Intelligent lubricant spraying device and spraying control method for rubber-tired road roller

Abstract

The application relates to the technical field of road engineering machinery, in particular to a separating agent spraying device for a rubber-tired road roller and a spraying control method. The device comprises an information acquisition unit, a spraying device and a controller. The controller dynamically determines a target spraying amount and controls the spraying device to execute based on the rubber wheel surface adhesion state information acquired by the information acquisition unit in real time, forming a real-time feedback control closed loop. The controller has a built-in mapping relationship model, and the model is iteratively updated after each spraying according to the state information collected again as effect data, realizing self-learning and continuous optimization of the system. The scheme realizes the leap of separating agent spraying from a fixed mode to dynamic and accurate regulation and control through the construction of an intelligent closed loop of "perception-decision-execution-optimization", solves the problem of material waste; at the same time, by introducing the model online updating mechanism, the system has self-adaptive ability and can continuously optimize the control strategy according to the working conditions, realizing long-term optimal control performance and material utilization.

Description

Technical Field

[0001] This invention relates to the field of road construction machinery technology, and in particular to a release agent spraying device and spraying control method for rubber-tired rollers. Background Technology

[0002] In asphalt pavement construction, using rubber-tired rollers to compact high-temperature asphalt mixtures is a crucial step in ensuring pavement density and final quality. However, high-temperature asphalt easily adheres to the surface of the roller's tires. This "tire adhesion" phenomenon not only damages the smoothness of the newly paved pavement but can also lead to construction interruptions due to the need to clean the adhered material, affecting project efficiency. Therefore, how to effectively and economically prevent asphalt from adhering to the tires has been a long-standing technical need in this field. To meet this need, the common practice in the prior art is to equip rubber-tired rollers with a release agent spraying device. For example, some solutions provide an automated spraying device that includes a reservoir and several nozzles, capable of spraying a release agent onto the tire surface during compaction to form a release film, thereby preventing direct contact between asphalt and the tires. These devices typically employ preset timed or continuous spraying modes, or the spraying system can be manually started or stopped by the operator based on their experience.

[0003] However, the aforementioned existing technical solutions have significant limitations in practical applications. First, their spraying control logic is preset and open-loop, and the execution of the spraying action is not directly related to the real-time asphalt adhesion status on the rubber tire surface. Second, this control method, decoupled from actual working conditions, makes it difficult to adaptively adjust the spraying volume according to the dynamic changes in adhesion risk. When the adhesion risk is low, a fixed spraying pattern will result in a significant waste of release agent; while when changes in working conditions cause a sudden increase in adhesion risk, the existing spraying volume may not be sufficient to provide effective protection, ultimately requiring manual intervention to prevent tire adhesion.

[0004] Therefore, existing spraying technologies cannot achieve on-demand and precise application of release agents, and are insufficient in terms of resource conservation and refined management of the construction process. Summary of the Invention

[0005] In order to solve the technical problems mentioned in the background, this application provides an intelligent spraying device for a rubber-tired roller and also relates to an intelligent spraying control method for a rubber-tired roller.

[0006] This application provides an intelligent release agent spraying device for rubber-tired rollers, comprising: an information acquisition unit for acquiring real-time adhesion status information of the rubber-tired roller surface; a spraying device for spraying release agent onto the rubber-tired roller surface; and a controller connected to the information acquisition unit and the spraying device respectively, wherein the controller is used to: determine the target spraying amount of the release agent based on the adhesion status information acquired from the information acquisition unit; and control the spraying device to spray at the target spraying amount, thereby achieving dynamic control of the release agent dosage.

[0007] Optionally, the information acquisition unit includes: a visible light camera for acquiring visible light images of the rubber wheel surface; and a thermal infrared camera for simultaneously acquiring thermal infrared images of the rubber wheel surface; wherein the controller is further configured to fuse the visible light image and the thermal infrared image to generate data characterizing the bonding state information.

[0008] Optionally, the device further includes a temperature sensor for acquiring temperature data of the rubber wheel surface; wherein, when determining the target spray volume, the controller combines the adhesion state information and the temperature data.

[0009] Optionally, the controller has a built-in mapping model, and the controller determines the target spraying amount by taking the adhesion state information and / or the temperature data as input to the model and taking the output of the model as the target spraying amount.

[0010] Optionally, the controller is further configured to: after the spraying device has sprayed, acquire adhesion state information again from the information acquisition unit as spraying effect data; and update the mapping relationship model based on the spraying effect data.

[0011] This application also provides a method for intelligent spraying control of release agent for rubber-tired rollers, comprising the following steps: acquiring adhesion state information of the rubber-tired roller surface in real time; determining the target spraying amount of release agent for preventing asphalt adhesion based on the adhesion state information; and controlling the spraying device to spray the release agent onto the rubber-tired roller surface at the target spraying amount, so as to achieve dynamic control of the release agent dosage.

[0012] Optionally, the step of acquiring the bonding state information of the rubber-tired roller surface in real time includes: acquiring a visible light image of the rubber-tired roller surface using a visible light camera; simultaneously acquiring a thermal infrared image of the rubber-tired roller surface using a thermal infrared camera; and fusing the visible light image and the thermal infrared image to generate data characterizing the bonding state information.

[0013] Optionally, the method further includes: acquiring temperature data of the rubber wheel surface; wherein, the step of determining the target spray amount of the release agent for preventing asphalt adhesion based on the adhesion state information is to combine the adhesion state information and the temperature data to jointly determine the target spray amount.

[0014] Optionally, the step of combining the adhesion state information and the temperature data to jointly determine the target spraying amount includes: inputting the adhesion state information and / or the temperature data into a pre-trained mapping relationship model; and using the mapping relationship model to output the target spraying amount.

[0015] Optionally, the method further includes: after spraying the release agent, acquiring the adhesion state information of the rubber wheel surface again as spraying effect data; and using the spraying effect data to update the mapping relationship model to continuously optimize the process of determining the target spraying amount.

[0016] The advantages of this application compared to existing technologies are: by setting up a closed-loop connection between the information acquisition unit, controller, and spraying device, a real-time dynamic control loop of "perception-decision-execution" is constructed for the first time in the field of release agent spraying for rubber-tired rollers. This fundamentally changes the traditional spraying mode that relies on manual experience or fixed procedures, enabling the spraying volume to be dynamically adjusted based on the real-time surface state of the rubber tire (adhesion state information). Thus, while ensuring the anti-sticking effect, it provides a basic solution to the core technical problem of the inability to accurately control the amount of release agent, which easily leads to waste.

[0017] This invention captures texture and color features using visible light images and temperature distribution features using thermal infrared images, and fuses this information to significantly improve the accuracy and environmental adaptability of identifying the adhesion state of asphalt on rubber tire surfaces. Specifically, it effectively overcomes the shortcomings of a single visible light camera, which suffers a sharp drop in recognition rate or even fails when there is insufficient light, shadows, or when the asphalt color is similar to the tire color. This provides more reliable and robust data input for subsequent control decisions and is a key prerequisite for achieving precise perception and intelligent control.

[0018] This invention introduces a temperature sensor to acquire surface temperature data of the rubber tire, enabling the controller's decision-making process to comprehensively consider two key influencing factors: "adhesion state" and "temperature." Since the viscosity of asphalt is extremely sensitive to temperature, temperature data provides the controller with crucial operating context information. Combining these two factors to determine the target spray volume makes the control strategy more closely aligned with physical reality. It allows for more precise differentiation between different scenarios, such as pre-spraying at high temperatures due to easy adhesion and spraying less at low temperatures due to low viscosity, thereby further improving the scientific rigor and accuracy of spray volume control and avoiding misjudgments based on single image information.

[0019] This invention determines the specific implementation method of the target spraying volume through a built-in mapping relationship model. Its technical advantage lies in encapsulating the complex multi-parameter decision-making process (bonding area, temperature, etc.) within a computable and executable mathematical model (such as a function or neural network). This method upgrades the controller's decision-making from a simple "if-then" rule to intelligent calculation based on multiple feature inputs, enabling the handling of more complex nonlinear relationships and achieving refined, adaptive matching between spraying volume and factors such as bonding state and temperature. This significantly improves the automation level and accuracy ceiling of the control.

[0020] This invention upgrades open-loop or fixed-model intelligent control to closed-loop intelligent control with self-learning and self-optimization capabilities by adding a model update mechanism based on spraying effect feedback. The system continuously trains and optimizes the mapping relationship model using real effect data by comparing the "actual effect after spraying" with the "expected effect." This allows the system to adapt to changes in different asphalt mixture formulations and environmental conditions, continuously self-correcting and improving during use, making the spraying volume control strategy increasingly precise and efficient, truly achieving long-term optimal operation of the intelligent system.

[0021] This invention employs specific steps for generating adhesion state information through dual-spectral image fusion. In the information acquisition stage, multi-source data fusion enhances the reliability and accuracy of state perception, laying a solid data foundation for subsequent decision-making steps. This is a key technical step that ensures the effectiveness of the entire intelligent control process from a methodological perspective.

[0022] This invention incorporates temperature data as a common decision-making basis in the decision-making process. When determining the spraying volume, it considers not only "how much sticky" (adhesion state) but also "why it's so sticky" (temperature influence), making the decision-making logic more complete and scientific. This methodologically ensures that the control strategy can dynamically adapt to changes in the construction environment temperature, improving the applicability of the entire method under different working conditions and the consistency of control effects.

[0023] This invention clarifies the methodological steps for achieving intelligent decision-making through a specific tool called a "mapping relationship model," concretizing the core intelligent step of "determining usage based on information" into an executable and reproducible model calculation process. This provides a clear path for the industrialization and software implementation of the method, enabling high-precision adaptive control to be stably achieved through algorithms, rather than relying solely on fuzzy empirical judgments.

[0024] This invention incorporates a feedback optimization step of "effect evaluation - model update" into the method flow, further forming an intelligent control method with continuous evolution capabilities. This method not only achieves precise control in a single operation but also iteratively optimizes its control model (mapping relationship) through continuously accumulated practical data (spraying effect) over long-term, multiple operations. This allows the overall control performance of the method to continuously improve over time, exhibiting long-term adaptability and superiority. This represents the highest level of "intelligence." Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the intelligent spraying device for release agent used in rubber-tired rollers in this application.

[0026] Figure 2 This is a control flowchart of the intelligent spraying device for release agent used in rubber-tired rollers in this application.

[0027] 1. Intelligent control module; 2. Control switch; 3. Alarm light; 4. Switch; 5. Frame; 6. Intelligent control pump; 7. Spray nozzle; 8. Wheel; 9. Thermometer; 10. Visible light camera; 11. Thermal infrared camera; 12. Liquid outlet pipe; 13. Pipe; 14. Valve; 15. Sensor; 16. Liquid level gauge; 17. Thermometer; 18. Liquid inlet; 19. Spraying execution module; 20. Pressure pump; 21. Thermal infrared camera; 22. Visible light camera; 23. Thermometer; 24. Intelligent control pump; 25. Spray nozzle. Detailed Implementation

[0028] The following are examples of specific implementation processes provided to illustrate the technical solutions to be protected in this application. However, this application may also be implemented in other ways different from those described herein. Those skilled in the art can implement this application by different technical means under the guidance of the concept of this application. Therefore, this application is not limited to the specific embodiments below.

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] This invention provides an intelligent spraying device and method for a release agent, the main implementing body of which can be an intelligent control system integrated on a rubber-tired roller. The core of this system can be one or more controllers. The controller can be a hardware unit with data processing and control capabilities, such as a microcontroller (MCU), programmable logic controller (PLC), central processing unit (CPU) in an embedded system, digital signal processor (DSP), or application-specific integrated circuit (ASIC). In the following embodiments, the controller is used as the executing entity for description.

[0031] Example 1: In existing asphalt pavement compaction construction, rubber-tired rollers generally use a timed or manually operated method for spraying release agent. This method cannot be adjusted in real time according to the actual adhesion of the asphalt on the tire surface, often resulting in insufficient release agent spraying and tire sticking, affecting construction quality and efficiency; or excessive spraying, which not only causes serious waste of release agent material, but may also adversely affect the performance of the asphalt mixture. To solve this technical problem, embodiments of the present invention provide an intelligent device capable of on-demand and precise spraying.

[0032] Please see Figure 1 , Figure 1 This embodiment provides a structural schematic diagram of an intelligent release agent spraying device for a rubber-tired roller. The device, which can be installed as a whole on the frame 5 of the rubber-tired roller, includes: an information acquisition unit, a spraying device, and a controller. The main function of the information acquisition unit is to sense and acquire the adhesion status information of the rubber tires 8 on the rubber-tired roller in real time.

[0033] like Figure 1As shown, this invention uses a vehicle frame 5 as the mounting base and integrates three major units: monitoring, control, and execution. Image acquisition includes visible light cameras 10 and 22 and thermal infrared cameras 11 and 21, respectively mounted on the inner side of the vehicle frame 5 and aimed at the front and rear wheels 8. The visible light cameras capture color and texture information of the wheel surface; the thermal infrared cameras acquire temperature distribution information of the wheel surface. The two cameras achieve synchronous acquisition through trigger signals. Temperature monitoring includes a first thermometer 9 for detecting road surface temperature, a second thermometer 17 for detecting the temperature of the front vehicle's tire surface, and a third thermometer 23 for detecting the temperature of the rear vehicle's tire surface. All thermometers are preferably infrared thermometers, operating in a non-contact manner. The liquid condition monitoring module integrates a level gauge 16 and a sensor 15 within the liquid storage tank. The sensor 15 is preferably a viscosity sensor for real-time monitoring of the physical state of the release agent; the level gauge 16 monitors the remaining amount of release agent in the liquid storage tank. The intelligent control module serves as the system's brain, i.e., the intelligent control unit or intelligent control module 1. It integrates a high-performance image processor and data processor, and is equipped with a control switch 2, an alarm light 3, an operation switch 4, and a display interface. Spraying execution module: Control commands generated by the system control module 1 are sent to the spraying execution module 19. The intelligent control pumps 6 and 24 of this module respond immediately and precisely adjust their flow output. The adjusted release agent, driven by the stable pressure provided by the pressure pump 20, is delivered through pipeline 13 to the spray nozzles 7 and 25, and sprayed in atomized form onto the surface of the wheel 8. The nozzle configuration ensures uniform coverage of the release agent.

[0034] In specific implementations, the information acquisition unit can be configured in various ways. For example, it can be one or more industrial cameras mounted on the frame 5, facing the contact surface of the rubber wheel 8. The industrial camera continuously or at a preset frequency (e.g., 10 times per second) captures images of the surface of the rubber wheel 8, and uses the captured image data as initial adhesion state information. Alternatively, the information acquisition unit can be an array of photoelectric sensors arranged along the width of the rubber wheel 8, which determines whether asphalt particles are attached by detecting changes in the reflectivity of the rubber wheel surface. In this embodiment, the information acquisition unit is preferably an image acquisition device mounted on the frame 5 and aimed at the rubber wheel 8, such as... Figure 1 The visible light cameras 10 and 22 are shown. The adhesion state information is quantitative or non-quantitative information used to characterize the degree of asphalt adhesion on the surface of the rubber wheel 8. This information can be raw data directly output by the information acquisition unit, such as an unprocessed digital image. It can also be data after preliminary processing, for example, obtained by binarizing the acquired image, representing the number of pixels with asphalt adhesion or the percentage of the total monitored area. The spraying device is a mechanism for receiving instructions from the controller and executing specific spraying actions. (Refer to...) Figure 1A typical spraying device may include a reservoir for storing a release agent, a pressure pump 20 that provides power for delivering the release agent, pipes 12 and 13 connecting the various components, and one or more spray nozzles 7, 25 that directly atomize and spray the release agent onto the surface of the rubber wheel 8. The pressure pump 20 or the valve 14 on the pipe 13 in this spraying device may be designed to have its operation controlled by an electrical signal; for example, a DC pump whose speed can be changed by adjusting the input voltage, or a solenoid valve whose opening can be controlled by a pulse width modulation (PWM) signal. The controller, such as... Figure 1 As shown in Figure 1, the intelligent control module serves as the core of the entire device's computation and control. It is connected to the information acquisition unit and the spraying device via signal lines. The controller can be any of the aforementioned hardware units, and its internal program logic implements the method of this invention.

[0035] In a complete workflow of this embodiment, combined with Figure 2 To understand the control workflow diagram shown, firstly, the controller command information acquisition unit acquires real-time information about the adhesion status of the rubber wheel 8 surface. For example, the controller triggers the visible light camera 10 to capture a real-time image of the rubber wheel 8 surface.

[0036] Next, the controller processes the acquired adhesion status information to determine the target spray amount of release agent to prevent asphalt adhesion under the current operating conditions. There are various ways to determine this target spray amount.

[0037] In a simpler implementation, the controller can make a judgment based on a preset threshold. For example, the controller analyzes the received image and calculates the area S covered by black asphalt particles. If S is less than the preset area threshold S_min, the controller determines that the target spraying amount is zero; if S is greater than S_min, the controller determines that the target spraying amount is a fixed non-zero value Q_fixed.

[0038] In a more refined implementation, the controller can calculate the target spray volume based on a preset mapping relationship. This mapping relationship can be a simple linear function, such as `Q = kA + Q_min`, where Q is the target spray volume, A is the calculated bond area, k is a proportionality coefficient, and Q_min is the minimum spray volume required to maintain basic wetting of the rubber wheel surface. The controller substitutes the calculated bond area A into this function to obtain a target spray volume Q that precisely matches the current bond level. Alternatively, this mapping relationship can be a lookup table stored in the controller's memory, recording the optimal spray volume corresponding to different bond area ranges.

[0039] Finally, the controller controls the spraying device to spray according to the determined target spray volume. For example, if the controller calculates a target spray volume of Q, it converts this into a specific electrical signal (e.g., a specific voltage V for a duration of T) and sends it to the intelligent control pump 6 or pressure pump 20 in the spraying device. Upon receiving the signal, the pump operates at a specific power for a corresponding time, thereby precisely pumping out a total amount of release agent Q, which is then evenly sprayed onto the surface of the rubber roller 8 through the spray nozzle 7, thus achieving dynamic control of the release agent dosage.

[0040] Through the above technical solution, the intelligent release agent spraying device provided in this embodiment of the invention constructs a closed-loop control system of perception-decision-execution. It can monitor the actual adhesion status of the rubber tires in real time, and based on this status, calculate and execute the spraying operation accurately and on demand, changing the extensive method of relying on manual experience or using fixed spraying patterns in existing technologies. This not only effectively prevents tire adhesion problems caused by insufficient spraying, ensuring the continuity of construction and the quality of road surface formation, but also significantly avoids the waste of release agent caused by excessive spraying, reducing construction costs and achieving efficient resource utilization.

[0041] Example 2: In Example 1, if only a single type of information acquisition unit is used, such as a visible light camera, in complex actual construction environments, such as insufficient lighting at night, strong shadow interference during the day, or when the asphalt material used is similar in color to the rubber of the tires, inaccurate identification or even missed detection of the asphalt bonding area may occur, thus affecting the accuracy of subsequent spraying decisions. This example makes specific improvements to the information acquisition unit. Please refer again. Figure 1In this embodiment, the information acquisition unit specifically includes visible light cameras 10 and 22 and thermal infrared cameras 11 and 21. The visible light cameras, such as high-resolution CMOS or CCD industrial cameras, primarily function to acquire color and texture information of the rubber wheel 8 surface. In the case of asphalt adhesion, the visible light cameras can capture the unique textures and contours formed on the rubber wheel surface after asphalt particles adhere. The thermal infrared cameras are devices capable of detecting the infrared radiation of an object and converting it into a temperature image. Because the temperature of freshly laid asphalt mixture is much higher than the surface temperature of the rubber wheel (e.g., the asphalt temperature may be above 120°C, while the rubber wheel surface is 60-80°C), this significant temperature difference causes the asphalt adhesion points to appear as high-brightness areas in the thermal infrared image, forming a strong contrast with the background rubber wheel surface. In this embodiment, the visible light cameras 10 and 22 and the thermal infrared cameras 11 and 21 can be mounted side-by-side on the frame 5 to ensure that their fields of view can collectively cover the key monitoring area of ​​the rubber wheel 8. To ensure that the two images accurately correspond to the physical state at the same moment, the controller sends a synchronized trigger signal to both, causing them to acquire images at exactly the same time. After acquiring the two images, a key task of the controller is to fuse these two complementary pieces of information from different sources to generate data that more accurately represents the adhesion state.

[0042] The fusion process may include the following steps: First, the controller may preprocess the two raw images. For the visible light image acquired by the visible light camera 10, as described in the technical disclosure, a contrast-limited adaptive histogram equalization (CLAHE) algorithm can be used to enhance the local contrast of the image and highlight the texture details of the asphalt particles. For the thermal infrared image acquired by the thermal infrared camera 11, temperature calibration and non-uniformity correction are performed to obtain an accurate temperature distribution matrix and eliminate the influence of sensor noise.

[0043] Next, the controller performs a fusion operation. There are several fusion methods: one is pixel-level fusion. For example, the controller can register two pre-processed images to ensure a one-to-one correspondence between pixels. Then, for each pixel, its brightness value in the visible light image and its temperature value in the thermal infrared image are weighted and averaged according to preset weights, thereby generating a new fused image containing both information modalities. Another, more efficient method is feature-level or decision-level fusion. For example, the controller can first quickly and accurately identify all potential high-temperature areas (i.e., areas that might be asphalt) in the thermal infrared image using a simple temperature threshold, forming a candidate region mask. Then, the controller further utilizes image processing algorithms such as texture analysis and edge detection only within the areas corresponding to this mask in the visible light image to confirm whether the area is indeed asphalt particles, thereby eliminating other possible interfering heat sources.

[0044] Through the above fusion process, the data finally generated by the controller (such as a binarized map of the confirmed asphalt bonding area, or a list containing the coordinates and areas of all bonding points) can reflect the true bonding state on the rubber wheel 8 with extreme reliability and accuracy.

[0045] The technical solution of this embodiment overcomes the recognition defects of single visible light imaging under adverse conditions such as low illumination, shadows, or low color contrast by applying visible light and thermal infrared dual-spectrum image fusion technology to the perception of adhesion state. Utilizing the significant differences in thermophysical properties between asphalt and rubber tires, the accuracy and robustness of adhesion recognition are greatly improved. This provides a high-quality, high-reliability data foundation for the subsequent controller to make precise decisions on the target spraying volume, thereby significantly improving the performance and environmental adaptability of the entire intelligent spraying device.

[0046] Preferably, in one of the preferred technical solutions of this embodiment, a temperature sensor is further added to the original device. While solutions that only consider bonding state information (such as bonding area) can achieve on-demand spraying, their decision-making dimensions are relatively singular. In actual asphalt compaction operations, the viscosity of the asphalt mixture is closely related to its temperature: the higher the temperature, the stronger the fluidity and viscosity of the asphalt, and the greater the risk of roller sticking. Therefore, for the same bonding area, the urgency of treatment and the required isolation dosage should differ at different temperatures. Ignoring temperature factors may result in insufficient spraying at high temperatures or redundant spraying at low temperatures. In this embodiment, multiple temperature sensors can be configured to obtain more comprehensive temperature information. For example, such as... Figure 1As shown, the device may include a first thermometer 9 for detecting road surface temperature, a second thermometer 17 for detecting the tire surface temperature of the preceding vehicle, and a third thermometer 23 for detecting the tire surface temperature of the following vehicle. These temperature sensors are preferably non-contact infrared thermometers, which are installed at appropriate locations on the vehicle frame 5 and are capable of measuring the surface temperature of the corresponding target in real time without interference, and transmitting the collected temperature data to the controller (intelligent control module 1).

[0047] The core improvement of this embodiment lies in the fact that, when determining the target spray volume, the controller no longer relies solely on adhesion state information, but also incorporates the acquired temperature data as a crucial decision input, combining these factors for determination. This combination can be implemented in several specific ways: One implementation method is rule-based decision logic. The controller has a pre-set two-dimensional decision rule matrix. For example, as mentioned in the technical disclosure, the controller can divide the bonding state (e.g., by calculating the bonding area) into three levels: "low value zone," "medium value zone," and "high value zone," and simultaneously divide the acquired rubber wheel surface temperature into three levels: "low temperature," "medium temperature," and "high temperature." Once the controller obtains the real-time bonding area and temperature, it searches for the corresponding target spray volume in the two-dimensional matrix based on their respective levels. For instance, the combination of a "high value zone" area and "high temperature" will trigger the highest level of spray volume, while the same "high value zone" area, combined with "low temperature," may only trigger a medium level of spray volume. In this way, temperature data acts as a regulating factor, dynamically adjusting the basic spraying strategy based on the bonding area.

[0048] Another approach is to use a multivariable function for calculation. The controller can use a function that includes two variables, area and temperature, to calculate the target spray volume, for example, `Q = k(T)`. A + Q_min. Here, A is the bonding area, Q_min is the base spray volume, and the proportionality coefficient `k(T)` is no longer a constant but a function of temperature T. This function `k(T)` can be designed to increase with increasing temperature T. Thus, when the controller receives the bonding area A and temperature T, it first calculates the risk coefficient `k(T)` matching the current temperature, and then calculates the final target spray volume Q. This allows the spray volume to change smoothly and accurately with variations in temperature and area.

[0049] Therefore, in the workflow of this embodiment, the controller acquires real-time temperature data from the temperature sensor while triggering the information acquisition unit. Subsequently, the controller takes the adhesion status information (such as adhesion area) and temperature data as input, and performs comprehensive calculations using the aforementioned rule matrix or multivariate function to finally determine a target spraying amount that considers both the adhesion range and the adhesion risk.

[0050] The technical solution in this embodiment adds a temperature sensing dimension to the device, expanding the controller's decision-making basis from one-dimensional to multi-dimensional. This allows the device to more accurately assess the actual risk of wheel adhesion, rather than simply identifying the area of ​​adhesion that has already occurred. This solution significantly improves the intelligence and accuracy of spraying decisions, enabling the amount of release agent to better adapt to temperature changes in different seasons, asphalt grades, and construction stages. Thus, while ensuring the anti-adhesion effect, it further optimizes and saves on the amount of release agent used.

[0051] Preferably, in one of the preferred technical solutions of this embodiment, in order to achieve a more advanced and flexible decision-making mechanism, the internal logic of the controller (intelligent control module 1) has been deepened. The core of this is that the controller has a pre-trained mapping relationship model built in, and uses this model to determine the target spraying amount. The controller can determine the target spraying amount based on preset rules or relatively simple functions. However, such hard-coded logic may struggle to accurately capture the complex, non-linear intrinsic relationship between the bonding state, temperature, and optimal spraying amount. Its adaptability and accuracy may be limited when facing varying asphalt materials, environmental humidity, and equipment aging. The mapping relationship model described in this solution is a mathematical or computational structure that can map a set of input data to a set of output data. Unlike fixed rules or formulas, this model is typically learned through a data-driven approach and can express more complex functional relationships. In this embodiment, the model can be an artificial intelligence model, such as a feed-forward neural network, a support vector regression model, or a gradient boosting decision tree.

[0052] Before the device is put into use, the mapping model needs to undergo a pre-training process. This process aims to teach the model the mapping relationship from operating parameters (input) to the optimal spraying rate (output). This training process may include the following steps: The first step is data acquisition. In a laboratory environment or at an actual construction site, a large number of training data samples are systematically collected. Each sample is a data pair, including the input working condition data and the corresponding, validated, optimal target spraying volume. The input working condition data is the bonding state information (e.g., asphalt bonding area) and / or temperature data. To accurately obtain training data, as mentioned in the technical disclosure, tracers such as fluorescent powder can be sprinkled onto the rubber tires to facilitate more accurate segmentation and calculation of the asphalt bonding area in image processing. The corresponding optimal spraying volume can be manually set by an experienced operator or determined through multiple trials (e.g., starting with a small spraying volume under specific working conditions and gradually increasing it until the amount at which the sticking phenomenon just disappears is observed). By collecting hundreds or even thousands of such data pairs, a rich dataset is formed.

[0053] The second step is model building and training. Select and build a suitable model structure. Taking a feedforward neural network as an example, it can include: an input layer with the number of neurons corresponding to the dimension of the input data (e.g., two input neurons if the input is adhesion area and temperature); one or more hidden layers, each containing several neurons, using activation functions such as ReLU (Rectified Linear Unit) to introduce non-linearity; and an output layer containing one neuron to output the scalar value of the target spray amount.

[0054] The collected dataset is then fed into the neural network. Using algorithms such as backpropagation and gradient descent optimizers (e.g., the Adam optimizer), the connection weights between layers are iteratively adjusted to minimize the error (e.g., mean squared error) between the model's predicted spraying amount and the optimal spraying amount marked in the dataset. When the error converges to a sufficiently small range, training is complete, and the weight parameters in the model are fixed, forming a "pre-trained" mapping model.

[0055] In actual operation, the pre-trained mapping model built into the controller begins to function. The controller first acquires real-time bonding state information (e.g., bonding area A) from the information acquisition unit and temperature data T from the temperature sensor. Then, the controller uses this data (A and / or T) as input to the model. This input data undergoes forward propagation calculations within the model's internal network layers. Finally, the model's output layer provides a numerical value, which the controller directly uses as the model's output, representing the target spraying volume under the current operating conditions.

[0056] The technical solution in this embodiment integrates a pre-trained mapping model into the controller, replacing simple rules or functions. This enables the device's decision-making core to learn complex patterns from large amounts of data. The model can more accurately fit the nonlinear relationship between multiple factors such as adhesion state and temperature and the optimal spraying rate, making the determination of the spraying rate more scientific and precise, and significantly improving the device's adaptability and spraying effect under various complex and variable operating conditions.

[0057] Preferably, in one of the preferred technical solutions of this embodiment, in order to endow the device with the ability to self-evolve and continuously adapt to new environments, the present invention adds online learning and optimization functions to the device. The key to this embodiment is that the controller has the ability to continuously update its built-in mapping relationship model based on the actual spraying effect. This process forms a complete closed loop of "prediction-execution-evaluation-learning". The device uses a pre-trained mapping relationship model to make decisions. Although this model is more accurate than simple rules, it is essentially a static model. Once trained and deployed on the device, its internal parameters no longer change. However, during long-term use, entirely new working conditions that were not covered in the training dataset may be encountered, such as the use of a new type of asphalt mixture, performance drift caused by batch changes in the release agent, or mechanical wear of the equipment itself. In these cases, a static model may not be able to make the optimal judgment.

[0058] Specifically, after completing one spraying action, the controller's work does not end; it continues with the following steps: First, after the spraying device sprays, the controller waits for a preset time (e.g., after the rubber wheel has rolled half a revolution or a full revolution). Then, the instruction information acquisition unit re-acquires the adhesion status information on the rubber wheel surface. This acquired information is defined by the controller as spraying effect data. This spraying effect data directly reflects the effectiveness of the previous spraying decision. For example, if the re-acquired adhesion area significantly decreases or becomes zero, it indicates that the previous spraying amount was effective, or even excessive; conversely, if the adhesion area does not change or continues to increase, it indicates that the previous spraying amount was insufficient. Next, the controller updates the mapping relationship model based on the spraying effect data. This update process can have various technical paths, but the core idea is to use new "(operating condition -> decision -> effect)" data samples to fine-tune or optimize the existing model.

[0059] One approach is online learning. The controller can use the most recent "(input condition, decision spray volume, spray effect)" as a new training sample. For example, if the spray effect is poor (adhesion area not reduced), the controller can generate a new training data pair with a higher target spray volume and use this single data pair to perform a small gradient update on the weights of the mapping model (e.g., a neural network). Conversely, if the spray effect is excellent (adhesion completely disappears and the surface is too wet), a data pair with a slightly lower target spray volume can be generated for updating. By continuously performing this small-step, real-time model fine-tuning during equipment operation, the model can gradually adapt to the changing operating conditions. Another approach is periodic batch updates. The controller can store all the "(condition, decision, effect)" data accumulated over a period of time (e.g., a workday or a construction section) locally. When the equipment is idle or when a certain number of samples are reached, the controller uses these newly accumulated samples to retrain or incrementally train the entire mapping model. This method of update is more stable and avoids noise interference from single samples.

[0060] For example, if the controller detects that its chosen spray volume fails to effectively prevent asphalt from bonding under a certain operating condition, it will immediately issue an alarm and temporarily increase the spray volume to address the emergency. Simultaneously, it will record this "failure": `{Input: Operating Condition A, Output: Spray Volume Q1, Effect: Poor}`. Subsequently, when encountering a similar operating condition A again, the updated model will tend to output a spray volume Q2 that is larger than Q1. Conversely, if the controller finds that the rubber tire surface remains clean after repeated spraying under a certain operating condition, it will tentatively reduce the spray volume and record the effect, thereby learning a more economical spraying strategy for that condition.

[0061] Next, the controller updates the mapping model based on the spraying effect data. This update process can take several technical approaches, but the core idea is to fine-tune or optimize the existing model using new "(operating condition -> decision -> effect)" data samples. One approach is online learning. The controller can use the most recent "(input operating condition, decision spraying amount, spraying effect)" as a new training sample. For example, if the spraying effect is poor (adhesion area not reduced), the controller can generate a new training data pair with a higher target spraying amount and use this single data pair to perform a small gradient update on the weights of the mapping model (e.g., a neural network). Conversely, if the spraying effect is excellent (adhesion completely disappears and the surface is too wet), a data pair with a slightly lower target spraying amount can be generated for updating. By continuously performing this small-step, real-time model fine-tuning during equipment operation, the model can gradually adapt to the changing operating conditions. Another approach is periodic batch updates. The controller can store all the "(operating conditions, decisions, effects)" data accumulated over a period of time (e.g., a workday or a construction section) locally. When the equipment is idle or when a certain number of samples are reached, the controller uses these newly accumulated samples to retrain or incrementally train the entire mapping model. This method of updating is more stable and avoids noise interference from single samples.

[0062] For example, if the controller detects that its chosen spray volume fails to effectively prevent asphalt from bonding under a certain operating condition, it will immediately issue an alarm and temporarily increase the spray volume to address the emergency. Simultaneously, it will record this "failure": `{Input: Operating Condition A, Output: Spray Volume Q1, Effect: Poor}`. Subsequently, when encountering a similar operating condition A again, the updated model will tend to output a spray volume Q2 that is larger than Q1. Conversely, if the controller finds that the rubber tire surface remains clean after repeated spraying under a certain operating condition, it will tentatively reduce the spray volume and record the effect, thereby learning a more economical spraying strategy for that condition.

[0063] The technical solution in this embodiment endows the intelligent spraying device with the ability to learn and continuously evolve. The controller is no longer a static decision-maker, but an intelligent agent capable of learning from its own experience. This closed-loop optimization mechanism based on actual effect feedback enables the device to automatically adapt to various long-term or sudden factors such as changes in material properties, environmental changes, and equipment aging, always maintaining optimal working conditions. This not only maximizes the utilization efficiency of the release agent but also gives the device strong robustness and long-term economic efficiency, truly achieving the intelligent goal of "becoming smarter with use."

[0064] Example 3: In this embodiment, the present invention also provides an intelligent spraying control method for release agent on rubber-tired rollers. In asphalt pavement construction, existing release agent spraying operations on rubber-tired rollers generally suffer from low levels of automation and intelligence. Operators typically rely on personal experience for manual spraying or use simple timed spraying strategies, neither of which can cope with the changing working conditions at the construction site. As a result, the amount of release agent used is difficult to control precisely, often leading to roller sticking and downtime due to insufficient spraying, or material waste and potential quality risks due to excessive spraying. The method provided in this embodiment aims to achieve on-demand and precise spraying of release agent through a closed-loop control process.

[0065] Please see Figure 2 It illustrates a control flowchart of a preferred embodiment of the method of the present invention, and can be combined with Figure 1 The structure of the device shown is understood. The method includes the following steps: First, the bonding status information of the rubber-tired roller surface is acquired in real time.

[0066] This is the data input and sensing stage of the entire intelligent control method. The purpose of this step is to accurately grasp the actual condition of the surface of the rubber wheel 8. The adhesion state information is any data or signal that can characterize the degree of adhesion of asphalt particles to the surface of the rubber wheel 8.

[0067] In specific implementations, this information can be acquired in various ways. For example, this step can be executed by triggering an image sensor (such as a visible light camera 10) mounted on the frame 5, which captures an image of the surface of the rubber wheel 8 and uses this image data as adhesion status information. This acquisition action can be continuous or periodically executed by the controller at preset time intervals (e.g., once every 0.5 seconds). In another embodiment, this step can also be accomplished by reading the measurement values ​​of a set of optical sensors, which determine the adhesion of asphalt by measuring changes in the light reflectivity of the surface of the rubber wheel 8.

[0068] Then, based on the adhesion state information, the target spraying amount of the release agent used to prevent asphalt adhesion is determined.

[0069] This is the core decision-making step of the method. In this step, the controller (intelligent control module 1) analyzes and calculates the adhesion state information obtained in the previous step to determine the most suitable release agent spraying amount, i.e., the target spraying amount.

[0070] The specific algorithm for determining the target spray volume can be configured according to the required control precision. In a basic implementation, this step may include: the controller analyzes bond state information (such as an image) and calculates a characteristic value representing the asphalt bond (e.g., bond area S). Then, this characteristic value is compared with a preset threshold. If S is less than the threshold, the target spray volume is determined to be zero or a minimum value to maintain basic wetting; if S is greater than or equal to the threshold, the target spray volume is determined to be a preset fixed value.

[0071] In a more refined implementation, this step can employ a function or lookup table to calculate the target spray volume. For example, as mentioned in the technical disclosure, a linear function `Q = kA + Q_min` can be used, where Q is the target spray volume to be determined, A is the bonded area calculated from the bonded state information, k is a proportionality coefficient, and Q_min is the base spray flow rate. The controller substitutes the area A calculated in real time into this formula to obtain a target spray volume Q that is proportional to the current bonded severity.

[0072] Finally, the spraying device is controlled to spray the release agent onto the surface of the rubber wheel at the target spraying amount, so as to achieve dynamic control of the release agent dosage.

[0073] This is the execution phase of the process, translating the decision into physical actions. In this step, the controller converts the target spray volume determined in the previous step into specific control commands and sends them to the spraying device.

[0074] For example, if the target spray volume is a volume value Q, the controller can calculate the required time T for turning on the spray device based on the calibrated flow characteristics of the spray device. Then, the controller outputs a high-level signal to the intelligent control pump 6 or solenoid valve 14 in the spray device and maintains it for time T, before outputting a low-level signal to turn it off, thus accurately spraying a total amount of release agent Q. Alternatively, if the target spray volume is a flow rate value Q', the controller can output a specific analog voltage or PWM (pulse width modulation) signal to the adjustable speed pressure pump 20, causing it to operate at a precise speed, thereby achieving continuous spraying at a flow rate of Q'.

[0075] By implementing the above method, this invention constructs a complete closed-loop control process from real-time perception to intelligent decision-making and precise execution. This method can dynamically and in real-time adjust the spraying amount of release agent based on the rapidly changing adhesion state of the rubber tire surface, achieving on-demand supply. This fundamentally changes the traditional extensive spraying mode, enabling the effective prevention of asphalt sticking to the tire while ensuring construction quality, and maximizing the conservation of release agent materials, reducing project costs, and demonstrating significant economic benefits and technological advancements.

[0076] In the above embodiments, if the step of obtaining bonding state information relies on only a single sensing mode, such as using only conventional visible light imaging, the obtained bonding state information may be biased or erroneous under actual working conditions such as poor lighting conditions (e.g., nighttime or tunnel construction), strong light and shadow changes, or low color contrast between asphalt and rubber tires. This directly affects the accuracy of subsequent spraying decisions and reduces the reliability of the entire method.

[0077] Preferably, in one of the preferred technical solutions of this embodiment, the acquisition step has been further refined and improved. (Refer to again...) Figure 1 The hardware configuration shown, specifically the step of obtaining the adhesion state information of the rubber wheel surface, includes the following sub-steps: First, a visible light image of the surface of the rubber wheel is acquired using a visible light camera.

[0078] This step is performed using a visible light camera 10, 22, for example, mounted on the frame 5. This visible light camera is able to capture color information and high-frequency texture details on the surface of the rubber wheel 8. When asphalt particles adhere to it, the color, roughness, and contour of its surface change, and these changes are recorded in the visible light image.

[0079] Simultaneously, a thermal infrared camera is used to acquire thermal infrared images of the surface of the rubber wheel.

[0080] This step is performed using thermal infrared cameras 11 and 21. The underlying principle is that a significant temperature difference exists between the freshly laid, high-temperature asphalt mixture (e.g., above 120°C) and the relatively cooler surface of the rubber tire 8 (e.g., between 60°C and 80°C). This temperature difference causes the asphalt bonding points to appear as bright hot spots in the thermal infrared image, creating a strong, unaffected contrast with the surrounding rubber tire background. The synchronous acquisition refers to the controller issuing a unified trigger signal to ensure that the visible light image and the thermal infrared image are captured at the same instant, thus guaranteeing precise temporal alignment of the two data sources.

[0081] Finally, the visible light image and the thermal infrared image are fused to generate data characterizing the bonding state information.

[0082] This is a key information processing step in the method of this embodiment, designed to combine the advantages of two images to generate a more reliable and accurate description of the adhesion state than a single image. This fusion step may include: 1. Image Preprocessing: Before fusion, the two images can be optimized. For example, as mentioned in the technical disclosure, contrast-limited adaptive histogram equalization (CLAHE) can be performed on the visible light image to enhance the texture features of asphalt particles; temperature calibration and non-uniformity correction can be performed on the thermal infrared image to obtain a more accurate temperature distribution map.

[0083] 2. Fusion computing: The specific implementation methods of fusion can be diverse.

[0084] In one implementation, feature-level fusion can be employed. The controller first analyzes the thermal infrared image, quickly segmenting all high-temperature regions as candidate regions for asphalt bonding by setting a temperature threshold (e.g., 90°C). Then, it returns to the visible light image, and only within these candidate regions is edge detection or texture analysis algorithms used for secondary confirmation to eliminate interference from other non-asphalt heat sources, ultimately determining the true asphalt bonding region.

[0085] In another implementation, pixel-level weighted fusion can be used. After spatially registering and aligning the two images, for each pixel in the image, the grayscale value of its visible light image and the temperature value of its thermal infrared image are weighted and summed to obtain a fused pixel value, thereby generating a new fused image. In this fused image, the features of the asphalt bonding region will be significantly enhanced.

[0086] Through the above fusion steps, the final generated data can be either a binary mask image, where pixels with a value of 1 represent asphalt bonding areas and pixels with a value of 0 represent the background; or a structured data list recording quantitative information such as the geometric center, area, and perimeter of each bonded patch. This fused data provides more accurate bonding state information and will be used in subsequent steps to determine the target spraying amount.

[0087] By implementing the method of this embodiment, dual-spectral image fusion technology is used to obtain bonding state information, effectively combining the detailed information of visible light images with the high contrast advantage of thermal infrared images. This method greatly improves the accuracy and robustness of asphalt bonding identification in various complex and harsh construction environments, providing high-quality data input for subsequent intelligent decision-making, thereby fundamentally improving the performance and reliability of the entire intelligent spraying control method.

[0088] In the aforementioned technical solution, the step of determining the target spraying amount mainly relies on the adhesion state information of the rubber wheel surface. However, making decisions solely based on geometric information such as adhesion area ignores the crucial physical characteristic that asphalt viscosity changes drastically with temperature. Asphalt adhesion points of the same area exhibit a stronger adhesion tendency and a greater risk of wheel sticking at high temperatures (e.g., during summer construction or immediately after discharge) than at low temperatures. Therefore, a decision-making method based solely on adhesion state still has room for improvement in accuracy and adaptability to different operating conditions.

[0089] Preferably, in one of the preferred technical solutions of this embodiment, in order to achieve more refined and intelligent decision-making, the present invention further introduces a temperature dimension in the decision-making process. The method first adds an extra step: acquiring the temperature data of the rubber wheel surface. This step is executed synchronously or asynchronously with the step of acquiring adhesion state information. Specifically, this can be achieved by reading one or more temperature sensors (such as...) mounted on the frame 5. Figure 1 The temperature is measured by thermometers 9, 17, and 23. These temperature sensors, preferably non-contact infrared thermometers, can be aimed at key locations such as the tire tread of the rubber tire 8 and the road surface to be compacted, to collect temperature readings in real time and transmit these temperature data to the controller for processing.

[0090] The core improvement of this method lies in the fact that the step of determining the target spray volume combines the adhesion state information and the temperature data to jointly determine the target spray volume. This means that the decision-making process is no longer a single-variable input, but a multi-variable input.

[0091] The steps determined by this combination can be implemented in several specific ways: One implementation approach is based on multi-dimensional rule-based decision-making. This step may include: First, classifying the bonding state information (e.g., bonding area) into levels, such as "low-value zone," "medium-value zone," and "high-value zone" as mentioned in the technical disclosure. Simultaneously, the acquired temperature data is also classified into levels, such as "low-temperature zone" (below 60℃), "medium-temperature zone" (60-80℃), and "high-temperature zone" (above 80℃). Then, in a pre-defined two-dimensional decision matrix or decision tree, the final target spraying volume is found or derived based on the combination of the current bonding area and temperature levels. For example, a bonding area in a "high-value zone" occurring in a "high-temperature zone" will trigger the highest level of spraying volume; while the same "high-value zone" area occurring in a "low-temperature zone" may only trigger a medium level of spraying volume.

[0092] Another approach is to use dynamic calculations with multivariable functions. This step may involve substituting the bonding state information (such as the bonding area A) and temperature data T into a pre-defined multivariable function for calculation. For example, a function of the form `Q = f(A, T)` could be used. This function `f` is designed so that the spraying rate Q increases with increasing area A and temperature T. A concrete example is `Q = (k_1 + k_2)`. T) A + Q_min`, where `k_1` and `k_2` are preset coefficients. In this function, temperature T directly affects the area-related spray volume coefficient, thereby dynamically adjusting the spraying strategy.

[0093] Therefore, in a complete workflow, after acquiring the adhesion status information, the controller further acquires temperature data. Subsequently, the controller uses these two different dimensions of information as common inputs to the decision-making system. By executing the aforementioned steps such as decision-making based on multidimensional rules or dynamic calculation of multivariable functions, it ultimately determines a target spray volume that reflects both the adhesion range and the adhesion risk (determined by temperature).

[0094] By implementing the method in this embodiment, temperature, a key physical quantity, is incorporated into the spraying decision model, upgrading the decision-making basis from a single dimension to a multi-dimensional one. This method can more scientifically and comprehensively assess the actual risk of wheel sticking, allowing the spraying strategy to intelligently adapt to construction conditions under different ambient and asphalt material temperatures. This significantly improves the accuracy and intelligence of spraying decisions, thereby further optimizing the consumption of release agent while ensuring the anti-sticking effect, achieving a higher level of resource conservation and cost control.

[0095] Example 4: In Example 3, although multi-dimensional decision-making was achieved by introducing temperature data, the decision matrix or simple multivariate function used is essentially a linear, rule-based system. Such a system may struggle to accurately depict the highly complex and non-linear relationship between operating conditions (bonding state, temperature, etc.) and the optimal spraying rate. Furthermore, these rules and parameters typically require manual setting and tedious on-site debugging, resulting in limited generalization ability. This example, from a methodological perspective, elaborates on how to replace simple rules or functions with a more advanced intelligent model to achieve higher-precision decision-making, focusing on the execution steps of the method.

[0096] To overcome the aforementioned limitations and achieve higher-order intelligent decision-making, a more preferred embodiment of the method of the present invention significantly improves the "determining the target spray volume" step. This step no longer relies on fixed rules or functions, but instead introduces a pre-trained model to perform the decision. The step of determining the target spray volume specifically includes: First, the bonding state information and / or the temperature data are input into a pre-trained mapping model. This is the data preparation and input step in the decision-making process. In this step, the controller first processes the raw information obtained from each sensor into a format acceptable to the model. For example, the obtained bonding state information (such as total bonding area, maximum patch size, number of patches, etc.) and the obtained temperature data (such as tire surface temperature, asphalt pavement temperature, etc.) are combined into a multi-dimensional input feature vector. For example, an input vector could be `[bonding area, number of patches, tire temperature]`.

[0097] The mapping relationship model is an algorithmic model trained using machine learning methods. The mapping relationship model in this embodiment can be of various types, including but not limited to: Feedforward Neural Network (FNN) or Multilayer Perceptron (MLP): Learns a complex nonlinear mapping from input features to output spray volume through multiple hidden layers.

[0098] Gradient Boosting Decision Trees (GBDT), such as XGBoost or LightGBM models, are adept at handling tabular data and can efficiently build powerful predictive models.

[0099] Support Vector Regression (SVR): A regression algorithm based on support vector machines, suitable for solving small-sample, non-linear regression problems.

[0100] The model is pre-trained. This means that the model was trained on a large amount of collected data before the device was deployed. The training dataset contains a large number of `(input feature vector, corresponding optimal spray rate)` data pairs. This data can be collected through operation records of expert drivers under real-world conditions, or calibrated through repeated experiments in a controlled environment. The goal of training is to optimize the model's parameters (such as the weights and biases of the neural network) so that it can output a prediction value as close as possible to the "optimal spray rate" when given any set of input features.

[0101] Next, the target spraying amount is output using the mapping relationship model.

[0102] This describes the calculation and output steps of the decision-making process. After the input feature vector is fed into the pre-trained mapping model, the model performs a forward propagation or inference calculation. This calculation is entirely determined by the model's internal mathematical structure and optimized parameters. The result is a single numerical value, which is defined by this method as the target spraying amount. For example, when the input vector is `[adhesion area = 25 cm², number of patches = 3, rubber wheel temperature = 78°C]`, a trained neural network model might calculate and output `12.5`, in milliliters (ml) or milliliters per second (ml / s). This output value is then passed to the next step of the method: controlling the spraying device to perform the spraying.

[0103] By implementing the method of this embodiment, the process of determining the target spray volume is upgraded from a "hard-coded" rule system to a "data-driven" machine learning system. This provides a "technical solution with a different concept." This method utilizes a pre-trained mapping relationship model to autonomously learn from historical data and extract deep patterns hidden between multi-dimensional operating conditions and optimal spraying strategies. This makes spraying decisions more accurate and robust, and can better generalize to various complex operating conditions not explicitly defined in the rules, thereby achieving more refined and intelligent management of the release agent at a higher level.

[0104] The above technical solution employs a pre-trained mapping model for decision-making. While this model is more accurate than simple rules or functions, it is essentially a static model. Once trained and deployed, its internal parameters remain unchanged. However, during long-term construction, entirely new working conditions not covered in the training dataset may arise, such as the use of new asphalt mixtures, performance drift due to batch variations in release agents, or mechanical wear of the equipment itself. In these situations, executing the decision-making steps of a static model may not yield optimal spraying results.

[0105] Preferably, in one of the preferred technical solutions of this embodiment, in order to endow the control method with the ability to self-evolve and continuously adapt to new environments, the present invention adds an online learning and optimization step based on the above-described implementation method. The key to this embodiment is that the method process has the ability to continuously update its built-in mapping relationship model based on the actual spraying effect.

[0106] This process forms a complete closed loop of "prediction-execution-evaluation-learning" at the methodological level. Specifically, after executing one spray control step, the method will continue to perform the following steps: First, after spraying the release agent, the adhesion status information of the rubber wheel surface is obtained again as spraying effect data.

[0107] This is a quantitative evaluation step of the effectiveness of the previous decision. The specific execution method is as follows: after the spraying device completes spraying, wait for a preset time (e.g., after the rubber roller has rolled half a revolution or one revolution), and then trigger the "acquire adhesion status information" step again. The information acquired this time is defined by the method as spraying effect data. This spraying effect data directly reflects the effectiveness of the previous decision. For example, if the acquired adhesion area significantly decreases or becomes zero, it indicates that the previously determined target spraying amount was effective, or even excessive; conversely, if the adhesion area does not change or continues to increase, it indicates that the previous target spraying amount was insufficient.

[0108] Next, the mapping relationship model is updated using the spraying effect data to continuously optimize the process of determining the target spraying amount.

[0109] This is the self-learning and optimization step of this method. The core idea of ​​this step is to use new "(working condition -> decision -> effect)" data samples to fine-tune or optimize the existing mapping relationship model. This step can have several technical approaches: One approach is online learning. This step might involve using the most recent "(input condition, decision spray volume, spray effect)" as a new training sample. For example, if the spray effect is poor (adhesion area not reduced), a new training data pair with a higher target spray volume is generated, and this single data pair is used to make a small gradient update to the weights of the mapping model (e.g., a neural network). Conversely, if the spray effect is excellent (adhesion completely disappears and the surface is too wet), a data pair with a slightly lower target spray volume is generated for updating. By continuously performing these small, real-time model fine-tuning steps during equipment operation, the model can gradually adapt to the changing operating conditions.

[0110] Another approach is periodic batch updates. This step may involve storing all the "(conditions, decisions, effects)" data accumulated over a period of time (e.g., a workday or a construction section). When the equipment is idle or a certain number of samples are reached, these newly accumulated samples are used to retrain or incrementally train the entire mapping model. As described in your knowledge base file "6G Network AI Applications," using real-world data as "fine-tuning training data" makes this update more stable and avoids noise interference from single samples.

[0111] For example, if the spraying amount determined by the method fails to effectively prevent asphalt bonding under a certain working condition, the method will immediately record this "failure": `{Input features: Vector of working condition A, Output: Spraying amount Q1, Effect: Poor}`. Subsequently, during the model update step, the model will be adjusted based on this record. The next time a similar working condition A is encountered, the updated model will tend to output a spraying amount Q2 that is larger than Q1 when performing the decision step. Conversely, if the method finds that the rubber tire surface remains clean after repeated spraying under a certain working condition, it will tentatively reduce the spraying amount and record the effect, thereby learning a more economical spraying strategy for that working condition.

[0112] By implementing the technical method of this embodiment, the entire intelligent spraying control process is endowed with the ability to learn and continuously evolve. This method is no longer a static decision-making process, but an intelligent process capable of learning from its own experience. This closed-loop optimization mechanism based on actual effect feedback provides a different conceptual technical solution for addressing long-term adaptability issues. It enables the spraying control method to automatically adapt to various long-term or sudden factors such as changes in material properties, environmental changes, and equipment aging, always maintaining optimal operating conditions. This not only maximizes the utilization efficiency of the release agent but also gives the method strong robustness and long-term economic efficiency, truly achieving the intelligent goal of "becoming smarter with use."

[0113] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A smart spraying device for a release agent on a rubber-tired roller, comprising: The information acquisition unit is used to acquire the bonding status information of the rubber-tired roller surface in real time; A spraying device for spraying a release agent onto the surface of the rubber wheel; and The controller is connected to the information acquisition unit and the spraying device respectively. The controller is used to: determine the target spraying amount of the release agent based on the adhesion state information obtained from the information acquisition unit. The spraying device is controlled to spray at the target spraying amount to achieve dynamic control of the amount of release agent used.

2. The intelligent spraying device for release agent for rubber-tired rollers according to claim 1, characterized in that, The information acquisition unit includes: A visible light camera is used to acquire visible light images of the surface of the rubber wheel; and A thermal infrared camera is used to synchronously acquire thermal infrared images of the surface of the rubber wheel; The controller is also used to fuse the visible light image and the thermal infrared image to generate data characterizing the bonding state information.

3. The intelligent spraying device for release agent for rubber-tired rollers according to claim 2, characterized in that, The device further includes: A temperature sensor is used to acquire temperature data of the surface of the rubber wheel; The controller determines the target spray volume by combining the adhesion state information and the temperature data.

4. The intelligent spraying device for release agent for rubber-tired rollers according to claim 3, characterized in that, The controller has a built-in mapping model, and the controller determines the target spraying amount by taking the adhesion state information and / or the temperature data as input to the model and taking the output of the model as the target spraying amount.

5. The intelligent spraying device for release agent for rubber-tired rollers according to claim 4, characterized in that, The controller is also configured to: after the spraying device has sprayed, acquire adhesion status information from the information acquisition unit again as spraying effect data; and update the mapping relationship model based on the spraying effect data.

6. A method for intelligent spraying control of release agent for rubber-tired rollers, characterized in that, Includes the following steps: Real-time acquisition of adhesion status information on the surface of the rubber-tired rollers on the rubber-tired roller; Based on the bonding state information, the target spraying amount of the release agent used to prevent asphalt bonding is determined; The spraying device is controlled to spray the release agent onto the surface of the rubber wheel at the target spraying amount, so as to achieve dynamic control of the release agent dosage.

7. The intelligent spraying control method for release agent in rubber-tired rollers according to claim 6, characterized in that, The step of acquiring the adhesion status information of the rubber-tired roller surface in real time includes: A visible light image of the surface of the rubber wheel is captured using a visible light camera; The thermal infrared image of the rubber wheel surface is simultaneously acquired using a thermal infrared camera; and The visible light image and the thermal infrared image are fused to generate data characterizing the bonding state information.

8. The intelligent spraying control method for release agent in rubber-tired rollers according to claim 7, characterized in that, The method further includes: Obtain the temperature data of the surface of the rubber wheel; The step of determining the target spray amount of the release agent for preventing asphalt adhesion based on the adhesion state information is to combine the adhesion state information and the temperature data to jointly determine the target spray amount.

9. The intelligent spraying control method for release agent in rubber-tired rollers according to claim 8, characterized in that, The step of combining the adhesion state information and the temperature data to jointly determine the target spraying amount includes: inputting the adhesion state information and / or the temperature data into a pre-trained mapping relationship model; and using the mapping relationship model to output the target spraying amount.

10. The intelligent spraying control method for release agent in rubber-tired rollers according to claim 9, characterized in that, The method further includes: After spraying the release agent, the adhesion state information of the rubber wheel surface is acquired again as spraying effect data; and the mapping relationship model is updated using the spraying effect data to continuously optimize the process of determining the target spraying amount.

Images