A method for delay compensation control of a caterpillar type ship unloader

Abstract

The application relates to the technical field of general control or regulation systems, and discloses a delay compensation control method for a crawler-type stripping machine, which comprises the following steps: time synchronization is performed between a driving end and a vehicle end, the driving end acquires control signals and derivatives thereof in real time, and the control signals and the derivatives are sent to the vehicle end after adding time stamps; the vehicle end receives data packets and analyzes the control signals, the derivatives and the time stamps, calculates one-way transmission delay according to the difference between the current time of the vehicle end and the time stamp, dynamically adjusts an overshoot suppression gain based on the transmission delay, combines a state tracking gain, generates a predicted control instruction through a double-parameter sliding mode predictor, converts the control instruction into left and right wheel speed and hydraulic rod speed instructions through a crawler differential model, performs real-time motion control on the stripping machine through the converted instructions, and monitors transmission delay to update the overshoot suppression gain. The problems of transmission delay, control lag, vehicle oscillation and trajectory deviation in the prior art are solved, and the purposes of overshoot suppression, delay compensation and stability enhancement are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control or regulation systems in general, and in particular to a delay compensation control method for a crawler-type cleaning machine. BACKGROUND

[0002] With the continuous improvement of the automation level of the wharf, unmanned cleaning machines, as key equipment in bulk cargo unloading operations, gradually replace the traditional manual cabin cleaning method, significantly improving the operation efficiency and safety. At present, unmanned cleaning machines mainly rely on remote driving to achieve human-machine separation control, and through remote operation of the driving end, the coordination between the vehicle body and the cabin cleaning operation is realized. Compared with the complete autonomous path planning scheme, the remote driving technology route deployment has low cost and strong control flexibility, and is more suitable for cabin operation under complex conditions, and has become one of the most practical intelligent schemes in the field of cleaning machines. The remote driving system is usually composed of a driving end and a vehicle end, and the two communicate with each other through a wireless network. The driver issues control instructions in real time according to the camera image and state feedback, and the vehicle end receives and executes the corresponding actions. However, this system generally faces three major challenges in actual operation: first, the communication link is prone to instability due to external interference, leading to connection interruption or data congestion; second, there is a time delay and accuracy loss in remote sensing, affecting the accuracy of the driver's judgment; third, there is a dynamic delay in the transmission of control instructions, causing the vehicle to execute and operate out of sync with the operator's intention, affecting trajectory accuracy and response speed. Among them, the time delay problem of the control link is the most prominent, and its causes are complex, including limited wireless channels, bandwidth fluctuations, node shielding, electromagnetic interference, etc. The cleaning machine usually works in a closed cabin, and is co-located with large equipment such as unloading machines in a limited space, which makes the wireless signal prone to shielding and attenuation, becoming the bottleneck of the entire communication link, thereby causing the transmission delay of the control instruction to increase significantly, and in severe cases, even causing the vehicle to run unstable.

[0003] For example, the Chinese patent with publication number CN118092286A discloses a vehicle control method, system, device and storage medium, which provides the following technical solution: a vehicle control method, system, device and storage medium, the method comprising: obtaining reference wheel deflection angle and current centroid position state data of the vehicle, the centroid position state data comprising vehicle position, position error, heading angle error and current vehicle speed; constructing a kinematic model based on the reference wheel deflection angle and the centroid position state data, the control quantity of the kinematic model being a target wheel deflection angle, and the state variable of the kinematic model comprising the position error and the heading angle error of the centroid position; defining a sliding mode surface of the sliding mode control of the kinematic model, the sliding mode surface comprising a first sliding mode surface component corresponding to the position error and its derivative and a second sliding mode surface component corresponding to the heading angle error and its derivative; and obtaining a target wheel deflection angle for vehicle control based on the sliding mode surface and the kinematic model. The present application improves the stability and precision of vehicle control. However, the above-mentioned vehicle control method, system, device and storage medium cannot handle network transmission delay problems, and in the remote driving scene, control command lag is easy to cause vehicle oscillation, trajectory deviation and operation stability decline. It lacks an effective delay compensation mechanism and cannot inhibit the overshoot phenomenon caused by signal mutation, and the operation efficiency is reduced due to frequent adjustment. SUMMARY

[0004] The present application solves the problems of network transmission delay, control lag, vehicle oscillation and trajectory deviation in the prior art, and proposes a crawler-type cleanout machine delay compensation control method, achieving the purposes of overshoot suppression, delay compensation, trajectory tracking precision improvement and operation stability enhancement.

[0005] Further, the present application can adapt to different network delay levels, inhibit the overshoot phenomenon caused by control command mutation in real time, effectively reduce vehicle bumping and trajectory oscillation. At the same time, combined with state tracking gain, stable compensation is maintained when the delay changes, significantly improving the trajectory tracking precision and operation safety of the cleanout machine.

[0006] To achieve the above-mentioned purposes, the present application adopts the following technical solution:

[0007] A crawler-type cleanout machine delay compensation control method, comprising:

[0008] The driving end and the vehicle end are time-synchronized, the driving end obtains control signals and their derivatives in real time, adds time stamps and sends them to the vehicle end;

[0009] The vehicle end receives data packets and parses the control signals, their derivatives and time stamps, calculates the one-way transmission delay according to the difference between the current time of the vehicle end and the time stamp, dynamically adjusts the overshoot suppression gain based on the transmission delay, and generates a predicted control command through a double-parameter sliding mode predictor combined with the state tracking gain.

[0010] Converting the control instruction into left and right wheel speed and hydraulic rod speed instruction through the tracked differential model;

[0011] Real-time motion control of the cleaning machine is performed through the converted instruction, and the transmission delay is monitored to update the overshoot suppression gain.

[0012] The application ensures accurate calculation of time synchronization delay, avoids control lag caused by error accumulation, dynamically adjusts the gain to optimize the prediction performance, reduces the trajectory error in the case of delay, adapts the tracked model to the vehicle motion characteristics, improves the trajectory tracking accuracy, monitors and updates the gain in real time to enhance the system stability, and prevents work interruption.

[0013] As a preferred, the vehicle end extracts the timestamp information after analyzing the data packet, compares it with the local system time, calculates the one-way transmission delay from the end to the end, and uses it as the input data for the subsequent delay compensation calculation of the predictor.

[0014] The accurate quantification of the end-to-end delay is realized, the predictor compensation input is ensured to be accurate, the delay calculation error is reduced, the trajectory deviation is avoided, and the control stability is improved.

[0015] As a preferred, the dynamic adjustment of the overshoot suppression gain based on the transmission delay specifically sets a reference overshoot suppression gain value and a reference delay, compares the actual transmission delay with the reference delay; when the actual transmission delay is greater than the reference delay, the overshoot gain value decreases exponentially with the increase of the delay, and when the actual transmission delay is less than the reference delay, the overshoot gain value increases exponentially with the decrease of the delay, and the specific rate of decay and growth is determined by the decay coefficient; the monitoring of the transmission delay to update the overshoot suppression gain also includes that when the system is in steady-state operation, the predictor will automatically degenerate into a single-parameter state, and only the state tracking gain is introduced.

[0016] Dynamic adaptation to network fluctuations, suppression of overshoot caused by sudden signal changes, in simulation comparison, the overshoot is greatly reduced, the risk of jolt is significantly reduced, the exponential regulation of the gain optimizes the response, the system remains stable when the delay changes, and the system automatically degenerates into a single-parameter structure when it enters steady-state operation, keeping the system stability unchanged, thereby realizing the dual characteristics of 'dynamic response + steady-state stability'.

[0017] As a preferred, the prediction control instruction specifically is that the derivative of the received control instruction is multiplied by the overshoot suppression gain and the historical prediction derivative error, the control instruction is multiplied by the state tracking gain and the historical prediction value error, the two are superimposed, and the derivative signal is integrated to generate the prediction control instruction after delay compensation; the double-parameter sliding mode predictor runs in real time at the vehicle end to suppress the overshoot caused by sudden signal changes.

[0018] Locally deploying the predictor at the vehicle end avoids relying on the network to cause error accumulation, integrates to generate smooth instructions, reduces control lag, and suppresses overshoot to improve vehicle stability.

[0019] As a preferred, the track differential model is specifically: according to the kinematic constraint, the predicted center line speed and angular velocity are input into the track differential model, the left drive wheel angular velocity and the right drive wheel angular velocity are calculated combined with the track center distance and the drive wheel radius, and the predicted value is used for the hydraulic rod speed of the rake joint.

[0020] As a preferred, the time synchronization adopts NTP network time protocol to clock calibrate the system of the driving end and the vehicle end, the driving end controller collects the operation joystick offset in real time, calculates the center line speed and angular velocity of the cleaning machine and the hydraulic rod speed of the rake joint, generates a four-dimensional control instruction vector and obtains its first derivative, and encapsulates the control instruction and the derivative and the high-precision timestamp generated by the NTP protocol into a data packet according to the communication protocol, and sends it to the vehicle end through the wireless network.

[0021] The NTP protocol is used to realize time synchronization, ensure accurate delay calculation, and support reliable prediction with high-precision timestamp. The four-dimensional instruction vector covers the full motion dimension, and the derivative information is combined to optimize the compensation effect and reduce the frequency of manual adjustment.

[0022] As a preferred, the prediction control instruction generated by the double-parameter sliding mode predictor is specifically: the historical prediction derivative error is multiplied by a dynamically adjusted overshoot suppression gain, and the historical prediction value error is multiplied by a fixed state tracking gain, and the two products are added to obtain the prediction instruction derivative correction amount at the current time. The correction amount is superimposed on the control instruction derivative received at the current time as the prediction instruction derivative value; the prediction instruction derivative value is accumulated with a fixed time step from the starting time of the predictor to generate the prediction control instruction value.

[0023] As a preferred, the historical prediction derivative error is specifically: the control instruction derivative received at the current time is subtracted from the last period prediction instruction derivative; and the historical prediction value error is specifically: the control instruction value received at the current time is subtracted from the last period prediction instruction value.

[0024] The error term is defined clearly to ensure that the predictor converges correctly, the historical error calculation improves the compensation accuracy, and the trajectory deviation is reduced.

[0025] As a preferred, the control signal includes the linear speed and angular velocity of the cleaning machine and the two joint drive hydraulic rod speeds of the rake, and the first derivative thereof is calculated in real time by the driving end and packaged and sent.

[0026] As preferred, the calculation of the left drive wheel angular velocity is specifically the predicted linear velocity minus the product of the track pitch and the predicted angular velocity divided by the wheel radius, and the calculation of the right drive wheel angular velocity is specifically the right drive wheel angular velocity equal to the predicted linear velocity plus the product of the track pitch and the predicted angular velocity divided by the wheel radius.

[0027] Compared with the prior art, the present application has the beneficial effects that.

[0028] 1. The present application effectively suppresses the vehicle bump caused by sudden change of control command by introducing a dynamic adjustment mechanism of overshoot suppression gain. The α term combines with the derivative error to form a correction factor, which automatically limits the output change rate when the signal suddenly changes. The hydraulic impact risk caused by sudden turning or sudden stopping is greatly reduced. The dynamic adjustment mechanism further optimizes the adaptability, and the α value increases or decreases exponentially when the actual delay deviates from the reference value, ensuring that the system still maintains stable response under network fluctuations. This design solves the problem of excessive rigidity of the traditional sliding mode predictor response, and avoids trajectory oscillation and operation interruption from the root.

[0029] 2. The double-parameter predictor deployed at the vehicle end of the present application combines the λ state tracking gain tracking term and the NTP timestamp synchronization to achieve high-precision delay compensation and steady-state tracking. The local embedded deployment avoids cloud error accumulation, directly outputs smooth instructions to the track differential model, and converts the predicted linear and angular velocities into left and right wheel speeds in real time through the kinematic model, ensuring that the motion constraints of the tracked vehicle are strictly met, thereby improving the trajectory tracking accuracy from the control link.

[0030] 3. The double-parameter predictor and the dynamic gain adjustment mechanism of the present application form an adaptive closed loop, which significantly improves the tolerance of the system to delay fluctuations. The overshoot suppression gain is dynamically adjusted according to the real-time delay, making the system very stable in both high-delay and low-delay situations. The four-dimensional instruction vector and its derivative are compensated synchronously, ensuring the coordinated control of the ship unloader's travel and operating mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The present application is a whole flow chart of a tracked ship unloader delay compensation control method.

[0032] Figure 2 The present application is a remote driving system model diagram.

[0033] Figure 3 The present application is a remote driving ship unloader kinematic model diagram.

[0034] Figure 4 The present application is a single-parameter sliding film state predictor simulation model diagram.

[0035] Figure 5 The present application is a double-parameter sliding film state predictor simulation model diagram.

[0036] Figure 6 Three signal comparison chart for single parameter slip film state predictor (100 ms delay) of the present application.

[0037] Figure 7 Oscillation stage overshoot (10%) chart for single parameter slip film state predictor (100 ms delay) of the present application.

[0038] Figure 8 Steady state delay compensation (40%) chart for single parameter slip film state predictor (100 ms delay) of the present application.

[0039] Figure 9 Steady state stage overshoot (3%) chart for single parameter slip film state predictor (100 ms delay) of the present application.

[0040] Figure 10 Three signal comparison chart for double parameter slip film state predictor (100 ms delay) of the present application.

[0041] Figure 11 Oscillation stage overshoot (10%) chart for double parameter slip film state predictor (100 ms delay) of the present application.

[0042] Figure 12 Steady state delay compensation (30%) chart for double parameter slip film state predictor (100 ms delay) of the present application.

[0043] Figure 13 Steady state stage overshoot (2%) chart for double parameter slip film state predictor (100 ms delay) of the present application.

[0044] Figure 14 Structure chart of experimental environment of the present application.

[0045] Figure 15 Remote driving position error (no increase in delay) chart of the present application.

[0046] Figure 16 Remote driving position error without delay compensation (200 ms delay) chart of the present application.

[0047] Figure 17 Remote driving position error based on delay compensation (200 ms delay) chart of the present application.

[0048] Figure 18 Remote driving position error without delay compensation (400 ms delay) chart of the present application.

[0049] Figure 19 Remote driving position error based on delay compensation (400 ms delay) chart of the present application. DETAILED DESCRIPTION

[0050] Referring to Figures 1-19 A caterpillar type of straddle carrier time delay compensation control method, comprising:

[0051] The driving end and the vehicle end are time synchronized, the driving end acquires control signals and their derivatives in real time, adds time stamps and sends them to the vehicle end;

[0052] The vehicle end receives data packets and parses control signals, their derivatives and time stamps, calculates one-way transmission delay according to the difference between the current time of the vehicle end and the time stamp, dynamically adjusts the overshoot suppression gain based on the transmission delay, and generates a predicted control instruction through a double-parameter sliding mode predictor in combination with the state tracking gain;

[0053] The control instruction is converted into left and right wheel speed and hydraulic rod speed instructions through a caterpillar differential model;

[0054] The converted instructions are used for real-time motion control of the straddle carrier, and the overshoot suppression gain is updated by monitoring the transmission delay.

[0055] As shown in Figure 1 An embodiment, Figure 1 is a whole flowchart of the caterpillar type of straddle carrier time delay compensation control method. First, the driving end and the vehicle end are time synchronized through the NTP network time protocol, ensuring that the clocks of the two ends are consistent. The driving end acquires the operation rocker offset in real time, calculates the center line speed, angular speed of the straddle carrier and the two joint drive hydraulic rod speed of the rake, generates a four-dimensional control instruction vector, and acquires its first derivative; then, high-precision time stamps are added to the control instruction and its derivative, packaged into data packets and sent to the vehicle end through a wireless network.

[0056] Next, the vehicle end parses the control signal, derivative signal and time stamp information after receiving the data packet, compares the time stamp with the local system time, calculates the one-way transmission delay from the end to the end as an input parameter for the subsequent predictor compensation. Then, the overshoot suppression gain is dynamically adjusted based on the calculated transmission delay: the reference overshoot suppression gain value and the reference delay are set, and the actual delay and the reference delay are compared; if the actual delay is greater than the reference delay, the overshoot gain decays exponentially; if the actual delay is less than the reference delay, the overshoot gain grows exponentially, and the specific rate of decay or growth is determined by the decay coefficient.

[0057] Subsequently, a predicted control instruction is generated through a double-parameter sliding mode predictor: the received control instruction derivative is multiplied by the historical prediction derivative error and then superimposed with the overshoot suppression gain, the control instruction is multiplied by the historical prediction value error and then superimposed with the state tracking gain, the two results are added together and the derivative signal is integrated to generate a predicted control instruction after delay compensation; the predictor runs in real time at the vehicle end, effectively suppressing the overshoot phenomenon caused by sudden signals.

[0058] After that, the predicted centerline velocity and angular velocity are input into the track differential model: according to the kinematic constraint, combined with the track center distance and the driving wheel radius, the left driving wheel angular velocity and the right driving wheel angular velocity are calculated; the predicted value is directly used for the hydraulic rod speed of the rake, without additional conversion. Finally, the converted left driving wheel angular velocity, right driving wheel angular velocity and rake hydraulic rod speed command are written into the vehicle controller to drive the real-time motion control of the cleaning machine through the Modbus TCP protocol; at the same time, the transmission delay data is continuously monitored, and the overshoot suppression gain is dynamically updated to adapt to network changes. When the system is in steady-state operation, the predictor automatically degenerates into a single parameter state, only the state tracking gain is retained to simplify the control, forming a closed-loop control process.

[0059] In another embodiment, the present application is to solve the problem of operation response lag of the remote driving system of the cleaning machine under high network delay, which leads to control overshoot and even oscillation. A remote driving controller based on delay compensation is designed. First, based on the state transmission and transmission delay between the driving end and the vehicle end, a state error equation is constructed, and based on the sliding mode state predictor, the remote control command is predicted at the vehicle end to compensate for the control lag caused by network delay. Second, combined with the kinematic model of the cleaning machine, a remote driving controller of the cleaning machine based on delay compensation is designed to realize the delay compensation of the remote driving of the cleaning machine. Then, based on MATLAB, a remote driving data transmission model based on delay is constructed to evaluate and compare the performance of the two state predictors. Finally, based on the designed remote driving controller of the cleaning machine based on delay compensation, remote driving tests are carried out under multiple levels of network delay. The experimental results show that the designed controller has the best improvement effect on the operation performance of the remote driving system under one-way 200ms (round trip 400ms) up and down, and the designed controller improves the response speed of remote operation and reduces the remote operation error, ensuring the stable and safe remote operation of the cleaning machine under high network delay.

[0060] As Figure 2 The network-based remote driving system model consists of two remote subsystems, the driving end and the vehicle end, coupled by a bidirectional communication network. The network delay causes the remote controlled vehicle to receive delayed control signals, and the response lags behind the driving end. When the delay is large, the driver will have difficulty operating and even cause instability of the remote driving system.

[0061] In the delay compensation control scheme proposed in the application, a sliding mode state predictor is deployed at the vehicle end of the cleaning machine to predictively compensate the remote driving control instructions, thereby alleviating the response lag caused by communication delay. In remote driving applications, sudden mutations or jumps in control input, such as sudden stops, sharp turns, etc., can cause significant overshoot in the predictor output. Such overshoot not only causes sudden acceleration or stoppage of the cleaning machine during travel, resulting in attitude jitter, but also can cause a series of safety risks such as unstable track control and sudden impact on the hydraulic system. The error is added to the predictor as a correction factor a, which is used to limit the rate of change of the predicted value. Specifically, the delayed original state value is subtracted from the predicted state value of the previous period to obtain the historical state error, the historical state error is multiplied by the overshoot suppression gain to generate an overshoot suppression correction amount, the historical state error is multiplied by the state tracking gain to generate a tracking correction amount, and the two correction amounts are added to obtain a total correction amount. The total correction amount is added to the delayed original state value, and the predicted state value at the current time is directly output.

[0062] This structure can effectively limit the output amplitude change of the predictor when the input instruction changes rapidly, avoiding system overshoot. When the system is in steady-state operation, the predictor will automatically degenerate into a single parameter structure, maintaining the system stability unchanged, thereby realizing the dual characteristics of "dynamic response + steady-state stability".

[0063] Then, the Network Time Protocol (NTP) is used to synchronize the time between the driving end and the vehicle end. The driving end controller is used to obtain the operator's operation signal, the first derivative of the signal, and the time stamp accurate to milliseconds, which are packaged and sent to the cleaning machine end controller according to the communication protocol. The controller at the cleaning machine end receives the communication data report and parses the control signal data and the time stamp carried by the signal according to the protocol, calculates the signal transmission delay based on the system time of the cleaning machine end, predicts the signal based on the sliding mode state predictor, and finally calculates the control amount based on the kinematic model of the cleaning machine, writes it into the vehicle controller through Modbus TCP to realize the motion control of the cleaning machine. The operator at the remote driving end operates the handle, and the driving end controller calculates the linear velocity, angular velocity of the cleaning machine, and the speed of the two joint drive hydraulic rods of the rake, to obtain the control instruction . And further calculate its first derivative. To support delay compensation, the controller timestamps the control amount and its derivative with high precision, packages them according to the set communication protocol, and sends them to the cleaning machine end through the wireless network.

[0064] After receiving the communication frame, the end of the stripping machine compares the timestamp with the current time to obtain the end-to-end delay, and then predicts the control instruction generated by the driving end at the next time through the state predictor. The predicted signal is used as the control instruction of the stripping machine end at the current time. Specifically, the historical prediction derivative error is multiplied by a dynamically adjusted overshoot suppression gain, and the historical prediction value error is multiplied by a fixed state tracking gain. The sum of the two products is added to obtain the prediction instruction derivative correction value at the current time. The correction value is superimposed on the current received control instruction derivative to obtain the predicted control instruction derivative value. The historical prediction derivative error is specifically the current received control instruction derivative minus the last period prediction instruction derivative; the historical prediction value error is specifically the current received control instruction value minus the last period prediction instruction value. The dynamic adjustment of the overshoot suppression gain is as follows: a reference overshoot suppression gain value and a reference delay are set, and the actual delay is compared with the reference delay; if the actual delay is greater than the reference delay, the overshoot gain decays exponentially; if the actual delay is less than the reference delay, the overshoot gain grows exponentially, and the specific rate of decay or growth is determined by the decay coefficient.

[0065] Then, from the time when the predictor is started to the time (t+t d1 ), integration is performed to obtain the delay-compensated control instruction. Then, the two sides of the tracked stripping machine are driven by two motors respectively, so the kinematics model of the stripping machine can be simplified as shown in Figure 3 . Since the vehicle body is a steel structure, according to the kinematics constraint, the expected vehicle center line speed v and angular speed ω can be solved and converted into the angular speed of the left drive wheel (v+Lω) / r and the angular speed of the right drive wheel (v-Lω) / r. By solving and converting the delay-compensated vehicle linear speed v 1p (t+t d1 ) and the vehicle angular speed ω p (t+t d1 ), the angular speed of the left drive wheel and the angular speed of the right drive wheel, as well as the v 2p (t+t d1 ) and v 3p (t+t d1 ) data, i.e. the angular speed and the speed of the two joint drive hydraulic rods, are written into the vehicle controller to realize motion control. At the same time, the transmission delay data is continuously monitored, and the overshoot suppression gain is dynamically updated to adapt to network changes. When the system is in steady-state operation, the predictor automatically degenerates into a single parameter state, only retaining the state tracking gain to simplify the control, forming a closed-loop control process.

[0066] The simulation experiment of the application is as follows: according to the simulation experiment, it is confirmed that, in the embodiment, the reference overshoot suppression gain value is 0.15, the reference delay is 100 ms, the attenuation coefficient is -1 / 100ln(2 / 3), the overshoot suppression gain takes the reference gain as the initial value, and is multiplied by the product of the negative attenuation coefficient of the natural constant e and the delay deviation (that is, the actual delay minus the reference delay) raised to the power:

[0067] The remote driving delay compensation method based on a model-free predictor first builds a simulation model in MATLAB / Simulink. The simulation model built as shown in Figure 4 and Figure 5 mainly includes the following modules:

[0068] (1) Signal source: used for simulating the control signal generated at the driving end.

[0069] (2) Time delay module: used for simulating network delay.

[0070] (3) Predictor: constructed based on a single-parameter sliding mode state predictor and a double-parameter sliding mode predictor.

[0071] (4) Oscilloscope: used for observing the original input signal, the delayed input signal, and the predicted signal.

[0072] Single-parameter sliding mode state predictor simulation experiment:

[0073] First, the single-parameter state predictor is set to have a 100 ms delay, a sine signal is used as the input, the non-delayed signal, the delayed signal, and the predicted signal are compared after being input into the predictor, the simulation step is 0.01 seconds, and the parameter λ is set to 13. As shown in Figure 6 the signal comparison from the initial stage of starting for 5 seconds, the delayed signal lags behind the non-delayed signal by 100 ms, the signal has no distortion, the predicted signal oscillates in the initial stage, then immediately stabilizes, and after stabilization, the predicted signal leads the delayed signal and can follow the original non-delayed signal, indicating that the predictor can make correct prediction on the original signal based on the delayed signal.

[0074] Figure 7 is a local enlarged view of No. 1 area in Figure 6 It can be seen that, in the oscillation stage of the predicted signal, about 10% overshoot is generated relative to the non-delayed signal; Figure 8 is a local enlarged view of No. 2 area in Figure 6 It can be seen that, in the stable stage after the predicted signal ends oscillation, the predicted signal lags behind the original signal by only 60 ms, while the set delay is 100 ms, so the predictor generates about 40% compensation on the delayed signal in the stable stage; Figure 9 is a local enlarged view of No. 3 area in Figure 6The enlarged view of region 3 shows the overshoot of the predicted signal in the steady state phase. The overshoot rate is about 2.6% relative to the original signal, and the peak overshoot still leads the peak of the delayed signal.

[0075] The overall results show that the single-parameter predictor can effectively compensate for delayed small signals and tend to stabilize after a short oscillation. However, the overshoot is large during the oscillation phase. In remote driving scenarios, large and sharp overshoot can easily cause sudden changes in speed or steering, resulting in vehicle bumps or vibrations.

[0076] Simulation experiment of two-parameter sliding mode state predictor:

[0077] A two-parameter state predictor was tested under delay, and its overshoot and delay compensation capabilities were analyzed. The delay was set to 100ms, and the input signal was a sinusoidal signal of the same frequency. An overshoot-suppressed predictor was used to predict the delayed signal, with a simulation compensation setting of 0.01 seconds, parameters λ=14, and α=0.15. Figure 10 As shown, similar to the single-parameter predictor, the predicted signal oscillates in the initial stage and then immediately stabilizes, and the predicted signal can follow the original time-delay signal.

[0078] from Figure 11 As can be seen, during the oscillation phase of the predicted signal, only about 5% overshoot was generated compared to the no-delay signal, a 50% reduction compared to the overshoot of the single-parameter predictor; from Figure 12 As can be seen, in the steady-state phase after the predicted signal ends its oscillation, the predicted signal lags behind the original signal by 70ms, which is 10ms longer than that of the single-parameter predictor. Therefore, the overshoot suppression predictor compensates for the delayed signal by about 30% in the steady-state phase, and the delay compensation performance is reduced. Figure 13 The figure shows the overshoot of the predicted signal in the steady-state phase. The overshoot rate is approximately 2% relative to the original signal, and the peak overshoot still leads the peak of the delayed signal. However, the peak signal lags slightly behind the predicted signal from the single-parameter predictor. Simulation results of the dual-parameter predictor show some loss of delay compensation performance compared to the single-parameter predictor, but it demonstrates good performance in suppressing the overshoot of the predicted signal.

[0079] Remote driving experiment:

[0080] The experimental environment structure for the remote piloting experiment of the cabin cleaning aircraft based on delay compensation is as follows: Figure 14As shown, the driving end and the cleaning machine end environment are respectively deployed in two computers. In the driving end computer, the uplink and downlink network delays are set through the network delay tool Network Link Conditioner. The driving test personnel drive the cleaning machine along the ground track according to the driving end picture feedback using the operation handle.

[0081] The cleaning machine end controller in the designed controller is deployed at the cleaning machine end. The control amount (left and right wheel speed and rake hydraulic rod speed) output by the controller is written to the Modbus TCP server. The cleaning machine virtual debugging system reads the control signal from the Modbus TCP server in real time to control the cleaning machine to drive. In the virtual debugging system, the driving track of the cleaning machine is collected and saved in real time, which is used for subsequent analysis of the control performance of the remote driving system designed in this chapter.

[0082] In this experiment, Network Link Conditioner network simulation tool is used to realize accurate regulation and control of network transmission parameters through network driver layer interception technology in the operating system kernel. The tool can simulate various weak network scenarios based on data link layer characteristics, including but not limited to transmission delay, packet loss and bandwidth limitation, etc. The graphical user interface (GUI) of the tool adopts a modular design architecture. The left side of the main control panel integrates the network link conditioner activation control, and the right side realizes the hierarchical call of the configuration interface through a special control unit.

[0083] In the parameter configuration module, researchers can model the network transmission channel bidirectionally and independently: the uplink and downlink are respectively provided with bandwidth threshold setting devices, discrete packet loss probability generators and transmission delay simulators. This decoupled architecture design makes it possible to accurately reproduce the asymmetric network characteristics in the request-response interaction process, providing an experimental basis for evaluating the robustness of the application program under different RTT (Round-Trip Time) scenarios. All configuration parameters are injected into the network protocol stack in real time through kernel extensions (KEXT), ensuring that the time accuracy of the simulation environment reaches milliseconds.

[0084] The cleaning machine virtual debugging system used in the present application adopts a joint development framework based on Unreal Engine 5 (Unreal Engine 5, UE5) real-time rendering platform and AGX Dynamics multi-body dynamics solver, forming a digital twin verification environment with industrial-level precision. Based on the high-precision numerical model of the cleaning machine transmission system constructed based on multi-body dynamics theory, the generalized coordinate method is used to describe the mechanism motion chain, and the implicit integration algorithm is used to solve the joint constraint equation, which can realize accurate simulation of the nonlinear dynamics characteristics of key components such as hydraulic transmission system and planetary gear box.

[0085] In terms of physical interaction modeling, the system realizes the physical simulation of the coupling of multi-rigid-body contact dynamics solution and nonlinear friction model through discrete element contact detection algorithm and continuous collision response mechanism. This architecture can accurately represent the collision energy dissipation, viscoelastic damping effect and Coulomb-viscous mixed friction characteristics between mechanical components during the operation of the stripping machine, and the simulation error of the motion trajectory is controlled within the millimeter level. To simulate the remote driving test through the vehicle-mounted camera, a camera module is added to the virtual stripping machine, which is rigidly connected to the stripping machine body and sends the picture to the driving end to provide environmental information for the operator.

[0086] In terms of geometric modeling, the virtual stripping machine strictly follows the technical specifications of the prototype machine: the overall size of the vehicle is 3m (wide) x 6m (long), and the maximum travel speed is limited to 11km / h. Through the parametric modeling method, more than 150 core parameters such as mass inertia characteristics and hydraulic system pressure-flow curve are configured, so that the dynamic similarity between the virtual prototype and the physical entity reaches more than 90%.

[0087] In this experiment, the designed remote driving system of the stripping machine based on delay compensation was tested under the conditions of no delay, setting a one-way 200ms (round trip 400ms) delay, and setting a one-way 400ms (round trip 800ms) delay. The reference trajectory set in the experiment is about 50 meters per lap, and the stripping machine collects the position and records it once every 0.1 meters of displacement. To reduce the impact of human factors on test results, the operator drives along the set reference trajectory for 10 laps in each test. The driving trajectory of the stripping machine is counted, and the current position error between the driving trajectory and the set trajectory, the current average position error during driving, the average position error and standard deviation of all trajectory points are calculated to analyze the delay compensation performance of the system. The experimental results are as follows:

[0088] (1) No delay: Without adding delay, the delay of control data is usually less than 50ms, and no delay compensation is needed. The purpose of this test is to provide a reference object for the analysis of data in the following cases. The experimental results are shown in Figure 15 Compared with the reference trajectory, the average error tends to be stable after driving about two laps, and the final total average position error is 28.81cm, and the standard deviation is 19.54cm. The position error of the overall driving trajectory is small. From the "reference trajectory and driving trajectory" figure, it can be seen that there is almost no large repeated adjustment of direction during driving.

[0089] (2) Set one-way 200ms (round trip 400ms) delay: In the case of setting 200ms delay, the control system without delay compensation and the control system with delay compensation were tested respectively. The test results of the control system without delay compensation are shown in FIG. 8, compared with the reference trajectory, the total average position error is 42.94 cm, and the standard deviation is 31.89 cm. Compared with the case without delay, the error of the driving trajectory is larger. From the reference trajectory and driving trajectory diagram, it can be seen that during driving, it is easy to deviate greatly, even deviate from the reference trajectory, and it is necessary to adjust the direction frequently to drive along the reference trajectory accurately. Figure 16

[0090] The test results of the control system with delay compensation are shown in FIG. 9, compared with the reference trajectory, the total average position error is 33.70 cm, and the standard deviation is 25.21 cm. Compared with the test results of the control system without delay compensation under 200ms, the total average position error is reduced by 21.5%, and the standard deviation is reduced by 20.9%. From the reference trajectory and driving trajectory diagram, it can be seen that compared with the case without delay compensation, the direction adjustment during driving is obviously reduced. Figure 17

[0091] (3) Set one-way 400ms (round trip 800ms) delay: In the case of setting 400ms delay, the control system without delay compensation and the control system with delay compensation were tested respectively.

[0092] The test results of the control system without delay compensation are shown in FIG. 10, compared with the reference trajectory, the total average position error is 59.26 cm, and the standard deviation is 44.05 cm. Compared with the case without delay, the error of the driving trajectory is doubled. Compared with the case of 200ms delay, the error is obviously increased. From the reference trajectory and driving trajectory diagram, it can be seen that during driving, it is difficult to drive along the reference trajectory, and it is necessary to adjust the driving direction constantly, and it is easy to deviate from the reference trajectory. Figure 18 The test results of the control system with delay compensation are shown in FIG. 11, compared with the reference trajectory, the total average position error is 55.32 cm, and the standard deviation is 42.05 cm. Compared with the test results of the control system without delay compensation under 400ms delay, the average position error is reduced by 6.64%, and the standard deviation is reduced by 4.54%. Under the delay of 400ms, there is a serious lag in the picture feedback, and the delay compensation of the control signal has a lower improvement effect on the overall remote driving compared with the case of 200ms delay.

[0093] Figure 19

[0094] ​​​​The present application is not limited to the above-mentioned embodiments, and any changes in shape or material composition are within the scope of the present application.

Claims

1. A delay compensation control method for a tracked tank cleaning machine, characterized in that, include: The driver's end and the vehicle end synchronize time. The driver's end obtains control signals and their derivatives in real time, adds timestamps, and sends them to the vehicle end. The vehicle receives data packets and parses control signals, their derivatives, and timestamps. It calculates the one-way transmission delay based on the difference between the current time and the timestamp. Based on the transmission delay, it dynamically adjusts the overshoot suppression gain and combines it with the state tracking gain to generate predictive control commands through a two-parameter sliding mode predictor. Set the reference overshoot suppression gain and reference delay, and compare the actual transmission delay with the reference delay. When the actual transmission delay is greater than the reference delay, the overshoot gain decreases exponentially with the increase of the delay. When the actual transmission delay is less than the reference delay, the overshoot gain increases exponentially with the decrease of the delay. The specific rate of decay and increase is determined by the decay coefficient. The control commands are converted into left and right wheel speeds and hydraulic rod speeds through the track differential model; The machine is controlled in real time by converting commands, and the overshoot suppression gain is updated by monitoring transmission delay.

2. The delay compensation control method for a tracked tank cleaning machine according to claim 1, characterized in that, After parsing the data packet, the vehicle terminal extracts the timestamp information and compares it with the local system time to calculate the end-to-end one-way transmission delay, which is then used as input data for the delay compensation calculation of the subsequent predictor.

3. A delay compensation control method for a tracked tank cleaning machine according to claim 1 or 2, characterized in that, The dynamic adjustment of overshoot suppression gain based on transmission delay specifically involves setting a reference overshoot suppression gain value and a reference delay, and comparing the actual transmission delay with the reference delay. When the actual transmission delay is greater than the reference delay, the overshoot gain value decreases exponentially with the increase of the delay. When the actual transmission delay is less than the reference delay, the overshoot gain value increases exponentially with the decrease of the delay. The specific rates of decay and increase are determined by the decay coefficient. The monitoring of transmission delay to update the overshoot suppression gain also includes that when the system is in steady-state operation, the predictor will automatically degenerate into a single-parameter state, introducing only the state tracking gain.

4. The delay compensation control method for a tracked tank cleaning machine according to claim 3, characterized in that, The predictive control command is specifically as follows: the error between the received control command derivative and the historical prediction derivative is multiplied by the overshoot suppression gain, and the error between the control command and the historical prediction value is multiplied by the state tracking gain. The two are superimposed and integrated onto the derivative signal to generate a delay-compensated predictive control command. The dual-parameter sliding mode predictor runs in real time at the vehicle end to suppress overshoot caused by sudden signal changes.

5. The delay compensation control method for a tracked tank cleaning machine according to claim 4, characterized in that, The track differential speed model is specifically as follows: based on kinematic constraints, the predicted center linear velocity and angular velocity are input into the track differential speed model, and the angular velocities of the left and right drive wheels are calculated by combining the track center distance and the drive wheel radius. The speed of the rake hydraulic rod is the predicted value.

6. A delay compensation control method for a tracked tank cleaning machine according to claim 4 or 5, characterized in that, The time synchronization uses the NTP network time protocol to calibrate the clocks of the driver's end and the vehicle end systems. The driver's end controller collects the offset of the operating joystick in real time, calculates the centerline velocity and angular velocity of the cleaning machine and the hydraulic rod velocity of the rake joint, generates a four-dimensional control command vector and obtains its first derivative, and encapsulates the control command and its derivative and the timestamp generated by the NTP protocol into a data packet according to the communication protocol, and sends it to the vehicle end through the wireless network.

7. The delay compensation control method for a tracked tank cleaning machine according to claim 4, characterized in that, The specific steps for generating predictive control commands using a dual-parameter sliding mode predictor are as follows: multiply the historical prediction derivative error by a dynamically adjusted overshoot suppression gain, then multiply the historical prediction error by a fixed-state tracking gain, add the two products together to obtain the current prediction command derivative correction amount, and add the correction amount to the currently received control command derivative as the prediction command derivative value; starting from the predictor startup time, accumulate the prediction command derivative value at fixed time steps to generate the predictive control command value.

8. The delay compensation control method for a tracked tank cleaning machine according to claim 7, characterized in that, The historical prediction derivative error is specifically the derivative of the currently received control command minus the derivative of the prediction command of the previous cycle; the historical prediction value error is specifically the value of the currently received control command minus the value of the prediction command of the previous cycle.

9. A delay compensation control method for a tracked tank cleaning machine according to claim 1 or 2, characterized in that, The control signals include the linear and angular velocities of the cleaning machine and the speeds of the hydraulic rods driving the two joints of the rake shovel. Their first derivatives are calculated in real time by the operator and sent in a package.

10. A delay compensation control method for a tracked tank cleaning machine according to claim 5, characterized in that, The calculation of the left drive wheel angular velocity is specifically the predicted linear velocity minus the product of the track spacing and the predicted angular velocity, divided by the wheel radius. The calculation of the right drive wheel angular velocity is specifically the predicted linear velocity plus the product of the track spacing and the predicted angular velocity, divided by the wheel radius.

Images