Adaptive optimization method and device based on repair effect in road repair process

Abstract

The application discloses a kind of self-adapting optimization method and equipment based on repair effect in road repair process, the method comprises the following steps: S1, based on the specified parameter in the robot operation process, the whole-process online perception and dynamic monitoring mechanism covering the specified index is constructed;S2, based on the real-time synchronous feedback based on twinborn environment and robot ontology state of preset digital twinborn platform, and construct the dynamic cooperative control strategy of repair robot chassis navigation and spraying action;S3, construct the closed-loop optimization system based on real-time evaluation of repair effect, propose the adaptive adjustment algorithm of repair robot repair parameter in the system, for realizing the adaptive adjustment of specified repair parameter in the process that repair robot executes repair operation.The application can realize the closed-loop regulation ability that spraying process is visible, path execution is adjustable, repair quality is controllable, to effectively break through the industry bottleneck that operation parameter is static, regulation lag, quality is unpredictable.

Description

Technical Field

[0001] This invention relates to the field of road repair data processing technology, specifically to an adaptive optimization method based on repair results during road repair and a road repair robot device employing this method. Background Technology

[0002] With the rapid growth of urbanization and transportation demands, road infrastructure has been operating under high loads for extended periods. Especially under the combined effects of heavy traffic, complex geological conditions, extreme weather, and environmental aging, asphalt pavement suffers from frequent cracks, fissures, and potholes, severely impacting road structural safety and service performance. In recent years, the frequency of serious traffic accidents caused by road collapses or worsening road conditions has been steadily increasing, becoming a significant threat to urban public safety. Improving the early identification and rapid repair capabilities of road defects has become an urgent need to ensure the resilience of transportation systems and extend the lifespan of infrastructure.

[0003] However, current road defect detection and repair still heavily rely on manual methods, resulting in key problems such as delayed detection, low efficiency, rudimentary repair practices, and uncontrollable quality. On the one hand, the accuracy of manual detection is limited, making it difficult to meet the needs of comprehensive, real-time, and high-precision defect identification in high-density road networks. On the other hand, crack repair processes are fragmented, with disjointed work path planning, rigid control strategies, and reliance on post-repair manual evaluation for repair quality. Overall, there is a lack of systematic and intelligent support mechanisms, which restricts operational efficiency and consistency.

[0004] Currently, crack repair robots suffer from problems such as static setting of operating parameters, rigidity of path execution, imperceptibility of process status, and delayed effect feedback. These issues result in inconsistent quality, such as uneven repair coverage and fluctuations in coating thickness, and heavily rely on manual inspection and repetitive work.

[0005] Therefore, this application proposes an adaptive optimization method and equipment based on repair effect in the road repair process to solve the technical problems of insufficient rigidity of traditional repair robot operation control, low chassis-robotic arm coordination capability and uncontrollable operation quality. Summary of the Invention

[0006] The main objective of this invention is to provide an adaptive optimization method and device based on repair effect during road repair. Based on twin environment perception and dynamic decision support, it studies the collaborative control technology of robot chassis and robotic arm, the adaptive adjustment mechanism of operation parameters and the closed-loop feedback optimization system of operation quality, so as to improve the autonomous optimization capability and repair consistency of the operation process and solve the technical problems mentioned in the background art.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solutions:

[0008] An adaptive optimization method and device based on repair effects in road repair processes, which includes the following steps executed by computer equipment:

[0009] S1. Based on specified key parameters including posture, position, speed, spraying pressure, and material flow during robot operation, a full-process online perception and dynamic monitoring mechanism covering specified key indicators including posture, position, speed, and spraying parameters is constructed to achieve real-time capture, risk warning, and operation assurance of the repair robot's repair operation status, and solve the problems of existing repair robots having a single dimension of status perception, slow response, and lack of continuous monitoring and anomaly identification capabilities during operation.

[0010] S2. Based on the real-time status feedback mechanism of the preset digital twin platform, the status of the twin environment and the robot body is synchronously fed back in real time. A dynamic collaborative control strategy for the chassis navigation and spraying action of the repair robot is constructed to optimize the movement speed, trajectory following and action execution accuracy of the repair robot during the repair process. This achieves a dual improvement in the accuracy of operation path following and the robustness of action execution in complex road scenarios. This solves the key problems of chassis and robotic arm action disconnection, control response lag and difficulty in autonomous adjustment according to environmental changes in the current repair robot operation process, and realizes autonomous decision-making and dynamic adjustment in complex environments.

[0011] S3. Construct a closed-loop optimization system based on real-time evaluation of repair effects, develop a closed-loop optimization mechanism based on real-time evaluation of work effects, and propose an adaptive adjustment algorithm for repair parameters, including the repair robot's trajectory step length, moving speed, and spraying density, in the system. This algorithm is used to achieve adaptive adjustment of specified key repair parameters (such as spraying density, trajectory step length, and moving speed) during the repair robot's operation, thereby improving the intelligence of the work process and the consistency of repair quality. This addresses the current problems of static parameter settings, lack of spraying effect feedback, and reliance on manual quality adjustment in the current repair robot operation process, ultimately achieving continuous improvement and intelligent optimization of work quality.

[0012] Preferably, in step S1, a multi-source sensing system consisting of an inertial measurement unit (IMU), an encoder, a GNSS module, a flow sensor, and a spraying pressure sensor is constructed to achieve synchronous acquisition of core operational parameters such as attitude, displacement, velocity, material flow rate, and spraying pressure.

[0013] For data processing, a moving average algorithm is used to fuse information from multiple sources to improve the stability and accuracy of attitude and velocity estimation. At this time, for abnormal situations such as spraying fluctuations and trajectory deviations, an anomaly detection algorithm based on time series is proposed to identify and report key anomalies such as spraying overpressure, flow fluctuations, and trajectory drift in real time during the operation.

[0014] Based on the processed data, a status assessment and risk level classification module is then constructed. The dynamic visualization of the status flow is then integrated into a preset digital twin platform and ROS visualization terminal to realize the display of operational health status, anomaly highlighting, risk warning and strategy intervention interface linkage.

[0015] Preferably, the real-time state feedback mechanism in step S2 is established collaboratively through state synchronization evaluation, collaborative decision modeling, and control command optimization, and closed-loop control is driven by twin environment feedback. The collaborative establishment process of the real-time state feedback mechanism is as follows:

[0016] S21. Construct a state perception and evaluation model. Based on specified parameters including robot position, attitude, trajectory deviation, and speed disturbance synchronized in real time in the preset digital twin platform system, use residual analysis and error weight model to evaluate the stability of the current repair operation state.

[0017] S22. Set up a chassis-robotic arm collaborative control strategy framework. Improve trajectory accuracy by using feedforward tracking and gain scheduling PID controller in the smooth section of the path. In the steering / sudden change area, introduce Q-learning adaptive control algorithm to adjust chassis speed and spraying posture to achieve spatiotemporal coordination and matching of speed and action.

[0018] S23. Construct a set of state-driven decision logic modules to trigger the execution of repair posture adjustment, spraying stop and recalibration instructions in advance based on the twin state prediction results in the digital twin platform system when trajectory deviation or environmental change is detected, so as to ensure action continuity and repair consistency.

[0019] At this time, control commands are sent to the robot controller in real time through the ROS 2 interface and can be visualized and verified in a preset digital twin platform, including the Unity3D twin platform.

[0020] Preferably, the specific operational procedure for evaluating the stability of the current repair operation status using residual analysis and error weighting models in step S21 includes:

[0021] S211. Based on the current repair operation status and the predicted repair operation status, construct a linear regression model and calculate the corresponding residuals. The smaller the residuals, the higher the stability.

[0022] S212. If the residual value is greater than the preset threshold, then construct observation weights for each group of error terms and minimize the weighted sum of squared residuals, where the weights are obtained from residual analysis.

[0023] Preferably, the weight acquisition process includes:

[0024] Fit an initial model using ordinary least squares;

[0025] Plot the residuals against the fitted values ​​to perform residual analysis and observe the pattern of heteroscedasticity;

[0026] Observe residuals absolute value or square with fitted value Or specify the argument. Establish a variance function to establish the relationship between them:

[0027] if and If they are directly proportional, then the variance and Proportional, at this point the weights are set to ;

[0028] if If it grows linearly, then the variance and Proportional, at this point the weights are set to .

[0029] Preferably, the variance function is replaced and used during the establishment process. Regression functions are performed on possible variables to establish an estimated variance function.

[0030] Preferably, in step S22, the Q-learning adaptive control algorithm initializes state S and performs the following specific cyclic update operation on state S:

[0031] S221. Select an action based on the current Q-value function and the specified exploration strategy. Among them, the exploration strategy is The probability of randomly selecting an exploration action, in order to The probability of choosing the current optimal action ;

[0032] S222. Perform the action And observe rewards and the next state At this point, the Q-value function represents the action performed in state S. And thereafter they all followed the execution strategy. The expected cumulative discount reward that can be obtained at that time;

[0033] S223. Update the Q value, with the following update formula: ;

[0034] in, It's the learning rate. It is the target Q value, which is composed of the actual reward obtained and the best estimate of the next state. It is the time-series difference error, used to represent the difference between the current estimate and the target. Represents the Q-value function, This represents the new action performed during the process of updating the Q-value by executing the optimal strategy;

[0035] S224. Change the state from Updated to .

[0036] Preferably, in process S223, a function approximator is used to approximate the Q-function for the Q-value function. In this case, the adaptive controller structure during the parameter update process is as follows:

[0037] Input layer: has state ;

[0038] Hidden layers: set as multi-layer neural networks;

[0039] Output layer: Outputs the Q-values ​​for all discrete actions, where the optimal action is set as... , For function approximation, It is the parameter vector of the approximator, which is updated by gradient descent to minimize the square of the time difference error.

[0040] Preferably, the specific operation process of using the adaptive adjustment algorithm to adaptively adjust the specified repair parameters in step S3 includes:

[0041] S31. Establish an online monitoring system that integrates material flow meters, visual feedback systems, and infrared thickness sensors to collect data on specified core indicators, including coating coverage, layer thickness uniformity, and path overlap accuracy, in real time.

[0042] S32. Based on the collected data, a repair effect evaluation model is constructed. The U-Net image segmentation algorithm is used to extract the contour of the repair area in the model. The path consistency index is calculated by combining the robot's movement trajectory execution record. The comprehensive quality score is achieved through a multi-factor weighted model.

[0043] S33. The scoring results are used as feedback input to the parameter scheduling module. By combining the genetic algorithm, the robot spraying parameters are dynamically searched and updated to form a closed-loop control mechanism of "parameter setting - effect feedback - adaptive adjustment".

[0044] The adaptive adjustment algorithm in the closed-loop optimization system will run in the ROS node and will be displayed through a preset digital twin platform to show the comparison of the state before and after the repair and the evolution of the adjustment trajectory, supporting the continuous optimization and stable operation of the intelligent control system.

[0045] Preferably, the U-Net structure in the U-Net image segmentation algorithm of step S32 includes:

[0046] The encoder is used to progressively extract features and capture the contextual information of the image. The encoder consists of 4 blocks, each block containing two sets of 3x3 convolutional blocks and ReLU activation functions, and a set of 2x2 max pooling blocks. The max pooling blocks are set with a stride of 2 and are used to halve the feature map size while doubling the number of channels.

[0047] The decoder is used to upsample the low-resolution, high-semantic features learned by the encoder back to the original image size in order to accurately locate the position of the target object and restore its details. The decoder consists of 4 blocks, each of which is skip-connected to the feature map of the corresponding layer of the encoder. Each block contains two sets of 3x3 convolutional blocks and ReLU activation functions, as well as a set of 2x2 upsampled convolutional blocks. The upsampled convolutional blocks are used to double the size of the feature map while halving the number of channels.

[0048] On the other hand, the present invention also discloses an adaptive optimization device based on repair effect during road repair, which is equipped with a computer data processing module for executing any of the above-described adaptive optimization methods based on repair effect during road repair after receiving the collected data from the device.

[0049] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0050] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0051] As can be seen from the above technical solution, the present invention provides an adaptive optimization method and device based on repair effect in the road repair process. Compared with the prior art, the present invention has the following advantages:

[0052] 1. This invention establishes a closed-loop optimization mechanism for the repair process based on twin feedback, integrating real-time monitoring of multiple parameters, online evaluation of repair effects, and adaptive parameter control technology. This enables closed-loop adjustment capabilities that allow for visibility of the spraying process, adjustable path execution, and controllable repair quality, thereby effectively overcoming industry bottlenecks such as static operation parameters, lagging control, and unpredictable quality.

[0053] 2. This invention constructs a multi-source sensing system consisting of an IMU, encoder, GNSS, flow and pressure sensors, and combines moving average fusion and time series anomaly detection algorithms to achieve continuous, high-precision, and real-time monitoring of key states such as robot posture, position, speed, and spraying parameters. This solves the problems of undetectable states and difficulty in early warning of anomalies in traditional repair robots, and provides a reliable data foundation for dynamic decision-making and closed-loop control.

[0054] 3. Based on a digital twin platform, this invention sets up a dynamic collaborative control strategy for chassis navigation and robotic arm painting actions, and integrates methods such as feedforward PID and Q-learning adaptive control to achieve a dual improvement in path tracking accuracy and action robustness. It can maintain high-precision trajectory following and action coordination even in scenarios with complex terrain and dynamic obstacles, thereby solving the technical pain points of disconnect between chassis and robotic arm control and lag response in traditional systems.

[0055] 4. This invention constructs a closed-loop optimization mechanism of parameter setting, effect feedback, and adaptive adjustment, integrating vision, infrared sensors, U-Net image segmentation, and genetic algorithms. It can achieve dynamic optimization of key parameters such as spray density, trajectory step length, and moving speed, thereby significantly improving the integrity of repair coverage, material utilization, and work consistency, and realizing an intelligent repair process that is perceptible in quality, adjustable in parameters, and evolvable in process.

[0056] 5. By setting up a Unity3D twin platform and ROS 2 interface, this invention can achieve real-time synchronization and visual verification of robot status, control commands, and repair effects, and supports strategy pre-playing, anomaly intervention, and system optimization, thereby enhancing the system's transparency, controllability, and debugging efficiency.

[0057] It should be understood that the descriptions in this section are not intended to identify key or essential features of embodiments of the invention, nor are they intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Of course, implementing any product of the invention does not necessarily require achieving all of the advantages described above simultaneously. Attached Figure Description

[0058] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0059] Figure 1 This is a schematic diagram of the overall operation process of the present invention. Detailed Implementation

[0060] 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 a part of the embodiments of the present invention, and not all of them. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] For details in the embodiments, please refer to Figure 1 .

[0062] like Figure 1 As shown in the embodiments of the present invention, the adaptive optimization method based on repair effect in the road repair process includes the following steps:

[0063] S1. Based on specified key parameters including posture, position, speed, spraying pressure, and material flow rate during robot operation, a full-process online perception and dynamic monitoring mechanism covering specified key indicators including posture, position, speed, and spraying parameters is constructed to achieve real-time capture, risk warning, and operational assurance of the repair robot's repair operation status, thereby solving the problems of existing repair robots having a single dimension of status perception, delayed response, and lack of continuous monitoring and anomaly identification capabilities during operation.

[0064] Specifically, in one set of embodiments, a multi-source sensing system consisting of an inertial measurement unit (IMU), an encoder, a GNSS module, a flow sensor, and a spraying pressure sensor is constructed to achieve synchronous acquisition of core operational parameters such as attitude, displacement, velocity, material flow rate, and spraying pressure.

[0065] For data processing, a moving average algorithm is used to fuse information from multiple sources to improve the stability and accuracy of attitude and velocity estimation. At this time, for abnormal situations such as spraying fluctuations and trajectory deviations, an anomaly detection algorithm based on time series is proposed to identify and report key anomalies such as spraying overpressure, flow fluctuations, and trajectory drift in real time during the operation.

[0066] Based on the processed data, a status assessment and risk level classification module is then constructed. The dynamic visualization of the status flow is then integrated into a preset digital twin platform and ROS visualization terminal to realize the display of operational health status, anomaly highlighting, risk warning and strategy intervention interface linkage.

[0067] In summary, this method can provide a complete state perception technology system for repair robot systems, from sensor acquisition to multi-source fusion, and from state estimation to anomaly diagnosis. It can significantly improve the observability, safety and decision support capabilities of the repair process, and lay a real-time data foundation for building intelligent control and closed-loop feedback mechanisms.

[0068] S2. Based on the real-time status feedback mechanism of the preset digital twin platform, the status of the twin environment and the robot body is synchronously fed back in real time. A dynamic collaborative control strategy for the chassis navigation and spraying action of the repair robot is constructed to optimize the moving speed, trajectory following and action execution accuracy of the repair robot in the repair process. This achieves a dual improvement in the accuracy of operation path following and the robustness of action execution in complex road scenarios. This solves the key problems of chassis and robotic arm action disconnection, control response lag and difficulty in autonomous adjustment according to environmental changes in the current repair robot operation process, and realizes autonomous decision-making and dynamic adjustment in complex environments.

[0069] Specifically, in one set of embodiments, the real-time state feedback mechanism is established collaboratively through state synchronization evaluation, collaborative decision modeling, and control command optimization, and closed-loop control is driven by twin environment feedback. The collaborative establishment process of the real-time state feedback mechanism is as follows:

[0070] S21. Construct a state perception and evaluation model. Based on specified parameters including robot position, attitude, trajectory deviation, and speed disturbance synchronized in real time in the preset digital twin platform system, use residual analysis and error weight model to evaluate the stability of the current repair operation state.

[0071] The specific operational procedures for evaluating the stability of the current repair operation status using residual analysis and error weighting models include the following.

[0072] S211. Based on the current repair operation status and the predicted repair operation status, construct a linear regression model and calculate the corresponding residuals. The smaller the residuals, the higher the stability.

[0073] This can be further explained by the fact that in regression models In the middle, residual It is the difference between the observed values ​​and the model predictions, i.e. ,in It is the first The actual value of each observation. The model is for the first The predicted value of each observation is used, and the residual can be regarded as the error term. The estimated value.

[0074] The residual analysis used here is mainly to verify the independence (error terms are independent of each other) and homoscedasticity (error terms have constant variance) in the linear regression model. The residuals are checked for normality (the error term follows a normal distribution), and linearity is also verified. Specific verification is achieved by comparing the residuals with various residual plots. When residual analysis reveals problems, especially heteroscedasticity, an error weighting model (usually implemented using weighted least squares) is employed.

[0075] S212. If the residual value is greater than the preset threshold, then construct observation weights for each group of error terms and minimize the weighted sum of squared residuals, where the weights are obtained from residual analysis.

[0076] Furthermore, the weight acquisition process includes:

[0077] Fit an initial model using ordinary least squares;

[0078] Plot the residuals against the fitted values ​​to perform residual analysis and observe the pattern of heteroscedasticity;

[0079] Observe the absolute value of the residuals or square with fitted value Or specify the argument. Establish a variance function to establish the relationship between them:

[0080] if and If they are directly proportional, then the variance and Proportional, at this point the weights are set to ;

[0081] if If it grows linearly, then the variance and Proportional, at this point the weights are set to .

[0082] Preferably, the variance function is replaced and used during the establishment process. Regression functions are performed on possible variables to establish an estimated variance function.

[0083] Because of the heteroscedasticity problem in the least squares method, the result of minimizing the sum of squared residuals can easily become invalid. Therefore, after the weighted least squares model is refitted with the estimated weights and residual analysis is performed again, it is necessary to re-examine the heteroscedasticity problem in the residual plot of the new model until the heteroscedasticity problem is eliminated or reduced. Otherwise, the weight function needs to be adjusted and this process needs to be repeated.

[0084] Furthermore, the residual analysis and error weighting model here represent the relationship between the specified independent variable "diagnosis - treatment", and the process is as follows:

[0085] Step 1: Establish the initial model: Construct a basic regression model using methods such as weighted least squares.

[0086] Step 2: Residual Analysis (Diagnosis): Using residual plots, normality tests, etc., determine whether there are problems such as heteroscedasticity or outliers.

[0087] Step 3: Error Weighting Model (Solution): If heteroscedasticity exists, set reasonable weights based on the residual analysis results and reconstruct the model to improve estimation accuracy.

[0088] Step 4: Verify the optimization effect: Analyze the residuals of the weight model again to confirm whether the heteroscedasticity problem has been solved.

[0089] S22. Set up a chassis-robotic arm collaborative control strategy framework. Improve trajectory accuracy by using feedforward tracking and gain scheduling PID controller in the smooth section of the path. In the steering / sudden change region, introduce Q-learning adaptive control algorithm to adjust chassis speed and spraying posture to achieve spatiotemporal coordination and matching of speed and action.

[0090] The Q-learning adaptive control algorithm here is essentially a set of Markov decision processes (MDPs), including:

[0091] 1) Status : The observed values ​​of the controlled system at time t. For example, the position and velocity of the robot, the angle and angular velocity of the motor, etc.

[0092] 2) Actions The control signal applied to the system by the controller at time t. For example, the applied voltage, torque, etc.

[0093] 3) Rewards : In performing the action Then, the system changes state. Transition to state The instantaneous performance score is obtained. The design of the reward function is crucial to the algorithm's success. For example:

[0094] • Target tracking problem: = (Negative squared error)

[0095] • The problem of energy conservation: =- (energy consumed)

[0096] 4) Q-function Q(s,a): This is the core of the algorithm. It represents the action a performed in state s, and thereafter follows a specific policy. The expected cumulative discount reward that can be obtained at that time.

[0097] (

[0098]

[0099] 5) Optimal Strategy A rule that maps states to actions that maximizes cumulative reward. Once the optimal Q-function is learned... The optimal strategy is quite simple: in state Next, choose the option that enables The biggest movement: =argma .

[0100] Furthermore, the Q-learning adaptive control algorithm initializes state S and performs the specific iterative update operation of state S as follows:

[0101] S221. Select an action based on the current Q-value function and the specified exploration strategy. Among them, the exploration strategy is The probability of randomly selecting an exploration action, in order to The probability of choosing the current optimal action ;

[0102] S222. Perform the action And observe rewards and the next state At this point, the Q-value function represents the action performed in state S. And thereafter they all followed the execution strategy. The expected cumulative discount reward that can be obtained at that time;

[0103] S223. Update the Q value, with the following update formula: ;

[0104] in, It's the learning rate. It is the target Q value, which is composed of the actual reward obtained and the best estimate of the next state. It is the time-series difference error, used to represent the difference between the current estimate and the target. Represents the Q-value function, This represents the new action performed during the process of updating the Q-value by executing the optimal strategy;

[0105] At this point, for the Q-value function, a function approximator is used to approximate the Q-function. The adaptive controller structure during the parameter update process is as follows:

[0106] Input layer: has state ;

[0107] Hidden layers: set as multi-layer neural networks;

[0108] Output layer: Outputs the Q-values ​​for all discrete actions, where the optimal action is set as... , For function approximation, It is the parameter vector of the approximator, which is updated by gradient descent to minimize the square of the time difference error;

[0109] For continuous action spaces, other algorithms are needed, such as:

[0110] Deep Deterministic Policy Gradient (DDPG);

[0111] Soft Actor-Critic (SAC);

[0112] Furthermore, it can be added that the network parameters here... Updating using gradient descent to minimize the squared time-series difference error, we have:

[0113] ;

[0114] S224. Change the state from Updated to .

[0115] In summary, the Q-learning adaptive control algorithm used here does not require a precise mathematical model of the system, can avoid complex system identification and modeling error problems, and is committed to learning the strategy that maximizes long-term cumulative reward. It is usually globally optimal or near-globally optimal, which is a significant improvement over traditional adaptive control (which is often only optimal near a local operating point). It also has a strong adaptability to internal parameter changes, unmodeled dynamics and external disturbances, and is naturally suitable for complex nonlinear system control.

[0116] S23. Construct a set of state-driven decision logic modules to trigger the execution of repair posture adjustment, spraying stop and recalibration commands in advance based on the twin state prediction results in the digital twin platform system when trajectory deviation or environmental change is detected, so as to ensure the continuity of actions and repair consistency.

[0117] At this time, control commands are sent to the robot controller in real time through the ROS 2 interface and can be visualized and verified in a preset digital twin platform, including the Unity3D twin platform.

[0118] In summary, this method will significantly improve the motion coordination and operational stability of repair robots in complex environments such as heterogeneous terrain, dynamic obstacles, and path disturbances. It can realize a fully intelligent operation control system with "path tracking, attitude controllability, and controllability", providing stable execution support for multi-target precision repair tasks.

[0119] S3. Construct a closed-loop optimization system based on real-time evaluation of repair effects, develop a closed-loop optimization mechanism based on real-time evaluation of work effects, and propose an adaptive adjustment algorithm for repair parameters, including the repair robot's trajectory step length, moving speed, and spraying density, in the system. This algorithm is used to achieve adaptive adjustment of specified key repair parameters (such as spraying density, trajectory step length, and moving speed) during the repair robot's operation, thereby improving the intelligence of the work process and the consistency of repair quality. This addresses the current problems of static parameter settings, lack of spraying effect feedback, and reliance on manual quality adjustment in the current repair robot operation process, ultimately achieving continuous improvement and intelligent optimization of work quality.

[0120] Specifically, in one set of embodiments, the specific operation process of adaptively adjusting specified repair parameters using an adaptive adjustment algorithm includes:

[0121] S31. Establish an online monitoring system that integrates material flow meters, visual feedback systems, and infrared thickness sensors to collect data on specified core indicators, including coating coverage, layer thickness uniformity, and path overlap accuracy, in real time.

[0122] S32. Based on the collected data, a repair effect evaluation model is constructed. The U-Net image segmentation algorithm is used to extract the contour of the repair area in the model. The path consistency index is calculated by combining the robot's movement trajectory execution record. The comprehensive quality score is achieved through a multi-factor weighted model.

[0123] At this point, the U-Net structure in the U-Net image segmentation algorithm includes an encoder (shrinking path), a decoder (expanding path), and skip connections. Specifically:

[0124] (1) The encoder, like traditional CNN classification networks (such as VGG), is used to progressively extract features and capture the contextual information of the image. As the network deepens, the spatial size (height and width) of the feature map becomes smaller and smaller, but the number of channels increases and the receptive field becomes larger and larger;

[0125] Structurally, the encoder consists of four blocks, each containing:

[0126] 1) Two 3x3 convolutional blocks (without padding, so the feature map size will be slightly reduced) with the ReLU activation function;

[0127] 2) A 2x2 max pooling block (with a stride of 2) is used to halve the feature map size while doubling the number of channels.

[0128] Ultimately, after multiple downsampling operations, the feature map is compressed into a low-resolution, high-dimensional feature representation that contains rich semantic information.

[0129] (2) Decoder, used to upsample (or “deconvolve”) the low-resolution, high-semantic features learned by the encoder back to the original image size in order to accurately locate the position of the target object and restore its details.

[0130] Structurally, the decoder consists of four blocks, each of which makes skip connections to the feature maps of the corresponding layer of the encoder, and each block contains:

[0131] 1) A 2x2 upsampled convolutional block (transposed convolution) doubles the feature map size while halving the number of channels;

[0132] 2) Two 3x3 convolutional blocks with ReLU activation function are used to fuse and refine features;

[0133] Finally, a segmentation map with a size similar to the input image is output.

[0134] S33. The scoring results are used as feedback input to the parameter scheduling module. By combining the genetic algorithm, the robot spraying parameters are dynamically searched and updated to form a closed-loop control mechanism of "parameter setting - effect feedback - adaptive adjustment".

[0135] (3) Skip connections are used to concatenate high-resolution, low-semantic feature maps at the same level in the encoder with low-resolution, high-semantic feature maps in the decoder along the channel dimension. This alleviates the gradient vanishing problem in deep networks, makes the network easier to train, and combines local information (from the encoder) and global information (from the decoder), thus achieving multi-scale feature fusion.

[0136] In a set of specific embodiments, the U-Net image segmentation algorithm can use an overlapping sliding window strategy to segment large images (such as electron microscope images), which can predict the segmentation results of image patches, use overlapping regions to eliminate prediction artifacts at the boundaries, and ensure the prediction accuracy of edge pixels by mirroring the context.

[0137] In one specific embodiment, the U-Net image segmentation algorithm, when segmenting in-touch objects in biomedical images, typically introduces a weight map into the standard cross-entropy loss function. This weight map assigns higher weights to boundary pixels between adjacent cells, forcing the network to pay more attention to these difficult-to-segment regions.

[0138] In one specific embodiment, the U-Net image segmentation algorithm can also use data augmentation techniques such as elastic deformation to improve the robustness and generalization ability of the model in medical image data.

[0139] The final output layer of the network is a 1x1 convolution that maps the 64-channel feature map to output channels for K classes. For example, for a binary classification problem (background and foreground), K=2.

[0140] In summary, the U-Net image segmentation algorithm benefits from skip connections, enabling it to generate segmentation results with very accurate boundaries, facilitating high-precision segmentation. Secondly, the algorithm structure is efficient, achieving good results even on relatively small datasets, making it suitable for small datasets. Overall, the input is an image, and the output is a segmentation map, making the entire process simple and efficient, facilitating end-to-end training. In terms of applicable fields, it can be used in most conventional technical fields.

[0141] The adaptive adjustment algorithm in the closed-loop optimization system will run in the ROS node and will be displayed through a preset digital twin platform to show the comparison of the state before and after the repair and the evolution of the adjustment trajectory, supporting the continuous optimization and stable operation of the intelligent control system.

[0142] In summary, this application's method constructs an intelligent closed-loop system with "perceptible quality, adjustable parameters, and evolvable process," proposing a quality closed-loop optimization mechanism for the repair process based on twin feedback. It integrates real-time multi-parameter monitoring, online evaluation of repair effects, and adaptive parameter control technology to achieve closed-loop adjustment capabilities of "visible spraying process, adjustable path execution, and controllable repair quality." This differs from the current traditional model with rigid control and reliance on manual quality assessment. It effectively reduces the need for manual intervention, improves the robot's coverage integrity, material utilization efficiency, and operational consistency in various crack repair tasks, and effectively overcomes industry bottlenecks such as static operational parameters, lagging control, and unpredictable quality. It provides key control support for building a highly consistent repair system that is "standardized, automated, and intelligent," and supports the full-process quality self-optimization capability under a digital twin platform.

[0143] On the other hand, the present invention also discloses an adaptive optimization device based on repair effect during road repair, which is equipped with a computer data processing module for receiving the collected data from the device and executing the adaptive optimization method based on repair effect during road repair in the above embodiments.

[0144] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0145] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0146] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute an adaptive optimization method based on the repair effect during any of the road repair processes described in the above embodiments.

[0147] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0148] This application also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other via the communication bus.

[0149] Memory, used to store computer programs;

[0150] When the processor executes the program stored in memory, it implements the adaptive optimization method based on the repair effect in the road repair process described above.

[0151] The communication bus mentioned in the above-mentioned electronic devices can be a standard bus for interconnecting peripheral components or an extended industrial standard structure bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0152] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0153] The memory may include random access memory or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0154] The processors mentioned above can be general-purpose processors, including central processing units, network processors, etc.; they can also be digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0155] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0156] 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.

[0157] Furthermore, it should be noted that if any directional indication (such as up, down, left, right, front, back, etc.) is involved in the embodiments of the present invention, the directional indication is only used to explain the relative positional relationship and movement of each component in a specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0158] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the meaning of "and / or" throughout the text includes three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, in the embodiments of this invention, "multiple" refers to two or more. Moreover, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

Claims

1. A method for adaptive optimization based on repair effectiveness in a road repair process, characterized in that, Comprise: S1. Based on the specified parameters in the robot operation process, a full-process online perception and dynamic monitoring mechanism covering the specified indicators is constructed to realize real-time capture, risk warning and operation guarantee of the repair robot repair operation state; S2. Based on the state real-time feedback mechanism of the preset digital twin platform, the real-time synchronous feedback based on the twin environment and the robot body state is realized, and a dynamic cooperative control strategy for the navigation of the repair robot chassis and the spraying action is constructed, which is used to optimize the moving speed, trajectory following and action execution accuracy of the repair robot in the repair process; S3. A closed-loop optimization system based on real-time evaluation of repair effect is constructed, and an adaptive adjustment algorithm for repair robot repair parameters is proposed in the system to realize adaptive adjustment of specified repair parameters during the execution of repair operation by the repair robot; The cooperative establishment operation process of the state real-time feedback mechanism in the S2 step is: S21. Based on the real-time synchronous robot specified parameters in the preset digital twin platform system, the stability of the current repair operation state is evaluated using residual analysis and error weight model; S22. A set of chassis-robot cooperative control strategy framework is set, the trajectory accuracy is improved by using feedforward tracking and gain scheduling PID controller in the path smooth section, and Q-learning type adaptive control algorithm is introduced in the steering / rapid change area to adjust the chassis speed and spraying posture, so as to realize the space-time coordination matching of speed and action; S23. A set of state-driven decision logic modules are constructed, which are used to trigger the execution of repair posture adjustment, spraying suspension and re-calibration instructions in advance based on the twin state prediction results in the digital twin platform system when trajectory deviation or environmental mutation is detected, so as to ensure the action continuity and repair consistency; At this time, the control command is issued to the robot controller in real time through the ROS 2 interface, and can be visualized and verified in the preset digital twin platform; The specific operation process of using residual analysis and error weight model to evaluate the stability of the current repair operation state in the S21 step includes: S211. Based on the current repair operation state and the predicted repair operation state, a linear regression model is constructed and the corresponding residual is calculated, and the smaller the residual is, the higher the stability is; S212. If the residual value is greater than the preset threshold, an observation weight is constructed for each error term, and the weighted residual sum of squares is minimized, wherein the weight is obtained according to residual analysis; The operation process of obtaining the weight includes: Fitting an initial model with ordinary least squares method; Draw residual and fitted value diagram for residual analysis, observe the pattern of heteroscedasticity; Observe residuals absolute value or square with fitted value Or specify the argument. Establish a variance function to establish the relationship between them: If is proportional to , then the variance is proportional to , in which case the weights are set to ; If variance is proportional to the weight is set to .

2. The method of claim 1, wherein, The variance function is built using the replacement A regression function is performed on the possible variables to build an estimated variance function.

3. The method of claim 1, wherein, The Q-learning type adaptive control algorithm in the S22 step initializes the state S, and performs specific loop update operation of state S as follows: S221. Select an action according to the current Q-value function and a specified exploration policy where the exploration policy randomly selects an exploration action with a probability of and selects the current optimal action with a probability of ;​ S222. Perform action and observe reward and next state At this point the Q-value function represents the expected cumulative discounted reward that can be obtained by performing action and then following the policy at state S. S223. Update Q value, with update formula as: ; wherein, is a learning rate, is a target Q value, constructed based on an actually obtained reward and a best estimate of a next state, is a temporal difference error, used to represent a difference between a current estimate and a target, represents a Q value function, represents a new action operated in an optimal policy during updating of a Q value; S224. update the state from to .

4. The method of claim 3, wherein the method further comprises: In the S223 process, the Q value function is approximated using a function approximator, and at this time the adaptive controller structure in the parameter update process is: Input layer: has state ; Hidden layer: set to multi-layer neural network; Output layer: output Q-values for all discrete actions, with the optimal action set as , is a function approximator, is the parameter vector of the approximator, and is updated by gradient descent to minimize the squared temporal difference error.

5. The method of claim 1, wherein, The specific operation process of using adaptive adjustment algorithm to adaptively adjust the specified repair parameters in the S3 step includes: S31. Real-time acquisition of specified indicator data; S32. Construct a repair effect evaluation model based on the collected data, extract the repair area contour in the model using the U-Net image segmentation algorithm, calculate the path consistency index combined with the robot movement trajectory, and realize the comprehensive quality score through the multi-factor weighted model; S33. The scoring results are used as feedback to dynamically search and update the robot spraying parameters by combining genetic algorithms; Wherein the adaptive adjustment algorithm will run in the ROS node, and the state comparison before and after repair and the adjustment trajectory evolution process will be displayed through the preset digital twin platform.

6. The method of claim 5, wherein the method further comprises: The U-Net structure in the U-Net image segmentation algorithm of the S32 step includes: An encoder for step-by-step feature extraction and capturing context information of the image, the encoder is composed of 4 blocks, each block contains two groups of 3x3 convolution blocks and ReLU activation functions, and a group of 2x2 maximum pooling blocks, the maximum pooling block is set with a step of 2, and is used to reduce the feature map size by half while doubling the channel number; A decoder for up-sampling the low-resolution, high-semantic features learned by the encoder back to the original image size to accurately locate the position of the target object and restore its details, the decoder is composed of 4 blocks, each block is connected with the feature map of the corresponding layer of the encoder, and each block contains two groups of 3x3 convolution blocks and ReLU activation functions and a group of 2x2 up-sampling convolution blocks, the up-sampling convolution block is used to double the feature map size while reducing the channel number by half.

7. A device for adaptive optimization based on repair effect in a road repair process, characterized in that A computer data processing module is provided for receiving the collected data of the device and executing the adaptive optimization method based on the repair effect in the road repair process according to any one of claims 1-6.

Images