Crawler sand shear state prediction and double motor cooperative control system and method

Abstract

This invention discloses a method and system for predicting the shear state of a tracked vehicle in sand and soil and for coordinated control of dual motors. The method includes: collecting vehicle operating status data; constructing a PINN-based digital twin model of track-sand physical information, using sand state parameters as implicit state variables, and embedding vehicle dynamics equations, shear models, and settlement models as physical constraints into the network training to achieve online estimation of sand state parameters and real-time reconstruction of the track-sand interaction state; using the digital twin model to predict the evolution trend of track slip ratio, shear displacement, and shear stress in the short-term time domain, and calculating the shear risk index; establishing a dynamic stability domain constraint boundary based on the current sand state and future shear risk, constructing a multi-objective optimization model, and generating the target torque of the left and right drive actuators in real time; and constructing an error feedback mechanism using the deviation between the actual response and the predicted results to correct the PINN network parameters and physical model parameters online.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a PINN-based system and method for predicting sand shear state and a dual-motor cooperative control system for tracked vehicles. Background Technology

[0002] When high-speed tracked vehicles travel on soft sandy surfaces such as deserts, beaches, and dry riverbeds, the low load-bearing capacity and weak shear strength of the sand make the tracks prone to excessive sinking, increased slip ratio, shear stress exceeding peak values, and even shear failure. This can lead to vehicle yaw instability, loss of steering control, or getting stuck. Existing high-speed tracked vehicle control systems have the following main shortcomings: (1) Lack of online identification capability for continuous mechanical parameters of sand. Existing technologies estimate the road surface adhesion coefficient through dynamic models to achieve anti-skid control, but can only output the global average adhesion coefficient. They have not constructed a sand-specific shear mechanics model, and cannot identify key continuous parameters such as sand shear deformation modulus and internal friction angle. Moreover, the control is a passive torsion limiting after slippage, and it does not have the ability to predict quicksand saturation.

[0003] (2) Lack of analysis of track shearing state based on soil mechanics model. Existing technology identifies the peak adhesion coefficient point by monitoring the slip ratio and using the slope of the real-time curve between the adhesion coefficient and the slip ratio, but it has not established a continuous evolution model of sand shear stress and shear displacement, and cannot comprehensively characterize the evolution process of shear displacement, shear stress and shear saturation.

[0004] (3) The control strategy lacks an active collaborative optimization mechanism that integrates road surface parameters and shear risk. Existing technologies classify and identify road surfaces, but can only identify category labels (such as sand and mud) and cannot identify continuous mechanical parameters such as sand shear modulus; torque distribution relies on an offline lookup table open-loop strategy, which has poor adaptive capability under gradually changing sand density conditions and lacks dynamic stability domain constraints.

[0005] (4) Existing data-driven prediction methods lack physical consistency constraints. Existing prediction methods based on pure neural network models such as LSTM and GRU have poor generalization in sparse sample regions, and the prediction results often violate classical mechanical laws. They do not establish physical constraints for track-sand coupling dynamics and cannot solve the control problems of high-speed tracked vehicles easily getting stuck in sand and slipping in quicksand on sandy roads. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for predicting shear state and coordinating dual-motor control of tracked vehicles on sandy roads based on PINN, overcoming the problems of existing high-speed tracked vehicles in complex sandy environments, such as difficulty in obtaining real-time changes in road surface parameters, difficulty in predicting the evolution of track shear state, and inability of dual-drive actuators to adapt to dynamic road surfaces.

[0007] To achieve the above objectives, the solution of the present invention is: a method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors, comprising the following steps: S1. Collect vehicle operation status data during the operation of the high-speed tracked vehicle. The vehicle operation status data includes vehicle body attitude information, track motion status information, drive execution unit output parameters and inertial measurement information to form a vehicle operation status dataset. S2. Construct a digital twin model of track-sand physical information based on PINN, and use sand state parameters as implicit state variables in the model; embed vehicle dynamics equations, Wong-Bekker sand shear model and Bekker subsidence model as physical constraints into the neural network training process; realize online estimation of sand state parameters through the mapping relationship between the actual vehicle operation response and the track-sand interaction mechanism, and reconstruct the current track-sand interaction state. The state parameters of the sandy soil include the shear deformation modulus, cohesion, internal friction angle, and settlement characteristic parameters of the sandy soil. The current track-sand interaction state includes track slip ratio, track shear displacement, and shear stress; S3. Based on the real-time updated sand and soil state parameters and vehicle operating status, the track-sand physical information digital twin model is used to predict the track-sand interaction state in the short time scale in the future, and the future shear risk index is calculated based on the prediction results. S4. Combining the current sand state parameters and future shear risk index output by the digital twin model of track-sand physical information, establish a dynamic stability domain constraint boundary, construct a multi-objective optimization function including energy consumption, tracking error, shear risk and load balancing of left and right drive execution units, and solve to obtain the target output torque of the left and right drive execution units. S5. Construct a prediction error feedback mechanism by utilizing the deviation between the actual vehicle operation response and the prediction results of the track-sand physical information digital twin model, and correct the network parameters of the PINN and the physical model parameters embedded in the track-sand physical information digital twin model online according to the error changes.

[0008] This invention also provides a PINN-based shear state prediction and dual-motor cooperative control system for tracked vehicles on sandy roads, comprising: The data acquisition layer is used to collect vehicle operating status data during the operation of high-speed tracked vehicles, forming a vehicle operating status dataset. The track-sand physical information digital twin layer is constructed based on the physical information neural network. The sand state parameters are used as the hidden state variables of the model, and the vehicle dynamics equation, Wong-Bekker sand shear model and Bekker settlement model are embedded as physical constraints in the neural network training process to realize online estimation of sand state parameters, real-time reconstruction of track-sand interaction state and prediction of future short-term shear state. The intelligent decision layer is used to calculate the shear risk index SRI based on the sand state parameters and shear risk prediction results output by the digital twin layer, generate dynamic stability domain constraints, construct a dual-motor collaborative torque optimization model, and solve the target output torque of the left and right drive motors in real time. The control execution layer is used to convert the target control quantity output by the intelligent decision layer into actual control commands for the left and right drive motors, and drive the high-speed tracked vehicle to complete the expected movement. The prediction error feedback update module is used to calculate the deviation between the actual vehicle operation response and the prediction results of the digital twin model, and to correct the PINN network parameters and physical model parameters online according to the error changes. The entire system constructs a dual-loop predictive control mechanism that combines a vehicle motion control closed loop with a model adaptive update closed loop.

[0009] After adopting the above solution, the beneficial effects of the present invention are as follows: 1. Possesses online state perception and adaptive capabilities in complex sandy soil environments. By constructing a digital twin model of track-sand physical information, and using sandy soil state parameters as implicit state variables, it achieves continuous online estimation of key mechanical parameters in complex sandy soil environments, overcoming the shortcomings of existing technologies that can only output discrete pavement categories or global average adhesion coefficients.

[0010] 2. Possesses the ability to predict the future evolution of track shear conditions, enabling early risk assessment. Utilizing a digital twin model to identify the state of sand and soil in real time, it predicts the evolution trends of track slip ratio, shear displacement, and shear stress over short timescales, achieving early assessment of track shear overload and failure trends. This overcomes the lag problem of existing technologies that can only passively limit torsion after slippage occurs.

[0011] 3. Achieve intelligent collaborative torque distribution between dual motors based on shear risk constraints. Based on the current state output by the digital twin model and the prediction results of future shear risk, a dynamic stability constraint boundary is established (which dynamically shrinks with SRI). The torque distribution between the left and right motors is adjusted in real time using a multi-objective optimization method, so that the high-speed tracked vehicle can maintain stable operation in complex sandy environments.

[0012] 4. Possesses online correction capability for digital twin models, improving long-term operational reliability. Through a prediction error feedback update mechanism, the PINN network parameters and physical model parameters are corrected online, enabling the model to continuously approximate the real track-sand interaction process and adapt to gradually changing working conditions of sand density, thickness, and moisture content.

[0013] 5. Construct a dual-closed-loop predictive control mechanism. This transforms the traditional passive control based on real-time state response into active control based on future state prediction, significantly improving the passability, steering stability, and energy utilization efficiency of high-speed tracked vehicles in complex sandy environments. Attached Figure Description

[0014] Figure 1 This is a block diagram of the overall system structure of the present invention.

[0015] Figure 2 This is the mechanism of the Physical Information Neural Network (PINN) of this invention.

[0016] Figure 3 This is a diagram of the PINN online adaptive update mechanism of the present invention.

[0017] Figure 4 This is a flowchart of the overall method of the present invention.

[0018] Figure 5 This is a flowchart of the online identification process of the present invention.

[0019] Figure 6 This is a flowchart of the shear prediction process of the present invention.

[0020] Figure 7 This is a flowchart of the dynamic stability domain calculation of the present invention.

[0021] Figure 8 This is a flowchart of the intelligent torque distribution process of the present invention. Detailed Implementation

[0022] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] This invention provides a method for predicting the sand shear state of tracked vehicles and for coordinated control of dual motors, such as... Figures 1 to 8 As shown, its core lies in constructing a digital twin model of track-sand physical information that integrates data-driven approaches and physical mechanisms, and realizing predictive control and adaptive optimization based on this model. The method includes the following steps: S1. Collect vehicle body attitude information, track motion state information, drive execution unit output parameters, and inertial measurement information during the operation of the high-speed tracked vehicle to form a vehicle operation state dataset; the drive execution unit is a motor. S2. Input the vehicle operating status dataset into the track-sand physical information digital twin model based on the Physical Information Neural Network (PINN). The model uses the sand state parameters as the model's hidden state variables and embeds the vehicle dynamics equations, track-ground interaction model, Wong-Bekker sand shear model, and Bekker subsidence model as physical constraints into the neural network training process.

[0024] By mapping the actual operating response of the vehicle to the interaction mechanism between the track and the sand, the key state parameters such as the shear deformation modulus, cohesion, internal friction angle and settlement characteristic parameters of the sand can be estimated online, and the current track-sand interaction state can be reconstructed.

[0025] S3. Based on real-time updated sand and soil condition information and vehicle operating status, a digital twin model of track-sand physical information is used to further predict the evolution of key state parameters such as track slip ratio, shear displacement, shear stress, and steering resistance torque in the short timescale of the future. The prediction results are used to calculate the future shear saturation risk, thereby enabling early judgment of track shear failure trends.

[0026] S4. Combining the current sand condition and future shear risk prediction results output by the digital twin model, a dynamic constraint boundary considering sand adhesion capacity, track shear stability, and vehicle motion performance is established to construct a dual-motor collaborative torque optimization model. By comprehensively considering vehicle stability, drive efficiency, and energy consumption indicators, the target output torque of the left and right drive motors is generated in real time.

[0027] S5. A prediction error feedback mechanism is constructed by utilizing the deviation between the actual vehicle operating response and the digital twin prediction results. Based on the error changes, the PINN network parameters and the track-sand coupling model are corrected online, enabling continuous updating of the digital twin model and improving the accuracy of future shear state predictions and adaptability to complex sandy environments. Ultimately, this transforms high-speed tracked vehicles from traditional passive control based on real-time state response to active control based on future state prediction, improving the vehicle's passability, stability, and intelligent control level in complex sandy environments.

[0028] The present invention also provides a tracked vehicle sand shear state prediction and dual-motor cooperative control system, which is applied to the above-mentioned control method.

[0029] like Figure 1As shown, the overall architecture of the system includes a data acquisition layer, a track-sand physical information digital twin layer, an intelligent decision-making layer, a control execution layer, and a prediction error feedback update module. The entire system works collaboratively according to the technical link of "vehicle operation status acquisition - track-sand physical information digital twin modeling - shear risk prediction - dual-motor intelligent control - prediction error feedback update", constructing a dual-closed-loop predictive control mechanism that combines vehicle operation control closed loop and model adaptive update closed loop.

[0030] (a) Data Acquisition Layer The data acquisition layer, located at the bottom layer of the entire control system, is responsible for acquiring real-time vehicle operating status and actuator feedback information, providing real-time input for the track-sand physical information digital twin layer. The data acquisition layer includes a vehicle operating status acquisition module and an execution feedback acquisition module. The vehicle operating status acquisition module is used to acquire vehicle operating status data in real time. The vehicle operating status vector is defined as follows:

[0031] Where ν represents the vehicle's center of gravity velocity; ω represents the yaw rate; a x Indicates longitudinal acceleration; a y Indicates lateral acceleration; ν L Indicates the speed of the left track; ν R Indicates the speed of the right track; n L Indicates the speed of the left drive motor; n R Indicates the speed of the right drive motor; T L T represents the output torque of the left motor. R This indicates the output torque of the right motor; SOC indicates the state of charge of the power battery.

[0032] Based on the collected data, calculate the slip ratio of the left track and the slip ratio of the right track: ,

[0033] Among them, the theoretical speed of the track ν t =rω m r is the radius of the drive wheel, ω m Indicates the angular velocity of the drive wheel; theoretical speed of the left track. ν tL =rω mL Theoretical speed of the right track ν tR =rω mR .

[0034] The vehicle's operating state vector is composed of vehicle motion parameters, track slippage state, drive motor output torque, and battery state of charge; among which, the travel speed... νThe yaw rate ω is obtained using an inertial sensor; the left and right track slip ratios are calculated using the track slip ratio definition formula, combined with the drive wheel speed and the vehicle's longitudinal speed; the motor output torque is collected in real time by the drive unit; and the power battery SOC is monitored by the battery management system. Therefore, the vehicle's operating status is uniformly represented as:

[0035] The vehicle's operating state vector serves as the input for subsequent prediction of the digital twin layer of track-sand physical information.

[0036] (ii) Track-soil physical information digital twin layer This layer is a digital twin model of the interaction between the vehicle and the sandy road surface, built based on PINN, such as... Figure 2 and Figure 5 As shown, real-time vehicle operating status is utilized. and control input By incorporating the physical mechanism constraints of the embedded PINN neural network training process, the model can simultaneously realize the state parameters of sandy soil. Online estimation, real-time reconstruction of track-sand interaction state (referring to a set of state variables describing the mechanical behavior of track-sand contact, including track slip ratio, track shear displacement, shear stress, etc.), and prediction of future short-term shear state.

[0037] 1. Input / Output Definition Input, current: Vehicle operating state vector

[0038] Desired control input vector

[0039] Sandy soil state parameters

[0040] Future prediction time domain length Time step Δt Output, Future Within the time domain: Tracked vehicle operating status prediction sequence ; Left and right track slip ratio prediction sequence , ; Left and right track shear displacement prediction sequence , ; Left and right shear stress prediction sequences , ; 2. State representation of digital twin models Establish a vehicle-sand dynamic evolution model:

[0041] The specific unfolding of the tracked vehicle's operating state X in the digital twin model is as follows:

[0042] in, ν x Indicates the speed of the vehicle's center of gravity; ω Indicates yaw rate; λ Indicates track slip ratio; j Indicates shear displacement; τ M represents shear stress; r This indicates the steering resistance torque.

[0043] Dual motor drive input Sandy soil state parameters

[0044] in, c is the shear modulus; c is the soil cohesion. It is the internal friction angle; For cohesive settlement parameters, For friction settlement parameters, These are the subsidence parameters.

[0045] Obtain current sand and soil state parameters P road Then, the digital twin model is based on the current vehicle operating status. X ( t ), control input U (t) and sand state parameters P road Predicting future vehicle operating status:

[0046] and left and right track slip ratio prediction sequences , Left and right track shear displacement prediction sequence , Left and right shear stress prediction sequences , The final result is the future trend of track-sand interaction. Here, k = 0, 1, 2, ..., K is the prediction step number, and Δt is the time step.

[0047] 3. Physical model constraints To prevent the neural network from deviating from actual vehicle behavior, multi-physics model constraints are introduced, and a PINN loss function is constructed. The physical model constraints include the vehicle dynamics equations, the Wong-Bekker sand shear model, and the Bekker subsidence model. The vehicle dynamics equations constrain the macroscopic motion of the vehicle; the Wong-Bekker sand shear model constrains the tangential interaction between the track and the sand, and the Bekker subsidence model constrains the normal interaction between the track and the sand; the Wong-Bekker sand shear model and the Bekker subsidence model together constitute the track-ground interaction model.

[0048] Vehicle dynamics constraints Longitudinal dynamics:

[0049] Lateral dynamics:

[0050] Yaw dynamics:

[0051] Where m represents the vehicle mass; ν x Indicates the longitudinal speed of the vehicle; ν y ω represents the vehicle's lateral velocity; ω represents the yaw rate. F represents yaw acceleration. xL F xR Indicates the longitudinal driving force of the left and right tracks; F yL F yR F represents the lateral force of the left and right tracks; f Indicates driving resistance; I z B represents the vehicle's moment of inertia about its vertical axis; M represents the track center distance; r This indicates the steering resistance torque.

[0052] Wong-Bekker sand shear model constraints

[0053] in, σ represents soil shear stress; σ represents normal stress; j represents shear displacement of the track contact point.

[0054] Bekker sinking model constraints

[0055] in, p Indicates the ground bearing pressure; b Indicates the width of the track pads; z This indicates the amount of subsidence.

[0056] Table 1. Parameters of a typical sandy pavement mechanical model

[0057] 4. PINN Loss Function The overall loss function is:

[0058] The definitions of each item are as follows: Data error terms: This ensures that the network prediction results are consistent with the training data.

[0059] Vehicle dynamics constraints: This ensures that the prediction results conform to the laws of vehicle motion.

[0060] Wong-Bekker shearing constraint: This ensures that the shear stress meets the shearing mechanism of sand and soil.

[0061] Bekker sinking constraint: This ensures that the subsidence prediction conforms to the soil bearing capacity law.

[0062] α, β, γ These are the weighting coefficients.

[0063] During the model initialization phase, the initial weights of the data constraint terms, vehicle dynamics constraint terms, shear model constraint terms, and sinking model constraint terms are set to 0.5, 0.25, 0.15, and 0.10, respectively, and are dynamically adjusted according to the risk status.

[0064] As the risks of slippage, shearing, and sinking increase during vehicle operation, the weights of each constraint are dynamically adjusted using operational status criteria to increase the effectiveness of shear model constraints and sinking model constraints under hazardous conditions.

[0065] Enhance vehicle operating state constraints (β) under low-risk conditions to improve motion state prediction accuracy; increase track-sand interaction constraint weights (γ and γ) under high shear risk conditions. This enhances the ability to predict shear states, enabling the PINN digital twin model to balance data fitting capabilities with track-soil physical consistency.

[0066] 5. Training and Online Updates Offline pre-training phase: Before the vehicle is actually put into operation, a high-fidelity simulation dataset covering a variety of typical sandy soil conditions is first constructed, and measured samples of soil shear and settlement in the soil trough are integrated. After normalization and hierarchical equalization sampling, the dataset is divided into training, verification and testing subsets.

[0067] A three-stage progressive weighted loss scheme is adopted for iterative training, focusing on data fitting, strengthening the physical constraints of Wong-Bekker sand shear and Bekker subsidence, and fine-tuning the accuracy of high-risk quicksand conditions in stages. Cosine annealing Adam optimizer is used to prevent overfitting. After training, the model accuracy is verified by two indicators: identification of sand mechanical parameters and prediction of short-time domain shear evolution. After passing the verification, the network weights and soil model baseline parameters are solidified as the initial model for online operation. This effectively improves the accuracy of sand state identification during the actual vehicle start-up stage, shortens the online adaptive convergence cycle, and avoids distorted predictions that violate the laws of sand mechanics in a purely data-driven network.

[0068] Online fine-tuning phase: During vehicle operation, the system performs online adaptive updates according to the following method: real-time state prediction – prediction error assessment – ​​condition-triggered update. Specifically, this includes: ① Collect the actual vehicle operation response at the current time t ; ②The current vehicle operating status and control input Input the current PINN model and calculate the prediction results. ; ③ Calculate the prediction error ; ④ If the prediction error This will trigger online fine-tuning; This is the preset prediction error threshold; ⑤ Retrieve the most recent N from the sliding time window buffer b Each sample constitutes a mini-batch training dataset, and one or more gradient descent updates are performed. ⑥ Update the network parameters of PINN, sand state parameters, and Wong-Bekker model parameters to achieve continuous learning and improved prediction accuracy of track-sand state in complex sandy environments.

[0069] ⑦ After the update is completed, the updated model parameters will be used for state prediction and torque optimization in the next control cycle.

[0070] 6. Prediction The track-sand physical information digital twin layer transmits the predicted future vehicle operating state and track-sand interaction state to the intelligent decision-making layer. Based on this predicted information, the intelligent decision-making layer identifies potential shear saturation risks, uses the prediction results as feedforward information to generate dynamic constraint boundaries, and performs collaborative optimization allocation of dual-motor torque, thereby improving the system's foresight and robustness.

[0071] (III) Intelligent Decision-Making Layer The intelligent decision-making layer is mainly used to complete the calculation of shear risk index, vehicle stability domain generation, and intelligent torque optimization control of dual motors.

[0072] 1. Shear state analysis First, based on the sand state parameters output by the digital twin layer of track-sand physical information. Based on the current vehicle operating status The Wong-Bekker sand shear model is updated in real time.

[0073] Based on the above model, combined with the vehicle's operating state at the current time t. Calculate the shear displacement j of the left and right tracks. L j R and shear stress τ L , τ R :

[0074]

[0075] Forming a real-time shear state vector:

[0076] Further prediction of the location where the shear stress reaches its peak value is used to obtain the shear saturation state of the track.

[0077] 2. Calculation of Shear Risk Index (SRI): Based on the output of the digital twin layer, λ, j, τ and Calculate SRI:

[0078] in, The critical slip ratio for shear failure. This indicates the proportion of the current slip ratio to the critical slip ratio for shear failure; This represents the critical shear displacement corresponding to the occurrence of significant shear failure in sandy soil. This indicates the degree to which the current shear displacement approaches the soil's ultimate shear state. Peak shear strength, This indicates the degree to which the current shear stress has reached the soil's maximum bearing capacity. This refers to the slip risk weighting coefficient; This refers to the shear deformation risk weighting coefficient; This is the shear stress risk weighting coefficient; and .

[0079] Shear risk assessment: Table 2 Shear Risk Assessment Indicators

[0080] 3. Generation of Dynamic Stable Domain (DSB): like Figure 7 As shown, by combining vehicle speed, yaw rate, adhesion shear stress utilization rate and shear risk index, the vehicle stability domain is dynamically calculated to obtain the vehicle's current maximum allowable yaw rate, maximum slip ratio and allowable motor output range.

[0081]

[0082] Maximum permissible yaw rate:

[0083] Maximum permissible slip ratio:

[0084] Maximum permissible output torque:

[0085] Where ω0, λ0, and T0 are the maximum permissible values ​​under the reference working condition.

[0086] 4. Dual-motor coordinated torque optimization like Figure 8 As shown, under the constraints of the Dynamic Stability Domain (DSB), a dual-motor multi-objective optimization control model is established with vehicle stability, steering tracking error, motor energy consumption, and load balancing of the left and right motors as optimization objectives. The objective function of intelligent torque optimization (TVA) is:

[0087] Constraints: , , .

[0088] in, E This indicates the energy consumption during the dual-motor drive process; This represents the error between the actual yaw rate and the target yaw rate; SRI is used not only for stability region constraints but also as a continuous optimization index to reduce the probability of vehicles entering high-risk areas; T L T R This refers to the output torque of the left and right motors; Optimize the weighting coefficients for energy consumption; For steering tracking weighting coefficients; This refers to the shear risk weighting coefficient; The weighting coefficient for load balancing between the left and right motors; and .

[0089] And satisfy the constraints: , , .

[0090] Optimization yields the optimal target output torque. and The signal is then input into the motor controller for execution, achieving coordinated distribution of driving force between the left and right tracks.

[0091] 5. Risk identification and prediction penalty mechanism To further ensure vehicle safety under extreme operating conditions, this solution introduces a risk prediction mechanism based on shear evolution trends. The system predicts future time t+Δt based on the current state. p (Δt) p (for predicting the shear stress in the time domain) With shear displacement .

[0092] When the predicted value exceeds the critical threshold, that is, when the condition is met... or When the vehicle enters a warning / dangerous state, a safety penalty term J is dynamically added to the torque optimization model. penalty The corrected overall objective function is:

[0093] in, The basic multi-objective optimization function (i.e., the one mentioned above) ), The definition is as follows:

[0094] In the formula, a and b are safety weight coefficients, and the max(0, x) function is used to ensure that a penalty is only generated when the threshold is exceeded, and the term is 0 when the threshold is not exceeded.

[0095] By introducing The optimizer will prioritize suppressing torque distribution schemes that may lead to soil shear failure or vehicle sideslip, while meeting basic driving requirements, thereby achieving a control strategy that "prioritizes vehicle stability".

[0096] (iv) Control Execution Layer The execution control layer includes left and right drive motor controllers, drive motors, and track actuators, and is responsible for transmitting the target control quantity output by the intelligent decision-making layer—the target output torque of the left and right motors. , The commands are converted into actual control commands for the left and right drive motors, which then drive the high-speed tracked vehicle to complete the expected movement.

[0097] After the vehicle completes its operation, its real-time operating status is transmitted back to the data acquisition layer via sensors, forming a closed loop. The real-time vehicle operating status data collected by the data acquisition layer is used to update the digital twin model of track-sand physical information and to provide closed-loop feedback for subsequent control decisions, thereby realizing real-time perception and control of the vehicle's operating status.

[0098] (v) Prediction error feedback and update mechanism like Figure 3 As shown, in order to further improve the control accuracy under complex road conditions, this invention establishes a digital twin adaptive update mechanism based on prediction error feedback, which compares the actual vehicle response with the shear prediction results in real time.

[0099] When the prediction error exceeds a preset threshold, the network parameters of PINN and the embedded physical model parameters in the digital twin model of track-sand physical information are corrected online according to the error change trend, so that the digital twin model can continuously approximate the real track-sand interaction process, thereby improving the prediction accuracy of future shear states and the adaptability to complex sandy environments.

[0100] The prediction error is calculated as follows:

[0101] Where represents the comprehensive prediction error at the current time t; N represents the number of samples involved in the error calculation or the dimension of the output variables; This represents the actual operating response of the i-th vehicle at time t. This is a predicted value; or

[0102] This represents the actual measured operating response vector of the vehicle. This represents the runtime response vector predicted by the digital twin model. This represents the deviation vector between the actual response and the predicted response; When e(t) > Model updates are triggered at specific times. This is the preset prediction error threshold.

[0103] The control closed loop is: (e.g.) Figure 3 (Solid line shown) Data acquisition layer → Track-sand physical information digital twin layer → Intelligent decision-making layer → Execution control layer → Data acquisition layer, realizing real-time perception and control of vehicle operating status.

[0104] Model adaptive update closed loop: (e.g.) Figure 3(As shown by the dashed line) Actual vehicle response → Predicted response → Error calculation → Correction of PINN network parameters and physical model → Road condition update → Intelligent decision layer, realizing online correction of road surface model and continuous improvement of prediction accuracy.

[0105] This architecture and dual closed-loop mechanism enable the present invention not only to achieve road surface adaptation and shear saturation prediction, but also to continuously correct model parameters according to the actual operating state of the vehicle, thereby significantly improving the steering stability, passability and energy utilization efficiency of high-speed tracked vehicles under complex road conditions.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the design of this case. All equivalent changes made based on the key design features of this case shall fall within the protection scope of this case.

Claims

1. A method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors, characterized in that, Includes the following steps: S1. Collect vehicle operation status data during the operation of the high-speed tracked vehicle. The vehicle operation status data includes vehicle body attitude information, track motion status information, drive execution unit output parameters and inertial measurement information to form a vehicle operation status dataset. S2. Construct a digital twin model of track-sand physical information based on PINN, and use sand state parameters as implicit state variables of the model; embed vehicle dynamics equations, Wong-Bekker sand shear model and Bekker subsidence model as physical constraints into the neural network training process. By mapping the actual operating response of the vehicle to the interaction mechanism between the track and the sand, the online estimation of the sand state parameters is achieved, and the current track-sand interaction state is reconstructed. The state parameters of the sandy soil include the shear deformation modulus, cohesion, internal friction angle, and settlement characteristic parameters of the sandy soil. The current track-sand interaction state includes track slip ratio, track shear displacement, and shear stress; S3. Based on the real-time updated sand and soil state parameters and vehicle operating status, the track-sand physical information digital twin model is used to predict the track-sand interaction state in the short time scale in the future, and the future shear risk index is calculated based on the prediction results. S4. Combining the current sand state parameters and future shear risk index output by the digital twin model of track-sand physical information, establish a dynamic stability domain constraint boundary, construct a multi-objective optimization function including energy consumption, tracking error, shear risk and load balancing of left and right drive execution units, and solve to obtain the target output torque of the left and right drive execution units. S5. Construct a prediction error feedback mechanism by utilizing the deviation between the actual vehicle operation response and the prediction results of the track-sand physical information digital twin model, and correct the network parameters of the PINN and the physical model parameters embedded in the track-sand physical information digital twin model online according to the error changes.

2. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: The vehicle operating state data mentioned in step S1 is in the form of a vehicle operating state vector. express; Where ν represents the vehicle's center of gravity velocity; ω represents the yaw rate; a x Indicates longitudinal acceleration; a y Indicates lateral acceleration; ν L Indicates the speed of the left track; ν R Indicates the speed of the right track; n L Indicates the speed of the left drive motor; n R Indicates the speed of the right drive motor; T L T represents the output torque of the left motor. R This indicates the output torque of the right motor; SOC indicates the state of charge of the power battery.

3. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: The sand state parameters mentioned in step S2 are expressed as follows: in, c is the shear modulus; c is the soil cohesion. It is the internal friction angle; For cohesive settlement parameters, For friction settlement parameters, These are the subsidence parameters.

4. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: The track-soil interaction state in the future short timescale in step S3 includes the track slip rate in the future time domain. Track shear displacement Track shear stress and steering resistance torque M r .

5. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: The comprehensive loss function of PINN mentioned in step S2 is: in, This is the data fitting error term. These are the constraint terms in the vehicle dynamics equations. For the Wong-Bekker shearing model constraint terms, For the Bekker sinkhole model constraint terms; α, β, γ, These are the weighting coefficients; The weighting coefficients are dynamically adjusted according to the vehicle's operating risk status: under low shear risk conditions, the β weight is increased to enhance the accuracy of vehicle operating status prediction; under high shear risk conditions, the γ and ξ weights are increased to enhance the constraint strength of the physical mechanism of track-sand interaction.

6. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: In step S3, the shear risk index SRI is calculated as follows: Where λ is the track slip ratio; j is the shear displacement; and τ is the shear stress. The critical slip ratio for shear failure; This represents the critical displacement corresponding to the occurrence of significant shear failure in sandy soil. Peak shear strength; , , These are the weighting coefficients; and ; When 0 < SRI < 0.4, it is considered a stable state; when 0.4 < SRI < 0.7, it is considered a warning state; and when 0.7 < SRI < 1, it is considered a dangerous state.

7. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: In step S4, the dynamic stability domain constraint boundary is dynamically adjusted according to the shear risk index, specifically including: Maximum permissible yaw rate: ; Maximum permissible slip ratio: ; Maximum permissible output torque: ; Where ω0, λ0, and T0 are the maximum allowable values ​​under the reference working condition; μ is the road adhesion coefficient; and SRI is the shear risk index.

8. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: In step S4, the objective function of the multi-objective optimization function is: Constraints: , , ; in, E This indicates the energy consumption during the dual-motor drive process; T represents the error between the actual yaw rate and the target yaw rate. L T R This refers to the output torque of the left and right motors; , , , These are the weighting coefficients, and .

9. The method for predicting the sand shear state of a tracked vehicle and for coordinated control of dual motors as described in claim 1, characterized in that: It also includes risk identification and prediction penalty mechanisms: When the predicted future shear stress Exceeding peak shear strength or predicted future shear displacement Exceeding the critical shear displacement At that time, a safety penalty term J is dynamically added to the torque optimization model. penalty : In the formula, Δt p For the prediction time domain; a and b are safety weight coefficients.

10. A tracked vehicle sand shear state prediction and dual-motor cooperative control system, applied to the control method as described in any one of claims 1 to 9, characterized in that: The data acquisition layer is used to collect vehicle operating status data during the operation of high-speed tracked vehicles, forming a vehicle operating status dataset. The track-sand physical information digital twin layer is constructed based on the physical information neural network. The sand state parameters are used as the hidden state variables of the model, and the vehicle dynamics equation, Wong-Bekker sand shear model and Bekker settlement model are embedded as physical constraints in the neural network training process to realize online estimation of sand state parameters, real-time reconstruction of track-sand interaction state and prediction of future short-term shear state. The intelligent decision layer is used to calculate the shear risk index SRI based on the sand state parameters and shear risk prediction results output by the digital twin layer, generate dynamic stability domain constraints, construct a dual-motor collaborative torque optimization model, and solve the target output torque of the left and right drive motors in real time. The control execution layer is used to convert the target control quantity output by the intelligent decision layer into actual control commands for the left and right drive motors, and drive the high-speed tracked vehicle to complete the expected movement. The prediction error feedback update module is used to calculate the deviation between the actual vehicle operation response and the prediction results of the digital twin model, and to correct the PINN network parameters and physical model parameters online according to the error changes. The entire system constructs a dual-loop predictive control mechanism that combines a vehicle motion control closed loop with a model adaptive update closed loop.

Images