Automatic Material Stacking Method and System for Unmanned Loaders

Abstract

This invention discloses an automated stacking method and system for an unmanned loader. The method includes: establishing a multi-sensor fusion architecture; constructing a digital twin model; updating material model parameters in real time based on measured pressure-displacement data during the stacking process; using the digital twin model as a predictor, establishing a state transition model with factors including stacking height, slope deviation, and bucket posture as state variables, and factors including pushing speed and hydraulic pressure as control variables, and setting an objective function; solving for the optimal control sequence for the next 3-5 cycles in each control cycle; receiving MPC commands through a coordinated electro-hydraulic proportional valve group and steering servo motor, combined with a hydraulic pressure-flow composite control strategy, and using real-time scanning feedback from a three-dimensional lidar to achieve control command execution and closed-loop error correction. This achieves stable and efficient automated stacking operations.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control of engineering machinery, and more specifically, relates to an automatic material stacking method and system for an unmanned loader. Background Technology

[0002] In the field of construction machinery, the intelligent transformation of loader stacking operations faces many thorny challenges. Currently, traditional loaders rely heavily on manual operation, which is not only inefficient but also fails to meet the stringent requirements of high-precision stacking, especially in complex operating scenarios such as mines and ports, where their shortcomings become glaringly apparent.

[0003] From an environmental perception perspective, existing solutions generally suffer from the drawback of relying on a single sensor configuration. Most systems rely solely on lidar to monitor the environment, and in the harsh environment of a dusty mine, the point cloud data loss rate soars to 35%-40%, undoubtedly posing a significant safety hazard. Pure vision solutions experience accuracy fluctuations exceeding 40% when lighting conditions change, and are completely ineffective when dealing with transparent or reflective materials. More critically, the vast majority of existing technologies lack predictive consideration for dynamic obstacles, resulting in an emergency stop rate exceeding 25% when equipment is moving material piles, severely disrupting operational continuity.

[0004] In terms of control accuracy and system adaptability, traditional PID control methods suffer from severe parameter lag when material properties change, resulting in stockpile height errors exceeding ±10cm. Furthermore, due to a lack of effective modeling of material rheological properties, 91% of existing systems still use fixed control parameters for materials with varying moisture content and particle sizes, ultimately leading to stockpile density differences exceeding 15% and significantly reduced stockpile stability. In addition, the coordination efficiency of the perception-decision-execution link is extremely low, with delays generally exceeding 200ms, causing a severe disconnect between control commands and real-time operating conditions.

[0005] Energy consumption and environmental adaptability are also bottlenecks restricting technological development. The control power consumption of a typical system can reach 8kW, resulting in a 35% reduction in the range of electric loaders. In low-temperature environments of -20℃, the hydraulic response speed of 65% of the system drops by more than 40%, making it impossible to meet the needs of all-weather operation. In summary, existing technologies have significant shortcomings in core areas such as multi-sensor fusion, intelligent control methods, and system collaborative architecture. This makes it difficult for stockpiling operations to meet the actual needs of modern engineering in terms of key indicators such as accuracy, efficiency, stability, and environmental adaptability, urgently requiring breakthroughs through technological innovation. Summary of the Invention

[0006] This invention aims to address the problems of existing loader-based material stacking operations, such as reliance on manual labor, low efficiency, insufficient accuracy, and poor environmental adaptability. It improves the reliability of environmental recognition through multimodal perception fusion technology, optimizes material stacking accuracy through digital twins and model predictive control, shortens system response latency by constructing an edge-cloud collaborative architecture, and innovates electro-hydraulic drive and safety mechanisms to achieve intelligent material stacking operations applicable to various scenarios such as mines and ports.

[0007] In view of the above-mentioned defects or improvement needs of the existing technology, as a first aspect of the present invention, the present invention provides an automatic material stacking method for an unmanned loader, comprising:

[0008] S1. Establish a multi-sensor fusion architecture to achieve all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while configuring device status monitoring sensors.

[0009] S2. Construct a digital twin model that includes equipment dynamics, hydraulic system and material rheological properties; the material rheological parameter library is classified and stored according to material type and working condition, and the material model parameters are updated in real time according to the measured pressure-displacement data during the stacking process.

[0010] S3. Using a digital twin model as a predictor, factors including stockpile height, slope deviation, and bucket posture are used as state variables, and factors including pushing speed and hydraulic pressure are used as control variables. A state transition model is established, and an objective function is set. The optimal control sequence for the next 3-5 cycles is solved within each control cycle, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver is used to optimize the control parameters, and sparse matrix operations are used to reduce the computational load.

[0011] S4. By receiving MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor, and combining the hydraulic pressure-flow composite control strategy, the execution of control commands and the correction of closed-loop errors are realized through real-time scanning feedback by three-dimensional lidar.

[0012] Furthermore, the multi-sensor fusion architecture in S1 is specifically as follows:

[0013] Point cloud data is collected by LiDAR and used for 3D environment modeling and geometric feature extraction to build terrain and obstacle outlines in real time.

[0014] By acquiring image information through a visual camera and processing it using deep learning methods, it can complete recognition tasks including color, texture and semantics, as well as classification tasks including personnel, equipment and material types, thus supplementing the shortcomings of LiDAR in material recognition.

[0015] By transmitting electromagnetic waves and receiving reflected signals through millimeter-wave radar, and combining signal processing methods for analysis, it can complete measurement tasks including target distance, speed and azimuth, as well as detection tasks including dynamic obstacles, transparent / reflective objects and material distribution, and compensate for the performance degradation of lidar in bad weather.

[0016] By emitting high-frequency sound waves and receiving echo signals through ultrasonic sensors, and processing them in conjunction with acoustic ranging methods, the system can complete detection tasks including the distance, orientation, and outline of close-range targets, as well as identification tasks including obstacles in narrow spaces, material pile edges, and bucket contact status, thus supplementing the perception blind spots of lidar and vision in close-range operations.

[0017] Furthermore, the specific method for updating the material model parameters in real time based on the measured pressure-displacement data during the stacking process in step S2 is as follows:

[0018] Let the material model parameter vector be ;

[0019] No. The measured pressure data at each moment is Displacement data and its derived features constitute the input vector;

[0020] ,

[0021] In the formula, Let be the feature vector at time k, where each element corresponds to a specific parameter. This represents the vector dimension and the specific number of parameters.

[0022] Assume the material model is linear:

[0023] ,

[0024] in, Representing the Output of the time-matter model prediction; For the first The parameter vector at time t, i.e., the first... The core parameters of the time-matter model;

[0025] Set the initial parameter vector Set the initial covariance matrix Where α is a large positive number, It is the identity matrix;

[0026] Calculate the gain matrix :

[0027] ,

[0028] In the formula, Let be the covariance matrix of the previous time step;

[0029] Update parameter vector :

[0030] ,

[0031] In the formula, This is the parameter vector from the previous time step;

[0032] Update covariance matrix :

[0033] ,

[0034] Acquire pressure sensor data in each sampling period. Displacement-related data Based on the above recursive formula, using and Update material model parameters The updated parameters Substitute the material model into the digital twin model to improve its predictive capabilities for decision-making in the next control cycle.

[0035] Furthermore, the state transition model in S3 is specifically as follows:

[0036] Let the state variable be... ,in For the height of the material stack, For slope deviation, The bucket attitude vector; control variables , For the feeding speed, For hydraulic pressure; establish the state transition equation using a digital twin model:

[0037] ,

[0038] This function describes the state change at the next moment based on the current state and control variables.

[0039] Furthermore, the objective function in S3 is specifically:

[0040] Let the target state variable be... objective function The aim is to minimize the deviation between the state variables and the target state, while limiting the magnitude of change in the control variables, and it is constructed as follows:

[0041] ,

[0042] ,

[0043] In the formula, To predict the number of steps; Is Time prediction The state at any given moment; The target state; Indicates state deviation; To control the change in quantity; For the first The control quantity of the step; For the first The control quantity of the step; where and This is a weight matrix used to adjust the importance of state tracking and control smoothing.

[0044] Furthermore, the method in S4 that combines the hydraulic pressure-flow composite control strategy is specifically as follows:

[0045] Incorporating the pressure-flow control strategy into the quadratic programming objective function:

[0046] ,

[0047] ,

[0048] In the formula, The rate of change of pressure between adjacent time points; The pressure fluctuation weighting coefficient is dynamically adjusted to balance response speed and stability. For the first The pressure of the hydraulic system in step; For the first The hydraulic system pressure is adjusted step by step; at the same time, pressure-flow coordination constraints are added to the constraints.

[0049] Furthermore, the pressure-flow coordination constraint specifically includes:

[0050] ,

[0051] In the formula, from top to bottom, they are pressure limit constraint, flow rate limit constraint, flow rate change constraint, and load balance constraint; Based on the current time Predicting the future The hydraulic system pressure at any given time; These represent the minimum and maximum allowable pressures of the hydraulic system, respectively. In order to be in At any given moment, the flow rate into the hydraulic actuator; This is the flow coefficient of the hydraulic valve; The valve port area gradient of the hydraulic proportional valve; This represents the maximum opening degree of the hydraulic proportional valve. The density of the hydraulic oil; The oil supply pressure of the hydraulic system; The inflow rate at the previous moment, and correspond; To control the time interval of the cycle; The maximum allowable rate of change in flow rate; In order to be in At any given moment, the load force borne by the hydraulic actuator; This refers to the effective working area of ​​the hydraulic actuator.

[0052] Furthermore, the pressure fluctuation weighting coefficient is specifically as follows:

[0053] ,

[0054] ,

[0055] in, This represents the pressure fluctuation weighting coefficient. For pressure error, and α represents the upper and lower bounds of the weights; α and β are adjustment parameters that control the steepness and threshold of the switching. For target pressure; This is the actual pressure.

[0056] As a second aspect of the present invention, an automatic stacking system for an unmanned loader is also provided, comprising:

[0057] The multi-sensor building unit is used to establish a multi-sensor fusion architecture, which enables all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while also configuring device status monitoring sensors.

[0058] The twin model building unit is used to build digital twin models including equipment dynamics, hydraulic systems and material rheological properties; the material rheological parameter library is classified and stored according to material type and working conditions, and the material model parameters are updated in real time based on the measured pressure-displacement data during the stacking process.

[0059] The rolling optimization control unit uses a digital twin model as a predictor, taking factors including stockpile height, slope deviation, and bucket posture as state variables, and factors including pushing speed and hydraulic pressure as control variables, to establish a state transition model and set an objective function. Within each control cycle, it solves for the optimal control sequence for the next 3-5 cycles, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver optimizes the control parameters, and sparse matrix operations reduce the computational load.

[0060] The closed-loop error correction unit is used to receive MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor, and combined with the hydraulic pressure-flow composite control strategy, to realize the execution of control commands and closed-loop error correction through real-time scanning feedback by three-dimensional lidar.

[0061] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor of any step of the above-described unmanned loader automatic stacking method.

[0062] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0063] 1. The unmanned loader automatic material stacking method of the present invention achieves all-round environmental perception by establishing a multi-sensor fusion architecture. It employs LiDAR, vision cameras, millimeter-wave radar, and ultrasonic sensors working in concert to accurately acquire spatial information, material distribution, and obstacle locations within the stacking area. Simultaneously, equipment status monitoring sensors collect real-time parameters of key components such as the hydraulic system and motors. This architecture effectively avoids the limitations of single sensors, improves the accuracy and reliability of environmental perception, provides a comprehensive and accurate data foundation for subsequent control decisions, and ensures the safe and stable operation of the loader under complex working conditions.

[0064] 2. The unmanned loader automatic stacking method of the present invention enhances the system's ability to accurately simulate the stacking process by constructing a digital twin model and updating material parameters in real time. The model integrates equipment dynamics, hydraulic system characteristics, and material rheological properties, while storing material rheological parameters in a database according to type and operating condition. During the stacking process, the material model parameters are dynamically adjusted based on measured pressure-displacement data, enabling the digital twin model to closely match actual operating condition changes. This technology can effectively predict state changes during stacking operations, providing a reliable basis for optimizing control strategies and improving the efficiency and quality of stacking operations.

[0065] 3. The unmanned loader automatic stacking method of the present invention achieves high-precision automation of stacking operations through rolling time-domain optimization combined with closed-loop feedback control. Using a digital twin model as a predictor, state variables such as stacking height and slope deviation, and control variables such as pushing speed and hydraulic pressure are determined. A state transition model is established and an objective function is set. A quadratic programming solver and sparse matrix operations are used to solve for the optimal control sequence for the next 3-5 cycles within each control cycle. Simultaneously, the electro-hydraulic proportional valve group and the steering servo motor collaboratively execute control commands, in conjunction with a hydraulic pressure-flow composite control strategy, and utilize real-time scanning feedback from a three-dimensional lidar to dynamically correct control parameters. This technology ensures that the loader accurately executes the stacking task, effectively reduces operational errors, and achieves stable and efficient automated stacking operations. Attached Figure Description

[0066] Figure 1 This is a flowchart of the automatic material stacking method for an unmanned loader according to an embodiment of the present invention;

[0067] Figure 2 This is a system unit diagram of an embodiment of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0069] Example 1

[0070] Please refer to Figure 1 This embodiment 1 provides an automatic material stacking method for an unmanned loader, including:

[0071] S1. Establish a multi-sensor fusion architecture to achieve all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while configuring device status monitoring sensors.

[0072] S2. Construct a digital twin model that includes equipment dynamics, hydraulic system and material rheological properties; the material rheological parameter library is classified and stored according to material type and working condition, and the material model parameters are updated in real time according to the measured pressure-displacement data during the stacking process.

[0073] S3. Using a digital twin model as a predictor, factors including stockpile height, slope deviation, and bucket posture are used as state variables, and factors including pushing speed and hydraulic pressure are used as control variables. A state transition model is established, and an objective function is set. The optimal control sequence for the next 3-5 cycles is solved within each control cycle, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver is used to optimize the control parameters, and sparse matrix operations are used to reduce the computational load.

[0074] S4. By receiving MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor, and combining the hydraulic pressure-flow composite control strategy, the execution of control commands and the correction of closed-loop errors are realized through real-time scanning feedback by three-dimensional lidar.

[0075] This embodiment 1 further elaborates on the above content.

[0076] (1) Multi-sensor construction

[0077] This embodiment 1 employs an innovative multi-sensor fusion architecture, achieving all-around environmental perception through the collaborative work of LiDAR, visual cameras, millimeter-wave radar, and ultrasonic sensors. Point cloud data is collected by LiDAR for 3D environment modeling and geometric feature extraction, constructing terrain and obstacle outlines in real time.

[0078] By acquiring image information through a visual camera and processing it using deep learning methods, it can complete recognition tasks including color, texture and semantics, as well as classification tasks including personnel, equipment and material types, thus supplementing the shortcomings of LiDAR in material recognition.

[0079] By transmitting electromagnetic waves and receiving reflected signals through millimeter-wave radar, and combining signal processing methods for analysis, it can complete measurement tasks including target distance, speed and azimuth, as well as detection tasks including dynamic obstacles, transparent / reflective objects and material distribution, and compensate for the performance degradation of lidar in bad weather.

[0080] By emitting high-frequency sound waves and receiving echo signals through ultrasonic sensors, and combining them with acoustic ranging methods, the system completes detection tasks including the distance, orientation, and outline of close-range targets, as well as identification tasks including obstacles in narrow spaces, material pile edges, and bucket contact status, supplementing the perception blind spots of lidar and vision in close-range operations.

[0081] In a preferred embodiment, the 905nm wavelength solid-state LiDAR has a 270° horizontal field of view and a 60° vertical field of view, with an angular resolution of 0.1°, achieving a positioning accuracy of ±3cm within a 50-meter range. Four 2-megapixel fisheye cameras form a panoramic vision system, supporting a 190° ultra-wide field of view and a 120dB high dynamic range, achieving a ≥98% accuracy rate in identifying personnel and equipment within a 5-meter range.

[0082] For equipment condition monitoring, the system is equipped with a triaxial MEMS vibration sensor (10kHz sampling) and a piezoresistive hydraulic pressure sensor (100Hz sampling), and uses advanced signal processing methods to achieve real-time monitoring of equipment health status. The hydraulic system life prediction model based on an LSTM neural network has a prediction error controlled within 8%.

[0083] (2) Twin model construction

[0084] The construction process of the digital twin model in this embodiment 1 mainly includes three steps. First, a basic model framework is built, and models are established for the loader's equipment dynamics, hydraulic system operation, and material rheological characteristics. The equipment dynamics model is based on the principle of mechanical motion and describes the motion law of each component of the loader; the hydraulic system model is based on the principle of fluid mechanics and describes the flow and pressure changes of hydraulic oil; the material rheological model simulates the deformation and flow of materials under different stress conditions according to their characteristics.

[0085] Next, a material rheological parameter library is established. Materials are categorized according to type (e.g., ore, sand, etc.) and operating conditions (humidity, temperature, etc.), and corresponding rheological parameters, such as the internal friction angle and viscosity coefficient, are collected and stored. These parameters provide initial data support for the model, enabling it to reflect the characteristics of different materials and operating conditions.

[0086] Finally, the model parameters are updated in real time. During the stockpiling operation, actual operating data such as hydraulic system pressure and loader displacement are continuously collected. By comparing the actual data with the model's predicted data, specific methods are used to adjust the material model parameters to ensure that the digital twin model can accurately reflect the changes in the actual stockpiling process, achieving dynamic matching between the model and actual working conditions.

[0087] In a preferred embodiment, the specific method for updating the material model parameters in real time based on the measured pressure-displacement data during the stockpiling process is as follows:

[0088] Let the material model parameter vector be ;

[0089] No. The measured pressure data at each moment is Displacement data and its derived features, such as velocity and acceleration, constitute the input vector;

[0090] ,

[0091] In the formula, Let be the feature vector at time k, where each element corresponds to a specific parameter. This represents the vector dimension and the specific number of parameters.

[0092] Assume the material model is linear:

[0093] ,

[0094] in, Representing the Output of the time-matter model prediction; For the first The parameter vector at time t, i.e., the first... The core parameters of the time-matter model;

[0095] Set the initial parameter vector Set the initial covariance matrix , where α is a relatively large positive number (e.g. ), It is the identity matrix;

[0096] Calculate the gain matrix :

[0097] ,

[0098] In the formula, Let be the covariance matrix of the previous time step;

[0099] Update parameter vector :

[0100] ,

[0101] In the formula, This is the parameter vector from the previous time step;

[0102] Update covariance matrix :

[0103] ,

[0104] Acquire pressure sensor data in each sampling period. Displacement-related data Based on the above recursive formula, using and Update material model parameters The updated parameters Substitute the material model into the digital twin model to improve its predictive capabilities for decision-making in the next control cycle.

[0105] (3) Rolling optimization control

[0106] The rolling optimization control process uses a digital twin model as its core and achieves precise control of the stockpiling operation through iterative iteration. First, factors such as stockpile height, slope deviation, and bucket posture are set as state variables, while pushing speed and hydraulic pressure are used as control variables. Based on the digital twin model, a state transition relationship is established to predict the state changes of the stockpiling operation in the next 3-5 control cycles.

[0107] Next, an objective function is defined, comprehensively considering the deviation between the state variables and the target value, as well as the costs incurred by changes in the control variables, striving to reduce fluctuations in the control quantity while meeting control requirements. Simultaneously, constraints are clearly defined, incorporating physical limitations such as the hydraulic system pressure limit and the bucket's range of motion to ensure the feasibility of the control scheme in actual operation.

[0108] Subsequently, within each control cycle, a quadratic programming solver is used to optimize the objective function, and sparse matrix operation techniques are combined to reduce the computational load, quickly solving for the optimal control sequence for the next 3-5 cycles. However, only the first control command of this sequence is actually executed, transmitting it to actuators such as the electro-hydraulic proportional valve group and steering servo motor to drive the loader to perform the stacking operation.

[0109] In a preferred embodiment, the state transition model is as follows:

[0110] Let the state variable be... ,in For the height of the material stack, For slope deviation, The bucket attitude vector; control variables , For the feeding speed, For hydraulic pressure; establish the state transition equation using a digital twin model:

[0111] ,

[0112] This function describes the state change at the next moment based on the current state and control variables.

[0113] The objective function is as follows:

[0114] Let the target state variable be... objective function The aim is to minimize the deviation between the state variables and the target state, while limiting the magnitude of change in the control variables, and it is constructed as follows:

[0115] ,

[0116] ,

[0117] In the formula, To predict the number of steps; Is Time prediction The state at any given moment; The target state; Indicates state deviation; To control the change in quantity; For the first The control quantity of the step; For the first The control quantity of the step; where and This is a weight matrix used to adjust the importance of state tracking and control smoothing.

[0118] When optimizing the objective function, the objective function and constraints in rolling optimization control must first be transformed into the standard form of a quadratic programming problem. Control variables such as the pushing speed and hydraulic pressure over the next few control cycles are taken as decision variables to be solved. Optimization objectives such as minimizing state tracking error and minimizing control quantity changes are mapped to quadratic and linear terms in the quadratic programming problem. Simultaneously, practical constraints such as the hydraulic system pressure limit and the bucket movement range are transformed into inequalities and equality constraints.

[0119] Next, analyzing the characteristics of the problem revealed that the correlation matrix of the constructed quadratic programming problem contained a large number of zero elements. In the coefficient matrix corresponding to the objective function, many zero values ​​were present due to the weak correlation between control variables in adjacent control periods; similarly, the constraint matrix, because each constraint is only related to a subset of control variables, also exhibited sparseness. Utilizing this characteristic, a special data storage format was adopted to record only the values ​​and positions of non-zero elements in the matrix, avoiding the storage of a large number of useless zero elements, thereby reducing memory usage and improving data processing efficiency.

[0120] Then, quadratic programming solvers capable of handling sparse matrices are selected. These solvers employ methods specifically designed for sparse matrices, such as sparse Cholesky decomposition, which significantly reduces computational complexity compared to traditional methods. Furthermore, the optimal solution from the previous cycle is used as the initial value for the current cycle, allowing the solution process to converge faster and reducing the number of iterations.

[0121] Ultimately, by using the above methods, the single optimization calculation time is controlled within a short time while ensuring the accuracy of the solution, thus meeting the requirements of real-time control. This enables the method to run efficiently in actual embedded controllers, achieving rapid optimization control of the automatic material stacking process of unmanned loaders.

[0122] (4) Closed-loop error correction

[0123] The closed-loop error correction process is centered on real-time feedback and achieves precise control of the stockpiling operation through multi-stage collaboration. First, a 3D LiDAR scans the stockpiling area at a fixed frequency to acquire 3D point cloud data of the stockpile surface. After preprocessing such as noise reduction and downsampling, key state parameters such as stockpile height and slope are extracted. These measured data are compared with the corresponding parameters predicted by the digital twin model to calculate the state error and its rate of change, such as stockpile height deviation and slope angle deviation.

[0124] Subsequently, the error information is transmitted to the control system. The electro-hydraulic proportional valve group and steering servo motor receive model predictive control (MPC) commands and execute operations in conjunction with a hydraulic pressure-flow composite control strategy, adjusting control variables such as pushing speed and hydraulic pressure to drive the loader to complete the stacking action. During this process, the hydraulic system operates according to pressure-flow constraints, ensuring that the pressure does not exceed the equipment limits and that the flow rate changes smoothly, avoiding equipment damage or operational deviations due to overly aggressive control.

[0125] Finally, the next scanning cycle begins, and the lidar again collects data on the stockpile status, repeating the process of error calculation, control command adjustment, and execution. By continuously cycling through the closed-loop process of "data acquisition - error calculation - command correction - action execution," the control strategy is continuously adjusted based on the difference between the actual stockpile status and the target status, gradually reducing the error and bringing the stockpile operation results closer to the preset target, thus achieving high-precision and stable automated stockpile operation.

[0126] In a preferred embodiment, the method combining the hydraulic pressure-flow composite control strategy specifically includes:

[0127] Incorporating the pressure-flow control strategy into the quadratic programming objective function:

[0128] ,

[0129] ,

[0130] In the formula, The rate of change of pressure between adjacent time points; The pressure fluctuation weighting coefficient is dynamically adjusted to balance response speed and stability. For the first The pressure of the hydraulic system in step; For the first The pressure of the hydraulic system in the step.

[0131] Simultaneously, add pressure-flow coordination constraints to the constraints:

[0132] ,

[0133] In the formula, from top to bottom, they are pressure limit constraint, flow rate limit constraint, flow rate change constraint, and load balance constraint; Based on the current time Predicting the future The hydraulic system pressure at any given time; These represent the minimum and maximum allowable pressures of the hydraulic system, respectively. In order to be in At any given moment, the flow rate into the hydraulic actuator; This is the flow coefficient of the hydraulic valve; The valve port area gradient of the hydraulic proportional valve; This represents the maximum opening degree of the hydraulic proportional valve. The density of the hydraulic oil; The oil supply pressure of the hydraulic system; The inflow rate at the previous moment, and correspond; To control the time interval of the cycle; The maximum allowable rate of change in flow rate; In order to be in At any given moment, the load force borne by the hydraulic actuator; This refers to the effective working area of ​​the hydraulic actuator.

[0134] These parameters collectively constrain the operating boundaries of the hydraulic system, ensuring that pressure and flow do not exceed equipment limits to avoid component damage; limiting the rate of flow change to avoid pressure shocks and bucket vibration; ensuring that the load force does not exceed the hydraulic thrust to ensure that the bucket can push the material; and keeping the hydraulic system within a safe and controllable operating range, making the material stacking process both powerful and stable.

[0135] In a preferred embodiment, the pressure fluctuation weighting coefficient is specifically as follows:

[0136] ,

[0137] ,

[0138] in, This represents the pressure fluctuation weighting coefficient. For pressure error, and α represents the upper and lower bounds of the weights; α and β are adjustment parameters that control the steepness and threshold of the switching. For target pressure; This is the actual pressure.

[0139] Example 2

[0140] Please refer to Figure 2 This embodiment 2 provides an automatic stacking system for an unmanned loader, comprising:

[0141] The multi-sensor building unit is used to establish a multi-sensor fusion architecture, which enables all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while also configuring device status monitoring sensors.

[0142] The twin model building unit is used to build digital twin models including equipment dynamics, hydraulic systems and material rheological properties; the material rheological parameter library is classified and stored according to material type and working conditions, and the material model parameters are updated in real time based on the measured pressure-displacement data during the stacking process.

[0143] The rolling optimization control unit uses a digital twin model as a predictor, taking factors including stockpile height, slope deviation, and bucket posture as state variables, and factors including pushing speed and hydraulic pressure as control variables, to establish a state transition model and set an objective function. Within each control cycle, it solves for the optimal control sequence for the next 3-5 cycles, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver optimizes the control parameters, and sparse matrix operations reduce the computational load.

[0144] The closed-loop error correction unit is used to receive MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor. Combined with the hydraulic pressure-flow composite control strategy, it realizes the execution of control commands and closed-loop error correction through real-time scanning feedback by three-dimensional lidar.

[0145] Example 3

[0146] This embodiment 3 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement any step of an automatic material stacking method for an unmanned loader.

[0147] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0148] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0149] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic material stacking method for an unmanned loader, characterized in that, include: S1. Establish a multi-sensor fusion architecture to achieve all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while configuring device status monitoring sensors. S2. Construct a digital twin model that includes equipment dynamics, hydraulic system and material rheological properties; the material rheological parameter library is classified and stored according to material type and working condition, and the material model parameters are updated in real time according to the measured pressure-displacement data during the stacking process. S3. Using a digital twin model as a predictor, factors including stockpile height, slope deviation, and bucket posture are used as state variables, and factors including pushing speed and hydraulic pressure are used as control variables. A state transition model is established, and an objective function is set. The optimal control sequence for the next 3-5 cycles is solved within each control cycle, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver is used to optimize the control parameters, and sparse matrix operations are used to reduce the computational load. S4. By receiving MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor, and combining the hydraulic pressure-flow composite control strategy, the execution of control commands and the correction of closed-loop errors are realized through real-time scanning feedback by three-dimensional lidar. The construction process of the digital twin model includes three steps: S201: Establish a basic model framework, creating models for loader equipment dynamics, hydraulic system operation, and material rheological properties. The material rheological model is built by simulating the deformation and flow of materials under different stress conditions based on their properties. The hydraulic system model is based on fluid mechanics principles to depict the flow and pressure changes of hydraulic oil. The equipment dynamics model is built based on mechanical motion principles to describe the motion laws of various components of the loader. S202: Establish a material rheological parameter library; S203: Real-time updates of model parameters are performed. During the stacking operation, actual operating data, including hydraulic system pressure and loader displacement, are continuously collected. The material model parameters are adjusted by comparing the actual data with the model prediction data. The specific method for updating the material model parameters in real time based on the measured pressure-displacement data during the stockpiling process in S2 is as follows: Let the material model parameter vector be ; No. The measured pressure data at each moment is Displacement data and its derived features constitute the input vector; In the formula, Let be the feature vector at time k, where each element in the vector corresponds to a specific parameter. This represents the vector dimension and the specific number of parameters; assuming the material model is a linear relationship: in, Representing the Output of the time-matter model prediction; For the first The parameter vector at time t, i.e., the first... The core parameters of the time-bound material model; setting the initial parameter vector. Set the initial covariance matrix ,in A relatively large positive number. It is the identity matrix; Calculate the gain matrix : In the formula, Let be the covariance matrix of the previous time step; Update parameter vector : In the formula, This is the parameter vector from the previous time step; Update covariance matrix : Acquire pressure sensor data in each sampling period. Displacement-related data Based on the above recursive formula, using and Update material model parameters The updated parameters Substitute the material model into the digital twin model to improve its predictive capabilities for decision-making in the next control cycle; The state transition model in S3 is specifically as follows: Let the state variable be... ,in For the height of the material stack, For slope deviation, The bucket attitude vector; control variables , For the pushing speed, For hydraulic pressure; establish the state transition equation using a digital twin model: This function describes the state change at the next moment based on the current state and control variables; The objective function in S3 is specifically: Let the target state variable be... objective function The aim is to minimize the deviation between the state variables and the target state, while limiting the magnitude of change in the control variables, and it is constructed as follows: In the formula, To predict the number of steps; Is Time prediction The state at any given moment; The target state; Indicates state deviation; To control the change in quantity; For the first The control quantity of the step; For the first The control quantity of the step; where and This is a weight matrix used to adjust the importance of state tracking and control smoothing; The specific method of combining hydraulic pressure-flow composite control strategy in S4 is as follows: Incorporating the pressure-flow control strategy into the quadratic programming objective function: In the formula, The rate of change of pressure between adjacent time points; The pressure fluctuation weighting coefficient is dynamically adjusted to balance response speed and stability. For the first The pressure of the hydraulic system in step; For the first The hydraulic system pressure of the step; at the same time, pressure-flow coordination constraints are added to the constraints; The pressure-flow coordination constraint is specifically as follows: In the formula, from top to bottom, they are pressure limit constraint, flow rate limit constraint, flow rate change constraint, and load balance constraint; Based on the current moment Predicting the future The hydraulic system pressure at any given time; These represent the minimum and maximum allowable pressures of the hydraulic system, respectively. In order to be in At any given moment, the flow rate into the hydraulic actuator; This is the flow coefficient of the hydraulic valve; The valve port area gradient of the hydraulic proportional valve; This represents the maximum opening degree of the hydraulic proportional valve. The density of the hydraulic oil; The oil supply pressure of the hydraulic system; The inflow rate at the previous moment, and correspond; To control the time interval of the cycle; The maximum allowable rate of change in flow rate; In order to be in At any given moment, the load force borne by the hydraulic actuator; This refers to the effective working area of ​​the hydraulic actuator.

2. The automatic material stacking method for an unmanned loader according to claim 1, characterized in that, The multi-sensor fusion architecture in S1 is specifically as follows: Point cloud data is collected by LiDAR and used for 3D environment modeling and geometric feature extraction to build terrain and obstacle outlines in real time. By acquiring image information through a visual camera and processing it using deep learning methods, it can complete recognition tasks including color, texture and semantics, as well as classification tasks including personnel, equipment and material types, thus supplementing the shortcomings of LiDAR in material recognition. By transmitting electromagnetic waves and receiving reflected signals through millimeter-wave radar, and combining signal processing methods for analysis, it can complete measurement tasks including target distance, speed and azimuth, as well as detection tasks including dynamic obstacles, transparent / reflective objects and material distribution, and compensate for the performance degradation of lidar in bad weather. By emitting high-frequency sound waves and receiving echo signals through ultrasonic sensors, and processing them in conjunction with acoustic ranging methods, the system can complete detection tasks including the distance, orientation, and outline of close-range targets, as well as identification tasks including obstacles in narrow spaces, material pile edges, and bucket contact status, thus supplementing the perception blind spots of lidar and vision in close-range operations.

3. The automatic material stacking method for an unmanned loader according to claim 1, characterized in that, The pressure fluctuation weighting coefficient is specifically as follows: in, This represents the pressure fluctuation weighting coefficient. For pressure error, and α represents the upper and lower bounds of the weights; α and β are adjustment parameters that control the steepness and threshold of the switching. For target pressure; This is the actual pressure.

4. An automatic material stacking system for an unmanned loader, characterized in that, include: The multi-sensor building unit is used to establish a multi-sensor fusion architecture, which enables all-round environmental perception through the collaborative work of devices including LiDAR, vision camera, millimeter-wave radar and ultrasonic sensor, while also configuring device status monitoring sensors. The twin model building unit is used to build digital twin models including equipment dynamics, hydraulic systems and material rheological properties; the material rheological parameter library is classified and stored according to material type and working conditions, and the material model parameters are updated in real time based on the measured pressure-displacement data during the stacking process. The rolling optimization control unit uses a digital twin model as a predictor, taking factors including stockpile height, slope deviation, and bucket posture as state variables, and factors including pushing speed and hydraulic pressure as control variables, to establish a state transition model and set an objective function. Within each control cycle, it solves for the optimal control sequence for the next 3-5 cycles, with constraints including hydraulic system pressure limits and bucket movement range. A quadratic programming solver optimizes the control parameters, and sparse matrix operations reduce the computational load. The closed-loop error correction unit is used to receive MPC commands in coordination with the electro-hydraulic proportional valve group and the steering servo motor, and combined with the hydraulic pressure-flow composite control strategy, it realizes the execution of control commands and closed-loop error correction through real-time scanning feedback by three-dimensional lidar. The construction process of the digital twin model includes three steps: S201: Establish a basic model framework, creating models for loader equipment dynamics, hydraulic system operation, and material rheological properties. The material rheological model is built by simulating the deformation and flow of materials under different stress conditions based on their properties. The hydraulic system model is based on fluid mechanics principles to depict the flow and pressure changes of hydraulic oil. The equipment dynamics model is built based on mechanical motion principles to describe the motion laws of various components of the loader. S202: Establish a material rheological parameter library; S203: Real-time updates of model parameters are performed. During the stacking operation, actual operating data, including hydraulic system pressure and loader displacement, are continuously collected. The material model parameters are adjusted by comparing the actual data with the model prediction data. The specific method for updating the material model parameters in real time based on the measured pressure-displacement data during the stacking process in the twin model construction unit is as follows: Let the material model parameter vector be ; No. The measured pressure data at each moment is Displacement data and its derived features constitute the input vector; In the formula, Let be the feature vector at time k, where each element in the vector corresponds to a specific parameter. This represents the vector dimension and the specific number of parameters; assuming the material model is a linear relationship: in, Representing the Output of the time-matter model prediction; For the first The parameter vector at time t, i.e., the first... The core parameters of the time-bound material model; setting the initial parameter vector. Set the initial covariance matrix ,in A relatively large positive number. It is the identity matrix; Calculate the gain matrix : In the formula, Let be the covariance matrix of the previous time step; Update parameter vector : In the formula, This is the parameter vector from the previous time step; Update covariance matrix : Acquire pressure sensor data in each sampling period. Displacement-related data Based on the above recursive formula, using and Update material model parameters The updated parameters Substitute the material model into the digital twin model to improve its predictive capabilities for decision-making in the next control cycle; The state transition model in the rolling optimization control unit is specifically as follows: Let the state variable be... ,in For the height of the material stack, For slope deviation, The bucket attitude vector; control variables , For the pushing speed, For hydraulic pressure; establish the state transition equation using a digital twin model: This function describes the state change at the next moment based on the current state and control variables; The objective function in the rolling optimization control unit is specifically: Let the target state variable be... objective function The aim is to minimize the deviation between the state variables and the target state, while limiting the magnitude of change in the control variables, and it is constructed as follows: In the formula, To predict the number of steps; Is Time prediction The state at any given moment; The target state; Indicates state deviation; To control the change in quantity; For the first The control quantity of the step; For the first The control quantity of the step; where and This is a weight matrix used to adjust the importance of state tracking and control smoothing; The specific method of combining the hydraulic pressure-flow composite control strategy in the closed-loop error correction unit is as follows: Incorporating the pressure-flow control strategy into the quadratic programming objective function: In the formula, The rate of change of pressure between adjacent time points; The pressure fluctuation weighting coefficient is dynamically adjusted to balance response speed and stability. For the first The pressure of the hydraulic system in step; For the first The hydraulic system pressure of the step; at the same time, pressure-flow coordination constraints are added to the constraints; The pressure-flow coordination constraint is specifically as follows: In the formula, from top to bottom, they are pressure limit constraint, flow rate limit constraint, flow rate change constraint, and load balance constraint; Based on the current moment Predicting the future The hydraulic system pressure at any given time; These represent the minimum and maximum allowable pressures of the hydraulic system, respectively. In order to be in At any given moment, the flow rate into the hydraulic actuator; This is the flow coefficient of the hydraulic valve; The valve port area gradient of the hydraulic proportional valve; This represents the maximum opening degree of the hydraulic proportional valve. The density of the hydraulic oil; The oil supply pressure of the hydraulic system; The inflow rate at the previous moment, and correspond; To control the time interval of the cycle; The maximum allowable rate of change in flow rate; In order to be in At any given moment, the load force borne by the hydraulic actuator; This refers to the effective working area of ​​the hydraulic actuator.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor according to any one of claims 1-3, which describes an automatic stacking method for an unmanned loader.

Images