΔLi of a specific wheel carriage i, increasing or decreasing the total telescopic force of the wheel carriage; i , increasing or decreasing the total telescopic force of the wheel carriage;

[0066] R is to calculate the reward value according to the state change after the action execution, used to evaluate the action effect, and the calculation formula is:

[0067] R = -(|θ| + δ|ΔG| + β∑|Pi - P0|) i -P0 target |)

[0068] where, |θ| is the absolute value of the tilt angle; |ΔG| is the absolute value of the center of gravity offset; |Pi - P0| is the deviation between the contact pressure of each wheel carriage and the target pressure; δ and β are both adjustment coefficients. i -P0 target | is the deviation between the contact pressure of each wheel carriage and the target pressure; δ and β are both adjustment coefficients.

[0069] Preferably, the LSTM prediction algorithm includes:

[0070] The input of the LSTM network is the sequence of real-time collected robot state information, and the output is the predicted value of the balance state of the power pipeline robot in the walking path;

[0071] LSTM network input layer:

[0072] The input sequence X t ={xt, xt - 1,..., xt - k + 1}, where t is the current moment, k is the time window length, and each input data includes the tilt angle, the center of gravity offset, and the pressure distribution; t-k ,xt - 1 t-k+1 ,...,xt - k + 1 t}

[0073] LSTM network hidden layer:

[0074] Adopt 2 layers of LSTM cells, each layer contains 64 hidden units, to capture the temporal dependence relationship;

[0075] LSTM cell formula:

[0076] ft t = σ(Wf · [ht - 1, xt] + bf)(forget gate) f ·[ht - 1 t-1 ,xt t +bf f )(forget gate)

[0077] it t = σ(Wi · [ht - 1, xt] + bi) i·[h t-1 ,x t +b i )(Input gate)

[0078]

[0079] o t =σ(W o ·[h t-1 ,x t +b o )(Output gate)

[0080] h t =o t ·tanh(C t )(Hidden state output)

[0081] LSTM network output layer:

[0082] Output the predicted values of the balance state for the next n steps, including:

[0083]

[0084] Among them, the tilt angle Predict the change in the tilt angle of the robot in the future to ensure a stable posture; the center-of-gravity offset Predict the future change in the center of gravity and adjust the wheel frame in advance to avoid imbalance; the path planning adjustment amount Dynamically optimize the telescopic length of the wheel frame according to the prediction results; the contact pressure The contact pressure at the future moment.

[0085] Furthermore, the dynamic adjustment of the robot walking path and the wheel frame control strategy based on the predicted balance state values includes:

[0086] Adjust the length of the wheel frame in advance according to the predicted tilt angle and center-of-gravity offset, and the formula is:

[0087]

[0088] Among them, L i is the telescopic length of the wheel frame i; k θ is the adjustment coefficient of the tilt angle: k G is the adjustment coefficient of the center-of-gravity offset;

[0089] Optimize the walking direction according to the prediction result of the pressure distribution and choose the direction with the least force to walk.

[0090] On the other hand, the present invention discloses an automatic telescopic wheel frame self - balancing control system for a power pipeline robot, including a sensor module arranged at the wheel frame support points of the robot. The sensor module is used to collect and process the robot's attitude, acceleration, tilt angle, and contact pressure data and perform data fusion on these data through a weighted average algorithm;

[0091] A data processing unit for generating robot state information from the fused sensor data;

[0092] A central control module for receiving the robot state information, outputting a control signal based on the PID control algorithm, and dynamically adjusting the length of each wheel frame through the control signal to ensure rapid balance restoration when tilt or instability is detected;

[0093] The central control module calculates the weights of different wheel frames and dynamically assigns telescopic priorities according to the center - of - gravity offset situation and tilt angle;

[0094] The central control module gradually optimizes the balance control strategy through the feedback of multiple walking data, and finally realizes rapid adaptation to complex pipeline environments;

[0095] The central control module predicts the balance state of the robot in the next few seconds through the LSTM network, and adjusts the telescopic length and angle of the wheel frame in advance according to the predicted data to improve walking stability;

[0096] A wheel frame drive module for receiving the control signal and driving the motor to adjust the telescopic length of the wheel frame.

[0097] Advantageous effects: Compared with the prior art, the present invention has the following remarkable advantages:

[0098] Through multi - sensor fusion technology, the robot of the present invention can obtain key information such as tilt angle, center - of - gravity position, and contact pressure in real time, and through a closed - loop feedback control system, automatically adjust the telescopic length and direction of the automatic telescopic wheel frame to ensure that the robot maintains balance in an irregular pipeline environment. The intelligent weighted algorithm dynamically assigns telescopic priorities according to real - time data, improving the accuracy and response speed of balance adjustment.

[0099] The introduction of reinforcement learning and the LSTM network enables the robot to optimize the balance control strategy and path planning through autonomous learning, improving the robot's adaptability and efficiency in unknown or changing environments. The self - learning balance adjustment algorithm can continuously accumulate and optimize from experience, making the robot walk more smoothly and quickly in the pipeline.

[0100] The present invention improves the stability and adaptability of the power pipeline robot in complex, tortuous and irregular pipeline environments; realizes the autonomous balance adjustment of the robot, reduces manual intervention; and improves the efficiency and safety of the robot when performing pipeline detection and repair tasks. The present invention provides efficient and reliable technical support for the intelligent detection and repair of power pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 It is a schematic flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0102] The technical solution of the present invention will be further described below in conjunction with the drawings.

[0103] As Figure 1 shown, an automatic telescopic wheel frame self-balancing control method for a power pipeline robot includes the following steps:

[0104] (1) Arrange the sensor module at the wheel frame support points of the robot, collect key information such as the posture, acceleration, tilt angle, contact pressure, etc. of the robot through multiple sensors, and perform data fusion through the weighted average algorithm to obtain accurate robot state information. The specific method is as follows:

[0105] Collection and preprocessing of multi-sensor data:

[0106] Acceleration sensor: Real-time collect linear acceleration a x , a y , a z , for estimating the motion state of the robot.

[0107] Gyroscope sensor: Measure angular velocity ω x , ω y , ω z , to obtain the rotational motion information of the robot.

[0108] Pressure sensor: Collect the pressure value P of the wheel frame in contact with the pipe wall i , for monitoring the change of the center of gravity.

[0109] Inclinometer sensor: Real-time measure the tilt angle θ of the robot and provide attitude data.

[0110] These data are synchronously collected by the sensor acquisition module, low-pass filtering is performed on the acceleration and gyroscope data to filter out high-frequency noise; median filtering is applied to the pressure and inclination data to remove outliers. The purpose of data preprocessing is to remove noise and outliers and ensure the reliability of the input data.

[0111] Fusion of multi-sensor data:

[0112] The weighted average algorithm assigns weights ω to each sensor datai , the weights are dynamically adjusted according to the real-time reliability and accuracy of the sensors, thereby achieving data fusion.

[0113] The weight ω of each sensor is determined according to the following formula i :

[0114]

[0115] where ω i is the weight of sensor i; is the noise variance of sensor i, representing data reliability;

[0116] The estimated value x of the robot state after fusion is calculated by the weighted average algorithm f , and the calculation formula is:

[0117]

[0118] where x i is the output data of sensor i, and x f is the fused data (such as tilt angle, acceleration, etc.).

[0119] Real-time weight dynamic adjustment:

[0120] The weight dynamic adjustment is based on the real-time feedback of sensor data reliability. When the robot is in a stable state, the weight of the gyroscope is relatively high for accurately measuring the angular velocity; when the environmental vibration is large, the weight of the acceleration sensor is increased to better reflect the motion state; the weight of the pressure sensor is adjusted according to the change of the contact point pressure to reflect the center of gravity shift.

[0121] For example: the tilt angle data provided by the acceleration sensor is θ a = 10°, and the noise variance is

[0122] The tilt angle data provided by the gyroscope is θ g = 11°, and the noise variance is

[0123] Calculate the weights:

[0124]

[0125] Calculate the fused tilt angle:

[0126] θ f = ω a θ a + ω g θ g = 0.33×10° + 0.67×11° = 10.67°

[0127] The fused multi-sensor data generates robot state information through the data processing unit, including:

[0128] Tilt angle: The fusion result is like θ f , which is used for attitude estimation;

[0129] Center of gravity offset: According to the pressure sensor data, the real-time change of the robot's center of gravity is calculated;

[0130] Motion state: Based on the acceleration and gyroscope data, the robot's speed and direction are estimated.

[0131] (2) Closed-loop feedback adaptive control system

[0132] After receiving the robot state information, the central control system dynamically adjusts the telescopic length of each wheel frame through the PID (Proportional-Integral-Derivative) control algorithm.

[0133] The system feeds back data once per second to ensure rapid balance restoration when tilt or instability is detected. By controlling the automatic adjustment of each wheel frame, the robot can maintain stable walking in pipe environments with different slopes and diameters.

[0134] PID control algorithm input and output:

[0135] Input the robot state information, including:

[0136] Tilt angle θ: Obtained by fusing the acceleration sensor, gyroscope, and inclination sensor;

[0137] Center of gravity offset ΔG: Calculated from the pressure sensor data;

[0138] Telescopic position L of the wheel frame i : The actual telescopic length of the current wheel frame.

[0139] Output the control signal u(t) to adjust the telescopic length of the wheel frame to restore the center of gravity and attitude.

[0140] Implementation of the PID control algorithm

[0141] According to the sensor fusion data, the error between the current state and the target state is calculated in real time:

[0142] e(t) = θ current - θ target

[0143] where, θ current is the current tilt angle of the robot, provided by the data collected in real time by sensors (such as acceleration sensors or gyroscopes), reflecting the current attitude state of the robot; θ targetIt represents the tilt angle of the robot's target, which is generally a preset value and is used to maintain the tilt angle of the robot in an ideal balanced state; t represents the current time point and is used to track the real-time state.

[0144] Generate the control signal u(t) using the PID formula based on the error e(t):

[0145]

[0146] Among them, K p is the proportional gain, K i is the integral gain, K d is the derivative gain;

[0147] Proportional control: Generate the control signal directly according to the current error, and quickly respond to the change of the tilt angle.

[0148] Integral control: Accumulate the historical error and correct the long-term deviation (the small offset caused by the hardware limitation of the wheel frame).

[0149] Derivative control: Detect the change trend of the error and respond to the rapid tilt change in advance.

[0150] The control signal u(t) is restricted within the retractable range of the wheel frame: u(t) ∈ [L min , L max

[0151] Among them, L min and L max are the minimum and maximum retractable lengths of the wheel frame respectively.

[0152] Transmit the control signal u(t) to the wheel frame drive module to drive the motor to adjust the retractable length of the wheel frame.

[0153] Dynamically adjust the lengths of each wheel frame through the control signal, and the initial balance strategy includes:

[0154] According to the tilt angle and the center of gravity offset amount feedback by the sensor, dynamically adjust the lengths of each wheel frame, and the specific strategy is as follows:

[0155] Center of gravity offset adjustment: When the center of gravity offset amount ΔG > 0, increase the length of the wheel frame opposite to the offset direction and shorten the length of the wheel frame in the offset direction.

[0156] Tilt angle adjustment: When the tilt angle θ > 0, restore the level by adjusting the lengths of the relatively high-side and low-side wheel frames.

[0157] The robot has 4 wheel frames, namely L1, L2, L3, L4, and the retractable length of each wheel frame is adjusted according to the following formula:

[0158] L i = L​i +u i (t)

[0159] Indicates the length of each new bogie plus u i , is the replacement value, where u i (t) is the adjustment value of the telescopic length of bogie i calculated by the PID control algorithm; for the low-side bogie, u i (t) > 0 (elongation), for the high-side bogie, u i (t) < 0 (shortening).

[0160] Judge the tilt direction: Judge the tilt direction according to θ and ΔG. If θ > 0 and the center of gravity is biased to the right, then the left-side bogies L1 and L3 need to be adjusted to shorten, and the right-side bogies L2 and L4 need to be adjusted to elongate. If θ < 0, the adjustment direction is opposite.

[0161] The PID algorithm calculates its control signal u i (t) for each bogie respectively, and updates the telescopic length L i of bogie i.

[0162] (3) Adjust the telescopic priority of the bogies based on the intelligent weighted distribution algorithm, specifically:

[0163] The central control system dynamically distributes the telescopic priority according to the calculated weights of different bogies and the center of gravity offset and tilt angle. The intelligent weighting mechanism of this algorithm plays a role in quickly restoring balance when the slope changes or the pipe diameter suddenly changes.

[0164] Obtain sensor data: including the tilt angle θ of the robot, the center of gravity offset ΔG, and the contact pressure P of bogie i i .

[0165] Calculate the dynamic weight ω i of bogie i, which reflects its priority in balance adjustment. The weight ω i is determined by multiple factors, and its calculation formula is:

[0166]

[0167] where is the influence of the tilt angle on the bogie, θ i is the tilt angle component corresponding to bogie i; θ j is the tilt angle component corresponding to the jth bogie; n is the number of adjustable bogies.

[0168] is the influence of the center of gravity offset on bogie i, ΔG i represents the influence component of the center of gravity offset on bogie i, ΔG jis the center-of-gravity offset corresponding to the j-th wheel bracket; n is the number of adjustable wheel brackets.

[0169] is the influence of the contact pressure on wheel bracket i, P i represents the contact pressure of wheel bracket i, P j is the contact pressure of the j-th wheel bracket; n is the number of adjustable wheel brackets.

[0170] α, β, and γ are the weight coefficients of each factor, which are dynamically adjusted according to the task requirements;

[0171] Allocate telescopic priorities: According to the weight ω i Allocate adjustment priorities to each wheel bracket.

[0172] Generate adjustment signals: Use the weights to guide the telescopic adjustment of the wheel brackets.

[0173] Calculate the telescopic adjustment signal u of wheel bracket i i , and the calculation formula is:

[0174] u i = ω i ·ΔL

[0175] where ΔL is the calculated total adjustment amount;

[0176] The wheel brackets with higher weights are given priority for larger adjustments.

[0177] (4) Reinforcement learning self-learning balance adjustment

[0178] During the training phase of the robot, the central control system uses the reinforcement learning algorithm to optimize the initial balance strategy, specifically:

[0179] After the initial balance strategy is generated, the robot continuously adjusts the wheel bracket expansion method during walking to adapt to the actual environment. Through the feedback of multiple walking data, the algorithm gradually optimizes the balance control strategy and finally achieves rapid adaptation to complex pipeline environments.

[0180] Use Q-Learning to optimize the initial balance strategy. Q-Learning is a value-function-based reinforcement learning method, and the update formula is as follows:

[0181]

[0182] where Q(S t ,A t ) is the Q value of executing action A t in state S t ;

[0183] α is the learning rate, which controls the update step size;

[0184] R t is the current reward value;

[0185] γ is the discount factor, measuring the weight of future rewards;

[0186] S t+1 is the new state after performing the action;

[0187] is the Q-value of the optimal future action; A’ is the action that can obtain the maximum Q-value in the new state S t+1 under which the maximum Q-value can be obtained.

[0188] S is the current environmental state of the robot, including the tilt angle θ, the center-of-gravity offset ΔG, the telescopic length L of the wheel carrier i i , the contact pressure P of the wheel carrier i i and the pipeline environmental parameters (such as slope, diameter change, etc.);

[0189] A is the control action that the robot can choose, including adjusting the telescopic length ΔL of a specific wheel carrier i i , increasing or decreasing the total telescopic force of the wheel carrier;

[0190] R is the reward value calculated according to the state change after the action execution, used to evaluate the action effect, and the calculation formula is:

[0191] R = -(|θ| + δ|ΔG| + β∑|P i -P target |)

[0192] where, |θ| is the absolute value of the tilt angle; |ΔG| is the absolute value of the center-of-gravity offset; |P i -P target | is the deviation between the contact pressure of each wheel carrier and the target pressure; both δ and β are adjustment coefficients.

[0193] (5) LSTM Balance Prediction and Path Optimization

[0194] The LSTM prediction algorithm is used to predict the balance state of the robot in the walking path in advance:

[0195] The central control system predicts the balance state of the robot in the next few seconds through the LSTM network, and adjusts the telescopic length and angle of the wheel carrier in advance according to the prediction data to improve the walking stability. This path optimization algorithm significantly improves the smoothness and energy efficiency of the robot.

[0196] LSTM Prediction Target:

[0197] Tilt Angle Predict the future tilt angle change of the robot to ensure a stable posture.

[0198] Center of gravity offset Predict the future change of the center of gravity and adjust the wheel frame in advance to avoid imbalance.

[0199] Path planning adjustment amount Dynamically optimize the telescopic length of the wheel frame according to the prediction result.

[0200] The input of the LSTM network is a sequence of robot state information collected in real time, and the output is the predicted value of the balance state of the power pipeline robot in the walking path;

[0201] LSTM network input layer:

[0202] Input sequence X t ={x t-k ,x t-k+1 ,...,x t}, where t is the current moment, k is the time window length, and each input data includes:

[0203] Tilt angle: θ t ; Center of gravity offset: ΔG; Pressure distribution: {P1, P2, P3, P4} t .

[0204] LSTM network hidden layer:

[0205] Adopt 2 layers of LSTM units, each layer contains 64 hidden units, to capture temporal dependencies;

[0206] LSTM unit formula:

[0207] f t =σ(W f ·[h t-1 ,x t +b f )(Forgetting gate)

[0208] i t =σ(W i ·[h t-1 ,x t +b i )(Input gate)

[0209]

[0210] o t =σ(W o ·[h t-1 ,x t +b o )(Output gate)

[0211] h t =o t ·tanh(C t(Hidden state output)

[0212] Output layer of the LSTM network:

[0213] Output the predicted values of the equilibrium state for the next n steps, including:

[0214]

[0215] Among them, the tilt angle Predict the change in the tilt angle of the robot in the future to ensure a stable posture; the center of gravity offset Predict the future change in the center of gravity and adjust the wheel frame in advance to avoid imbalance; the path planning adjustment amount Dynamically optimize the telescopic length of the wheel frame according to the prediction results; the contact pressure The contact pressure at the future moment.

[0216] Collect the time-series state data of the robot in different pipeline environments through sensors to construct a training dataset.

[0217] Dynamically adjust the walking path and wheel frame control strategy of the robot based on the predicted equilibrium state values. The dynamic adjustment strategy is specifically:

[0218] Adjust the length of the wheel frame in advance according to the predicted tilt angle and center of gravity offset. The formula is:

[0219]

[0220] Among them, L i is the telescopic length of the wheel frame i; k θ is the adjustment coefficient of the tilt angle: k G is the adjustment coefficient of the center of gravity offset; optimize the walking direction according to the prediction result of the pressure distribution and choose the direction with the least force to walk.

Claims

1. An automatic telescopic wheel frame self - balancing control method for a power pipeline robot, characterized in that, It includes the following steps: Collect multi-sensor data of the power pipeline robot, and fuse the multi-sensor data to obtain the robot state information; Based on the PID control algorithm, input the robot state information, output the telescopic adjustment length of each wheel frame as the control signal, and dynamically adjust the length of each wheel frame through the control signal to form an initial balance strategy; Adjust the telescopic priority of the wheel frame based on the intelligent weighted distribution algorithm; Use the reinforcement learning algorithm to optimize the initial balance strategy to enable the power pipeline robot to quickly adapt to complex pipeline environments; Based on the LSTM prediction algorithm, input the sequence of robot state information collected in real time, and output the predicted value of the balance state of the power pipeline robot in the walking path; Dynamically adjust the robot walking path and the wheel frame control strategy based on the predicted balance state value.

2. The automatic telescopic wheel frame self-balancing control method for the power pipeline robot according to claim 1, characterized in that, The multi-sensor data includes synchronously collecting the gyroscope, acceleration, tilt angle, and contact pressure data of the robot through multiple sensors, performing low-pass filtering on the acceleration and gyroscope data to filter out high-frequency noise; applying median filtering to the pressure and tilt angle data to remove outliers; The robot state information includes the tilt angle, center of gravity offset, and motion state.

3. The automatic telescopic wheel frame self-balancing control method of the power pipeline robot according to claim 1, characterized in that, The fusion of multi-sensor data includes: Determine the weight ω of each sensor according to the following formula i : where ω i is the weight of sensor i; is the noise variance of sensor i, representing data reliability; Calculate the fused robot state estimate value \(x\) through the weighted average algorithm f , and the calculation formula is: where x i is the output data of sensor i.

4. The automatic telescopic wheel frame self-balancing control method of the power pipeline robot according to claim 1, characterized in that, The PID control algorithm includes: According to the robot state information, calculate the error e(t) between the current state and the target state in real time: e(t) = θ current -θ target Among them, θ current is the current tilt angle of the robot; θ target is the tilt angle representing the target of the robot; t is Represents the current time point for tracking the real-time state; Generate the control signal u(t) using the PID formula according to e(t): Among them, K p is the proportional gain, K i is the integral gain, K d is the derivative gain; Proportional control: Generate the control signal directly according to the current error to quickly respond to the change in the tilt angle; Integral control: Accumulate historical errors to correct long-term deviations; Derivative control: Detect the change trend of the error and respond in advance to rapid tilt changes; The control signal u(t) is restricted within the telescopic range of the wheel frame.

5. The automatic telescopic wheel frame self-balancing control method for the power pipeline robot according to claim 2, characterized in that, The dynamic adjustment of the length of each wheel frame through the control signal to form an initial balance strategy includes: Dynamically adjust the length of each wheel frame according to the tilt angle and center of gravity offset feedback by the sensor, and the initial balance strategy is: Center of gravity offset adjustment: When the center of gravity offset ΔG>0, increase the length of the wheel frame opposite to the offset direction and shorten the length of the wheel frame in the offset direction; Tilt angle adjustment: When the tilt angle θ>0, restore the level by adjusting the lengths of the relatively high-side and low-side wheel frames; The telescopic length of each wheel frame of the robot is adjusted according to the following formula: L i = L i + u i (t) where, u i (t) is the adjustment value of the telescopic length of the wheel carriage i calculated by the PID control algorithm; for the lower side wheel carriage, u i (t) > 0, and for the upper side wheel carriage, u i (t) < 0; Judge the tilt direction: Judge the tilt direction according to θ and ΔG; The PID algorithm calculates its u i (t) for each bogie and updates the telescopic length L i of bogie i.

6. The automatic telescopic wheel frame self-balancing control method of the power pipeline robot according to claim 1, wherein The adjustment of the telescopic priority of the wheel frame based on the intelligent weighted distribution algorithm includes: Obtain the tilt angle θ of the robot, the center-of-gravity offset ΔG, and the contact pressure P of the wheel carrier i i ; Calculate the dynamic weight ω of the wheel carrier i i , and the calculation formula is as follows: Among them, is the influence of the tilt angle on the wheel carrier, θ i is the tilt angle component corresponding to the wheel carrier i; θ j is the tilt angle component corresponding to the j-th wheel carrier; n is the number of adjustable wheel carriers; Effect of center of gravity offset on wheel set i, ΔG i Component representing the effect of center of gravity offset on wheel set i, ΔG j Center of gravity offset corresponding to the j-th wheel set; n is the number of adjustable wheel sets; For the influence of the contact pressure on the wheel carrier i, P i Indicates the contact pressure of the wheel carrier i, P j Is the contact pressure of the j-th wheel carrier; n is the number of adjustable wheel carriers; α, β, γ are the weight coefficients of each factor, which are dynamically adjusted according to the task requirements; Calculate the telescopic adjustment signal u of the wheel carriage i i , and the calculation formula is: u i = ω i ·ΔL Among them, ΔL is the calculated total adjustment amount; the wheel frame with a higher weight is given priority to make a larger adjustment.

7. The automatic telescopic wheel frame self-balancing control method of the power pipeline robot according to claim 1, characterized in that The optimization of the initial balance strategy using the reinforcement learning algorithm includes: Use Q-Learning to optimize the initial balance strategy, and the update formula of Q-Learning is: Among them, Q(S t , A t ) is the Q-value of executing action A t in state S t ; α is the learning rate, which controls the update step size; R t is the current reward value; γ is the discount factor, which measures the weight of future rewards; S t+1 The new state after the execution of the action; The Q-value for the optimal action in the future; S is the current environmental state of the robot, including the tilt angle θ, the center-of-gravity offset ΔG, the telescopic length L of the wheel carrier i i , the contact pressure P of the wheel carrier i i and the pipeline environmental parameters; A is a control action that can be selected by the robot, including adjusting the telescopic length ΔL of a specific wheel bracket i i , increasing or decreasing the total telescopic force of the wheel bracket; R is the reward value calculated according to the state change after the action is executed, which is used to evaluate the action effect, and the calculation formula is: R = -(|θ| + δ|ΔG| + β∑|P i -P target |) where, |θ| is the absolute value of the tilt angle; |ΔG| is the absolute value of the center-of-gravity offset; |P i - P target | is the deviation between the contact pressure of each wheel set and the target pressure; both δ and β are adjustment coefficients.

8. The automatic telescopic wheel frame self-balancing control method for the power pipeline robot according to claim 1, characterized in that The LSTM prediction algorithm includes: The input of the LSTM network is a sequence of robot state information collected in real time, and the output is the predicted value of the balance state of the power pipeline robot in the walking path; Input layer of the LSTM network: Input sequence X t = {x t-k , x t-k+1 ,..., x t}, where t is the current time, k is the length of the time window, and each input data includes the tilt angle, the center of gravity offset, and the pressure distribution; Hidden layer of the LSTM network: Two layers of LSTM units are adopted, with each layer containing 64 hidden units to capture temporal dependencies; LSTM unit formula: f t = σ(W f · [h t-1 , x t + b f )(Forgotten Gate) i t = σ(W i · [h t-1 , x t + b i )(input gate) (Candidate status) (Unit status) o t = σ(W o ·[h t-1 , x t + b o (Output gate) h t = o t ·tanh(C t )(Hidden state output) Output layer of the LSTM network: Output the predicted values of the balance state for the next n steps, including: Among them, the tilt angle Predict the future change of the tilt angle of the robot to ensure a stable posture; the center-of-gravity offset Predict the future change of the center of gravity and adjust the wheel frame in advance to avoid imbalance; the path planning adjustment amount Dynamically optimize the telescopic length of the wheel frame according to the prediction result; the contact pressure The contact pressure at the future moment.

9. The automatic telescopic wheel frame self-balancing control method of the power pipeline robot according to claim 8, characterized in that, The dynamic adjustment of the robot walking path and wheel frame control strategy based on the predicted balance state values includes: Adjust the length of the wheel frame in advance according to the predicted tilt angle and center of gravity offset, and the formula is: Among them, L i is the telescopic length of the wheel carrier i; k θ is the adjustment coefficient of the tilt angle: k G is the adjustment coefficient of the center of gravity offset; Optimize the walking direction according to the pressure distribution prediction result and choose the direction with the least force to walk.

10. An automatic telescopic wheel frame self-balancing control system for a power pipeline robot, characterized in that, It includes a sensor module arranged at the wheel frame support points of the robot. The sensor module is used to collect and process the robot's attitude, acceleration, tilt angle, and contact pressure data and perform data fusion on these data through a weighted average algorithm; Data processing unit, used to generate robot state information from the fused sensor data; Central control module, used to receive the robot state information, output control signals based on the PID control algorithm, and dynamically adjust the length of each wheel frame through the control signals to ensure rapid recovery of balance when tilt or instability is detected; The central control module calculates the weights of different wheel frames and dynamically allocates the telescopic priority according to the center of gravity offset and tilt angle; The central control module gradually optimizes the balance control strategy through the feedback of multiple walking data, and finally realizes rapid adaptation to complex pipeline environments; The central control module predicts the balance state of the robot for the next few seconds through the LSTM network, and adjusts the telescopic length and angle of the wheel frame in advance according to the prediction data to improve walking stability; Wheel frame drive module, used to receive control signals and drive the motor to adjust the telescopic length of the wheel frame.

Images