A current prediction-based inspection robot slip identification and control method

Abstract

A kind of current prediction-based patrol robot skidding identification and control method, first, arrange experimental site, simulate the skidding situation that patrol robot can encounter in task, collect current data and carry out pre-processing by experiment;Build training network model, predict training to the current data collected, performance evaluation is carried out to the trained model;The change trend of future n current data is predicted using the model after training is completed, and whether the patrol robot has skidding phenomenon is judged;According to whether the patrol robot skids, make corresponding control strategy adjustment, reduce the influence brought by skidding.The present application predicts the probability of skidding in advance by prediction, and makes control strategy adjustment in advance.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of mobile robot travel control, and particularly relates to a mobile robot slip identification and control method. BACKGROUND

[0002] The inspection robot is a machine device for automatically performing a specified work task, and is generally used to replace human work and can work uninterruptedly for 24 hours, and is often used in dangerous places such as pipe corridors, chemical plants and power plants.

[0003] When the inspection robot performs a work task, the driving wheel output torque is greater than the tire adhesion force of the ground in some ground with a low friction coefficient (such as muddy and slippery road sections), which causes the driving wheel to slip, and the actual travel linear speed of the inspection robot is less than the linear speed of the driving wheel rotation, so that the inspection robot deviates from the original work path and affects the work, and in severe cases, the inspection robot may even roll over and other dangerous situations.

[0004] In the prior art, there are various methods for identifying the slip of the mobile robot, such as comparing the displacement output by the encoder odometer with the displacement output by the IMU, GPS and other sensors, and determining that the slip phenomenon occurs when the difference between the two exceeds a threshold value; continuously collecting the driving wheel motor current and the encoder speed signal, and determining whether the slip phenomenon occurs according to a certain change rate threshold; using multi-sensor fusion to improve the positioning accuracy, and determining whether the slip phenomenon occurs according to the relationship between the robot walking trajectory and the planned trajectory. As can be seen, the existing technology has a certain lag in determining the slip. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the present application provides a mobile robot slip identification and control method based on current prediction, which can predict the occurrence of slip to a certain extent and help the controller make appropriate control strategies earlier.

[0006] In order to solve the above technical problems, the present application provides the following technical solutions:

[0007] A mobile robot slip identification and control method based on current prediction, comprising the following steps:

[0008] Step S1, obtaining the parameters of the inspection robot, including the center of mass position, mass and running slope, and calculating the normal pressure of the driving wheel;

[0009] Step S2, arranging an experimental site to simulate the slip conditions that the inspection robot may encounter in the task, collecting current data through experiments and pre-processing the current data;

[0010] Step S3, build a training network model to predict and train the collected current data, and evaluate the performance of the trained model;

[0011] Step S4, use the trained model to predict the trend of the next n current data, and determine whether the inspection robot has slipped;

[0012] Step S5, adjust the control strategy according to whether the inspection robot slips or not, and reduce the impact of slipping.

[0013] Further, in step S1, the center of mass position of the inspection robot includes the distance from the center of mass to the front axle, the distance from the center of mass to the rear axle, and the height of the center of mass, which is obtained by SolidWorks software.

[0014] Preferably, in step S1, the mass of the inspection robot is obtained by weighing or three-dimensional software, and the running slope is obtained by converting the acceleration signal collected by the IMU sensor.

[0015] In step S1, after obtaining the related parameters of the inspection robot, the vertical load of the driving wheel is calculated based on the robot dynamics model.

[0016] Further, in step S2, the experimental site selects the common running site of the inspection robot, and sets the corresponding slope according to the technical requirements of the robot. The actual low ground adhesion coefficient scene is simulated by adding water or laying sand in different road sections.

[0017] Preferably, in step S2, the driving wheel motor data is collected by a servo driver, and is sent to the vehicle controller through CAN communication and printed output.

[0018] In step S2, the preprocessing process is:

[0019] (21) Based on the idea of sliding window, the current data is filtered by Gaussian weighted average method;

[0020] (22) The collected current data is single column time series data, which is re-divided according to the prediction step m and the prediction step n, and is divided into m+n column data;

[0021] (23) The divided data is normalized for network training.

[0022] Further, in step S3, the neural network structure used for training is TCN-GRU-ATTENTION network, and after training, the next n data can be predicted according to the first m data.

[0023] The trained model is evaluated in performance by root mean square error, mean absolute error and determination coefficient in the step S3.

[0024] In the step S4, the torque output by the motor is related to the current, the torque constant of the motor is obtained by querying the motor manual, and the reduction ratio is obtained by querying the reduction gear nameplate, and the motor torque is the product of the parameters.

[0025] In the step S4, after obtaining the motor output torque, the driving wheel angular acceleration information is collected through the motor encoder, and the longitudinal force size can be calculated according to the single wheel dynamics model, and the adhesion coefficient size is the ratio of the driving wheel longitudinal force and the vertical load in step S1.

[0026] In the step S4, the corresponding current change threshold is set to judge the slip condition of the inspection robot at this time.

[0027] In the step S5, when the slip occurs, the size of the motor output driving force is changed by adjusting the running speed of the robot, and the difference between the encoder wheel speed and the linear speed obtained by the IMU is adjusted, and when the difference is within a certain threshold, it is considered that the slip condition is improved.

[0028] The technical concept of the application is that when the inspection robot slips, the driving wheel motor adopts speed feedback control, so the speed is generally stable, and when the slip occurs, the friction coefficient decreases, the driving wheel load torque decreases, and the driving wheel motor current also decreases, and according to the different degrees of friction coefficient reduction, the degree of motor current reduction is also different.

[0029] The beneficial effects of the application are that the model trained in advance can predict the change trend of the current to judge the occurrence of the slip earlier, and the controller can be intervened and adjusted when the inspection robot does not appear serious deviation, so that the occurrence of more dangerous conditions is greatly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 It is the overall framework diagram of the application.

[0031] Figure 2 It is the force analysis diagram of the longitudinal climbing of the inspection robot in the application.

[0032] Figure 3 It is the network model training flowchart based on the current in the application.

[0033] Figure 4 It is the slip state judgment flowchart based on the current prediction in the application.

[0034] Figure 5 It is the single wheel force analysis diagram of the driving wheel in the application.

[0035] Figure 6 The speed adjustment strategy block diagram in the application. DETAILED DESCRIPTION

[0036] The application will be further described below with reference to the drawings.

[0037] Reference Figures 1-6 A slip identification and control method for an inspection robot based on current prediction, which improves the foresight of the slip identification of the inspection robot by introducing the idea of time series prediction.

[0038] In an embodiment of the application, a slip identification and control method for an inspection robot is provided, the drive wheel of the inspection robot is driven by a servo motor, the servo motor is driven by a servo driver, and the servo driver works in a speed loop.When the drive wheel is in a slip state, the motor encoder output value remains unchanged under the adjustment of the speed loop, while the tangential force of the ground on the drive wheel decreases, which is equivalent to the decrease of the motor load torque, at this time, the drive motor current will show a downward trend, according to which it can be judged whether the slip occurs.

[0039] To achieve the above object, as Figure 1 shown, the application provides a slip identification and control method for an inspection robot based on current prediction, comprising the following steps:

[0040] Step S1, obtaining the parameters of the inspection robot, including the mass center position, mass and running slope, calculating the normal pressure of the drive wheel, the process is as follows:

[0041] Step S11, in the SolidWorks software, the materials and masses of each part of the inspection robot are assigned, and the mass and gravity center position of the inspection robot are obtained through the "mass attribute" function;

[0042] The robot adopts symmetrical design, and the mass center can be assumed to be located at the center of the left and right drive wheel shafts, and the obtained parameters are mainly the distance a from the mass center to the front shaft, the distance b from the mass center to the rear shaft and the height h of the mass center;

[0043] Step S12, obtaining the acceleration information through the IMU sensor arranged on the inspection robot, and calculating the slope of the running road at this time through conversion;

[0044] Step S13, the longitudinal force analysis of the inspection robot when climbing a slope is as Figure 2 shown;

[0045] In the figure: G is the gravity of the robot; θ is the climbing angle; a is the distance from the mass center to the front shaft; b is the distance from the mass center to the rear shaft; h is the height of the mass center; F NF , F NRF fF , F fR F

[0046] The longitudinal dynamics calculation of the inspection robot is shown in the following formula:

[0047] F t = F f + F a + F i + F w

[0048] In the formula, F t is the driving force; F f is the rolling resistance; F a is the acceleration resistance; F i is the slope resistance; and F w is the air resistance.

[0049] The inspection robot is generally used for indoor operation and has a low running speed, so the air resistance can be ignored; the moment balance is taken at the front wheel contact point with the ground, and the calculation is shown in the following formula:

[0050]

[0051] Further, the vertical load size of the driving wheel is calculated, and the calculation is shown in the following formula:

[0052]

[0053] Step S2: as shown in the following formula, the present application provides a data acquisition and prediction model training process; Figure 3

[0054] An experimental site is arranged to simulate the slipping conditions that the inspection robot may encounter in the task, current data is collected through the experiment and is preprocessed, and the process is as follows:

[0055] Step S21, according to the scenes that may cause slipping during the operation of the inspection robot, such as water, oil stains, and sand, the corresponding experimental site is built, the task is issued to the inspection robot to make it run according to a certain trajectory, the motor current signals of the inspection robot during smooth running and running on the slipping road section are collected through the servo driver, the driver sends them to the vehicle controller VCU through CAN communication, and the VCU saves the data in the form of printing log files;

[0056] Step S22, due to the existence of various interferences, the collected current data needs to be filtered, the Gaussian weighted moving average method is adopted, the sliding window size is n, and the step is 1, and the process is as follows:

[0057] ​(221) Define the filter window: its size depends on the characteristics of the signal and the noise level;

[0058] (222) Calculate the Gaussian weight: calculate the weight of each point in the window, the closer the point to the current sample point, the greater the weight, the farther the point, the smaller the weight;

[0059] (223) Weighted average: weighted average of the value of each point in the window, to get the preliminary filtering result of the current sample point;

[0060] (224) Move the window: move the filter window forward by 1 step, repeat the above steps until all sample points are processed;

[0061] (225) Output results: output the smoothed and noise-reduced signal.

[0062] Step S23, the collected current data is single column time series data, which needs to be re-divided before using network training. Let the prediction step be m, and the prediction number be n (consistent with the filter window size in step S22), then the sliding window size is m+n, and the sliding step is set to 1, the specific steps are as follows: from the first data in the time series data, select m+n data as the first row of new data, then move the window by 1 step, select m+n data from the second data as the second row of new data, and so on, finally get new data m+n columns, the first m columns as the input of the network, the last n columns as the output of the network, so as to realize the prediction of n future data with m historical data;

[0063] In order to improve the performance of model training, the new data is normalized, and the calculation formula is as follows:

[0064]

[0065] x′ is the normalized input data;

[0066] x max is the maximum original data value;

[0067] x min is the minimum original data value;

[0068] By normalizing the input data, it can be ensured that the value range of the input original data is within [0, 1], and the influence of inconsistent data on the test is eliminated;

[0069] Step S3: building a training network model, predicting and training the collected current data, and evaluating the performance of the trained model; the network model is a TCN-GRU-ATTENTION combined model, mainly including a TCN module, a GRU module and an ATTENTION module; the process is as follows:

[0070] Step S31, building a training network model, specifically:

[0071] The time convolution network (TCN) is based on a one-dimensional convolutional neural network, which integrates multiple mechanisms such as dilated convolution, causal convolution and residual connection module. The core of TCN is dilated causal convolution and residual connection block. Causal convolution is a one-way structure, that is, the output of a certain time point of the previous level is only related to the input of the same time point and the previous level of the next level, and the calculation is as follows:

[0072]

[0073] In the formula: t is the time step; F is the convolution operation; F(t) is the output value of time point t; k is the convolution kernel size; f(i) is the i th element in the convolution kernel; d is the dilated coefficient; x t-d*i is the historical data in the input sequence;

[0074] The gated recurrent unit network (GRU) has two key gating mechanisms, reset gate and update gate, which integrates the functions of the forget gate and output gate in LSTM. The reset gate determines the amount of information lost at the previous moment, and the update gate controls the amount of state information transmitted from the previous moment to the current moment. Through these two gating mechanisms, important features can be retained and long and short-term dependencies in sequences can be effectively captured, as shown in the following formula:

[0075] z t =σ(W z h t-1 +U z x t +b z )

[0076] r t =σ(W r h t-1 +U r x t +b r )

[0077]

[0078] In the formula: z t is the output of the update gate; r t is the output of the reset gate; h tis the output of the hidden state; x t is the input at time t; h t-1 is the state at time t-1; σ is a sigmoid function; Tanh is a hyperbolic tangent function; W z , W r , W h is a weight vector of the recurrent connection; U z , U r , U h are weight vectors between the input layer, the update gate, the reset gate and the candidate state, respectively; b z , b r , b h are the corresponding bias matrices;

[0079] The self-attention mechanism can give different weights to different features in the training process, reduce the dependence on external information, and effectively improve the feature extraction capability of the model. The attention mechanism can not only capture the global feature information of the data, but also capture the feature information between the same group of data vectors, identify important trends in changes, and calculate as shown in the formula:

[0080]

[0081] In the formula: Attention is the attention weight obtained; Q is the length of the query matrix; K is the length of the key matrix; V is the length of the key vector; T is the transpose of the matrix; dk is the dimension of the key; Softmax is a normalization equation;

[0082] In step S32, a network model is built in MATLAB software, and corresponding training parameters are set, such as the maximum number of training times, the gradient threshold, the initial learning rate, and the regularization parameter, etc. The data preprocessed in step S2 is used for training, and the output value is restored by using the inverse normalization.

[0083] Further, the root mean square error (RMSE), the mean absolute error (MAE), and the determination coefficient (R 2 ) are used as evaluation indexes of the training effect of the model. The calculation of the effect evaluation indexes is shown in the formula:

[0084]

[0085] In the formula: Y i is the true value; is the predicted value; is the average value of all sample true values; m is the total number of test set samples. Among them, RMSE represents the deviation between the predicted value and the true value, MAE represents the error between the two, and R 2 represents the fitting degree index of the model, and the value range is 0-1. The closer to 1, the better the fitting effect of the model.

[0086] Step S4, as shown in the figure, the application proposes a current prediction-based inspection robot slip state determination method; Figure 4

[0087] The trained model is used to predict the trend of the future n current data, and to determine whether the inspection robot has slipped, the process is as follows:

[0088] Step S41, using the trained model, input m historical data, output n historical data, get motor current data prediction set {x1 x2 … xn}, further, the output torque value of the DC servo motor can be calculated by the current value, the calculation is as shown in the formula: n

[0089] T=I A K t i

[0090] In the formula: T is the motor output torque; I A is the driving motor current; K t is the motor torque constant; i is the size of the reducer reduction ratio;

[0091] Step S42, variance calculation is performed on the current prediction value obtained in step S41, and the calculation is as shown in the formula:

[0092]

[0093] In the formula: var is the variance size calculated by the data; n is the data number; x i is the data point; μ is the mean of the data point;

[0094] The reason for choosing variance as the judgment index: after trying various time series data feature indexes, it is found that the variance value of the current data when slipping occurs is rapidly rising compared with the smooth running, and there is an obvious dividing line between the two, so the corresponding threshold is designed, when the threshold is exceeded, it is considered that slipping has occurred (note to distinguish the sudden change caused by input change);

[0095] Step S43, according to the motor output torque prediction value calculated in step S41, combined with the single wheel dynamics model of the driving wheel (single wheel force analysis diagram as Figure 5 shown), the change of the longitudinal force of the driving wheel can be obtained, and the calculation is as shown in the formula:

[0096]

[0097] In the formula: F x ​​is the longitudinal force applied to the driving wheel by the ground; T is the motor output torque; r is the driving wheel diameter; J is the equivalent rotational inertia of the driving wheel; is the driving wheel angular acceleration;

[0098] Further, the driving wheel normal load size obtained in step S13 can be used to obtain the size of the friction coefficient μ s at this time, which is calculated as shown in the formula:

[0099]

[0100] Further, when the friction coefficient is lower than a certain value, the inspection robot will be in an uncontrollable state due to slipping, and at this time, applying control may cause the robot to be completely out of control, so a maximum threshold value Δx max of current change is set according to the friction coefficient calculation formula, and when this threshold value is exceeded, it can be considered that the robot is in an out-of-control state at this time, and should be immediately stopped and an alarm is reminded, waiting for the staff to handle;

[0101] Step S5: as shown in Figure 6 , the application proposes a control strategy adjustment method after slipping occurs;

[0102] According to whether the inspection robot slips or not, the corresponding control strategy is adjusted to reduce the impact of slipping, and the process is as follows:

[0103] Step S51, after the inspection robot slips, the controller first reduces the preset speed by a certain value, i.e. v c =kv, wherein v is the original running speed of the robot, v c is the adjusted input speed value, and the value of k can be given according to the current change rate in the current prediction value set in step S41, and it should be noted that the left and right driving wheels need to be adjusted synchronously;

[0104] Step S52, after the initial speed reduction is completed, the motor encoder wheel speed signal and the speed signal obtained by integrating the IMU acceleration are collected, and the difference between the two is compared, and when the difference between the two is too large, it is considered that the robot is still in a slipping state at this time, and the controller needs to continuously adjust the input speed value until the difference between the two is controlled within a certain range, i.e. v c =kv+Δv; At the same time, a model predictive controller is designed in the motion control aspect to make the inspection robot still run according to the predetermined trajectory when it deviates.

[0105] The content described in the embodiments of the present specification is only a list of implementation forms of the inventive concept, and is only for illustrative purposes. The protection scope of the present application should not be regarded as being limited to the specific forms described in the present embodiments, and the protection scope of the present application also extends to equivalent technical means that can be thought of by those skilled in the art according to the inventive concept.

Claims

1. A method for slippage identification and control of an inspection robot based on current prediction, characterized in that, The method includes the following steps: Step S1: Obtain the parameters of the inspection robot, including its center of mass position, mass, and running slope; calculate the magnitude of the normal force on the drive wheel; and calculate the vertical load on the drive wheel based on the robot's dynamics model. size; Step S2: Set up the experimental site to simulate the slippage situation that the inspection robot may encounter during the task, collect current data through the experiment and preprocess it. Step S3: Build a training network model, perform prediction training on the collected current data, and evaluate the performance of the trained model. Step S4: Use the trained model to predict the trend of the current data for the next n days and determine whether the inspection robot has slipped. In step S4, after obtaining the motor output torque, the angular acceleration information of the drive wheel is acquired through the motor encoder. The magnitude of the longitudinal force can be calculated based on the single-wheel dynamics model, and the magnitude of the friction coefficient is the longitudinal force of the drive wheel. and the vertical load in step S1 The ratio of friction coefficient is used to set a maximum current change threshold; a corresponding current change threshold is set to determine the slippage of the inspection robot at this time. Step S5: Adjust the control strategy based on whether the inspection robot slips or not to reduce the impact of slippage.

2. The method for slippage identification and control of an inspection robot based on current prediction as described in claim 1, characterized in that, In step S1, the position of the center of mass of the inspection robot includes the distance from the center of mass to the front axis, the distance from the center of mass to the rear axis, and the height of the center of mass, which are obtained by solving using SolidWorks software.

3. The slippage identification and control method for inspection robots based on current prediction as described in claim 2, characterized in that, In step S1, the mass of the inspection robot is obtained by weighing or solving with three-dimensional software, and the running slope is calculated based on the acceleration signal collected by the IMU sensor.

4. A method for slippage identification and control of an inspection robot based on current prediction as described in any one of claims 1 to 3, characterized in that, In step S2, the experimental site is selected from the common operating sites of the inspection robot, and corresponding slopes are set according to the robot's technical requirements. The actual low ground adhesion coefficient is simulated by adding water or laying mud and sand in different road sections. The data from the drive wheel motor is acquired by a servo driver, sent to the vehicle controller via CAN communication, and printed out. In step S2, the preprocessing process is as follows: (21) Based on the idea of ​​sliding window, the current data is filtered by Gaussian weighted average method; (22) The collected current data is a single-column time series data. It is re-divided into m+n columns of data according to the prediction step size m and the prediction step number n. (23) Normalize the data after partitioning to facilitate network training.

5. A method for slippage identification and control of an inspection robot based on current prediction as described in any one of claims 1 to 3, characterized in that, In step S3, the neural network structure used for training is a TCN-GRU-ATTENTION network. After training, it can predict the next n data points based on the first m data points. The performance of the trained model is evaluated by the root mean square error, mean absolute error, and coefficient of determination.

6. The slippage identification and control method for an inspection robot based on current prediction as described in claim 5, characterized in that, In step S4, the torque output by the motor is related to the current. The torque constant of the motor can be obtained by consulting the motor manual, and the reduction ratio can be obtained by consulting the nameplate of the reducer. The motor torque is the product of the current, the motor torque constant, and the reduction ratio.

7. A method for slippage identification and control of an inspection robot based on current prediction as described in any one of claims 1 to 3, characterized in that, In step S5, when slippage is detected, the magnitude of the motor output driving force is changed by adjusting the robot's running speed. The adjustment is based on the difference between the encoder wheel speed and the linear speed obtained by the IMU. When the difference is within a certain threshold, the slippage is considered to have improved.

Images