Brush roller self-adaptive lifting control method based on photovoltaic module surface obstacle identification

Abstract

The present application relates to a brush roller adaptive lifting control method based on photovoltaic module surface obstacle identification, belonging to the technical field of photovoltaic cleaning robot. The method comprises: identifying the type, size, spatial position and projection interval of the obstacle in the axial direction of the brush roller; obtaining the rotation speed, torque current and robot travel speed of the brush roller, and calculating the collision risk coefficient; when the collision risk coefficient exceeds the threshold value, a three-dimensional obstacle avoidance trajectory containing a variable speed lifting curve and a lateral offset curve is dynamically generated according to the height, slope, projection interval of the obstacle and the elastic compression allowance of the brush roller; before contacting the obstacle, the brush roller is controlled to perform lifting and axial offset according to the three-dimensional obstacle avoidance trajectory; after confirming that the obstacle is crossed, the brush roller is controlled to reset to the initial working height and axial position; the characteristic parameters of the obstacle and the three-dimensional obstacle avoidance trajectory parameters are recorded to the local knowledge base as support for subsequent similar situations. The adaptive three-dimensional obstacle avoidance of the brush roller is realized, and the cleaning continuity and safety are improved.

Description

Technical Field

[0001] This invention belongs to the field of photovoltaic cleaning robot technology, specifically relating to an adaptive lifting control method for brush rollers based on obstacle recognition on the surface of photovoltaic modules. Background Technology

[0002] Photovoltaic power generation, as a clean and renewable energy source, has been widely adopted in recent years. Photovoltaic modules are typically installed outdoors, and their surfaces easily accumulate pollutants such as dust, bird droppings, and fallen leaves, severely impacting power generation efficiency. Therefore, regular cleaning of the photovoltaic module surfaces is a crucial aspect of power plant operation and maintenance. Photovoltaic cleaning robots have emerged to address this need, with tracked cleaning robots becoming one of the mainstream models for large-scale photovoltaic power plants due to their excellent maneuverability and stability.

[0003] The core component of a photovoltaic cleaning robot is the brush roller, which contacts the surface of the photovoltaic modules and removes dirt through rotation. In actual operation, the surface of photovoltaic modules is not completely flat and often contains various protruding obstacles, such as: frame edges between modules, junction boxes, fragments warped from the module surface due to heat spots or external forces, hard bumps formed by long-term dust accumulation, and bird droppings. When the brush roller rotates at high speed and approaches these protruding obstacles, the bristles collide violently with them. Because existing tracked photovoltaic cleaning robots generally lack the function of lifting and avoiding obstacles with their brush rollers, the brush rollers can only forcibly cross or crush obstacles, leading to the following problems: First, the brush bristles are prone to bending, breaking, and falling off after impacting hard protrusions, significantly shortening the service life of the brush rollers; Second, when the height of the obstacle exceeds the elastic compression margin of the brush rollers, the brush rollers may be stuck, causing the robot's drive motor to overload or even burn out, or causing the tracks to slip and the robot to stop; Third, even if they can barely cross the obstacle, the impact vibration of the brush rollers at the obstacle will be transmitted to the robot body, which may cause structural loosening or damage to electronic components in the long run.

[0004] To address the aforementioned issues, some existing technologies attempt to detect obstacles using visual or laser sensors and control the robot to either navigate around them or stop and wait for manual intervention. However, navigating around obstacles requires significant space and path planning capabilities, which is often difficult to implement in densely packed photovoltaic arrays; stopping and waiting severely impacts cleaning efficiency. Other technologies propose installing elastic suspension mechanisms on the brush rollers, allowing them to passively compress when encountering obstacles. However, passive suspension suffers from lag in response, limited lifting capacity, and difficulty adapting to obstacles of varying heights and shapes, and it cannot actively prevent wear and tear on the brush bristles caused by obstacles.

[0005] Therefore, how to enable photovoltaic cleaning robots to actively and adaptively avoid various protruding obstacles on the surface of photovoltaic modules while maintaining cleaning efficiency, and to prevent brush roller wear and equipment jamming, is a technical problem that urgently needs to be solved in this field. To address the above-mentioned technical deficiencies, this invention proposes a brush roller adaptive lifting control method based on obstacle recognition. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides an adaptive lifting control method for brush rollers based on obstacle recognition on the surface of photovoltaic modules. The objective of this invention can be achieved through the following technical solution: An adaptive lifting control method for brush rollers based on obstacle recognition on the surface of photovoltaic modules includes: Identify the type, geometry, spatial location, and projection range of obstacles along the brush roller axis; obtain the brush roller rotation speed, torque current, and the sweeping robot's travel speed; and calculate the collision risk coefficient. When the collision risk coefficient exceeds the threshold, a three-dimensional obstacle avoidance trajectory is dynamically generated based on the height and slope of the obstacle, the projection range, and the elastic compression margin of the brush roller; the three-dimensional obstacle avoidance trajectory includes a variable speed lifting curve in the lifting direction and a lateral offset curve in the brush roller axis. Before contacting the obstacle, the brush roller is controlled to lift and axially offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory. The height of the brush roller exceeds the height of the obstacle and the axial position of the brush roller moves out of the axial projection range of the obstacle. After confirming that the obstacle has been passed, the brush roller is controlled to return to the initial working height and axial position according to the surface flatness behind the obstacle and the empty space of the axial projection range. The feature parameters and 3D obstacle avoidance trajectory parameters of the current obstacle are recorded in the local knowledge base. Based on the local knowledge base, when similar obstacles are subsequently identified, the corresponding obstacle avoidance trajectory parameters are directly called for prediction and strategy adjustment.

[0007] Specifically, the obstacle identification includes: using at least one RGB camera and one depth camera, based on the fusion of image semantic segmentation results and depth information, identifying the type, geometric size, spatial location, and projection range of the obstacle along the brush roller axis; the obstacle types include at least: photovoltaic module frames, junction boxes, dust accumulation bumps, bird droppings, and broken or warped fragments of the module; when identified as a flexible obstacle, the calculation weight of the collision risk coefficient is reduced; when identified as a rigid protruding obstacle, the calculation weight of the collision risk coefficient is increased.

[0008] Specifically, the collision risk coefficient is determined by a combination of the following factors: The difference between the actual height of the obstacle and the elastic compression margin of the brush roller, the ratio between the length of the obstacle along the direction of travel of the sweeping robot and the current braking distance of the sweeping robot, and the instantaneous rate of change of the brush roller torque current per unit time. The above factors are assigned different evaluation levels, and a collision risk coefficient between 0 and 1 is output through fuzzy logic rules. Among them, the instantaneous change rate of torque current is used as an input of fuzzy logic. The larger the value, the higher the output collision risk coefficient.

[0009] Specifically, the variable speed lifting curve is an S-shaped speed curve, which includes a uniform acceleration segment, a uniform speed segment, and a uniform deceleration segment. The acceleration of the uniform acceleration segment is dynamically adjusted according to the change of the obstacle slope: when the obstacle slope is less than a first slope threshold, a first acceleration value is used; when the obstacle slope is greater than a second slope threshold, a second acceleration value less than the first acceleration value is used.

[0010] Specifically, the target offset of the lateral offset curve is at least equal to the sum of the width of the projection range of the obstacle on the brush roller axis and the safety margin reserved at both ends of the brush roller, and the offset direction is selected as the direction of the nearest axial free area; if the width of the axial free area on both sides of the brush roller is less than the preset minimum value, only lifting to avoid the obstacle is performed and no axial offset is performed.

[0011] As a preferred technical solution of the present invention, the control brush roller performs lifting and axial offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory, including: dynamically selecting the execution order of lifting and axial offset according to the shape characteristics of the obstacle; When the ratio of the height of the obstacle to the diameter of the brush roller is less than the first threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is greater than the second threshold, the brush roller is controlled to first perform axial offset and then lift. Conversely, when the ratio of the height of the obstacle to the diameter of the brush roller is greater than the third threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is less than the fourth threshold, the brush roller is controlled to first perform lifting and then axial offset. For obstacles whose height-to-diameter ratio and axial projection width-to-brush roller length ratio are both within the intermediate threshold range, the lifting and axial offset are activated simultaneously.

[0012] Specifically, the control of resetting the brush roller to the initial working height and axial position includes: using segmented reset, firstly restoring the height to a preset ratio of the initial working height at a first reset speed, pausing the reset action and visually confirming again that the surface flatness behind the obstacle and the absence of residual obstacles in the axial projection range, and then fully resetting the brush roller to the initial working height and axial working position at a second reset speed; wherein, the first reset speed is greater than the second reset speed; and simultaneously, the axial reset and height reset are synchronized in time.

[0013] Specifically, when the length of the obstacle in the direction of the cleaning robot's travel exceeds a preset length threshold, and there are flat areas on both sides of the obstacle that allow the cleaning robot to pass, while controlling the brush roller to lift and axially deflect to avoid the obstacle, the travel path of the cleaning robot is further adjusted. The adjustment of the travel path includes: planning a local arc trajectory that bypasses the obstacle, and while the cleaning robot is walking on this trajectory, the brush roller remains in a lifted and axially deflected state, and the original straight travel path is restored after the cleaning robot has completely passed the obstacle.

[0014] Specifically, the local knowledge base stores multiple obstacle feature templates and their corresponding optimal three-dimensional obstacle avoidance trajectory parameters. The feature templates include the type of obstacle, typical geometric size range, and common position distribution patterns. When it is found that the matching degree between the current obstacle and a certain feature template exceeds a preset threshold, the optimal lifting start time, target lifting height, axial offset direction, and offset amount stored in the template are directly called as initial parameters. During the execution process, online fine-tuning is performed based on the actual detected brush roller torque current and contact feedback. The fine-tuned parameters are used to update the template.

[0015] Specifically, when visual detection identifies multiple obstacles on the same photovoltaic module surface, the collision risk coefficient of each obstacle is calculated sequentially according to the direction of travel of the cleaning robot. If the distance between adjacent obstacles is less than the minimum distance required for the brush roller to reset its axial offset, the brush roller is controlled not to perform axial reset after passing the first obstacle, but to maintain the current axial offset and continuously pass subsequent obstacles. If the distance between adjacent obstacles is greater than or equal to the minimum distance, the brush roller is controlled to reset after passing the current obstacle, and then a three-dimensional obstacle avoidance trajectory is regenerated for the next obstacle.

[0016] Specifically, it also includes an anomaly handling process: if any of the following anomalies are detected during the lifting or axial offset process: the brush roller torque current exceeds the safety threshold, visual detection feedback indicates that the brush roller is still in obvious contact with the obstacle, or the actual displacement of the brush roller is detected to deviate from the target three-dimensional obstacle avoidance trajectory beyond the allowable range, an emergency stop is immediately executed. At the same time, the brush roller is controlled to reset to the initial position in reverse order at an emergency reset speed, and an alarm message containing the obstacle location, anomaly type, and timestamp is sent to the remote terminal. After the emergency stop, the cleaning robot pauses its movement and waits for remote instructions or manual intervention.

[0017] Specifically, during the lifting and axial offset obstacle avoidance of the brush roller, the rotational speed of the brush roller is simultaneously reduced to a preset ratio of the normal rotational speed; at the same time, the traveling speed of the cleaning robot is increased according to the length of the obstacle avoidance path; after obstacle avoidance is completed, visual inspection confirms that the brush roller has passed the obstacle and the rear surface is flat, and then the rotational speed of the brush roller and the traveling speed of the cleaning robot are gradually restored to normal working parameters. The speed restoration process adopts a slow start method, wherein the acceleration of the slow start is a preset fixed value or a curve that increases linearly with time.

[0018] The beneficial effects of this invention are as follows: This invention achieves active three-dimensional obstacle avoidance with brush rollers, significantly reducing brush roller wear and the risk of equipment jamming: Existing photovoltaic cleaning robots lack brush roller lifting capabilities, forcing them to rigidly pass through protruding obstacles, leading to brush bristle wear and even equipment jamming. This invention uses visual detection to identify the type, geometry, and spatial location of obstacles in real time. It combines brush roller rotation speed, torque current, and robot speed to comprehensively assess collision risk, proactively generating a three-dimensional obstacle avoidance trajectory including variable-speed lifting curves and lateral offset curves before contact with obstacles. The brush roller can not only lift above the highest point of the obstacle in the height direction but also offset axially to avoid the obstacle's projection area in the axial direction, achieving active obstacle avoidance within a three-dimensional spatial range.

[0019] This invention dynamically optimizes obstacle avoidance strategies based on obstacle shape, avoiding excessive lifting or ineffective offset: It dynamically selects the execution order of lifting and axial offset based on the obstacle's height, slope, and axial projection width: long, flat obstacles are prioritized for axial offset, tall obstacles are prioritized for lifting, and intermediate obstacles are activated simultaneously. The variable-speed lifting curve adjusts acceleration according to the slope, avoiding excessive lifting (creating blind spots) or ineffective offset caused by traditional single lifting methods, thus improving obstacle avoidance accuracy and cleaning continuity.

[0020] The invention combines segmented reset with surface flatness detection to prevent secondary collisions: The invention adopts a segmented reset strategy: First, it quickly restores the target height to a preset ratio, pauses and re-detects the surface flatness and axial projection interval clearance behind the obstacle, and after confirming that there are no remaining obstacles, it slowly and completely resets the object. The axial and height resets are carried out simultaneously, which effectively avoids secondary collisions during the reset process.

[0021] Knowledge base self-learning enables continuous strategy optimization: This invention records obstacle characteristic parameters (type, size, position) and the adopted 3D obstacle avoidance trajectory parameters in a local knowledge base during each obstacle avoidance process. When a similar obstacle is identified again, the optimal parameters in the matching template can be directly called as the initial values, and fine-tuned online based on torque current and contact feedback during execution. The fine-tuned parameters update the template in reverse. As the running time increases, the obstacle avoidance strategy is continuously optimized, and parameters such as the lifting start time, target height, and offset tend to be optimal, with response speed and processing accuracy gradually improving. Attached Figure Description

[0022] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart illustrating the overall method of the present invention; Figure 2 This is a schematic diagram of the three-dimensional obstacle avoidance trajectory of the present invention; Figure 3 This is the lifting and offset timing decision diagram of the present invention; Figure 4 This is a flowchart of the segmented reset process of the present invention; Figure 5 This is a schematic diagram of the continuous multi-obstacle processing of the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0025] Please see Figures 1-5 An adaptive lifting control method for brush rollers based on obstacle recognition on the surface of photovoltaic modules includes: Identify the type, geometry, spatial location, and projection range of obstacles along the brush roller axis; obtain the brush roller rotation speed, torque current, and the sweeping robot's travel speed; and calculate the collision risk coefficient. When the collision risk coefficient exceeds the threshold, a three-dimensional obstacle avoidance trajectory is dynamically generated based on the height and slope of the obstacle, the projection range, and the elastic compression margin of the brush roller; the three-dimensional obstacle avoidance trajectory includes a variable speed lifting curve in the lifting direction and a lateral offset curve in the brush roller axis. Before contacting the obstacle, the brush roller is controlled to lift and axially offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory. The height of the brush roller exceeds the height of the obstacle and the axial position of the brush roller moves out of the axial projection range of the obstacle. After confirming that the obstacle has been passed, the brush roller is controlled to return to the initial working height and axial position according to the surface flatness behind the obstacle and the empty space of the axial projection range. The feature parameters and 3D obstacle avoidance trajectory parameters of the current obstacle are recorded in the local knowledge base. Based on the local knowledge base, when similar obstacles are subsequently identified, the corresponding obstacle avoidance trajectory parameters are directly called for prediction and strategy adjustment.

[0026] Specifically, the obstacle identification includes: using at least one RGB camera and one depth camera, based on the fusion of image semantic segmentation results and depth information, identifying the type, geometric size, spatial location, and projection range of the obstacle along the brush roller axis; the obstacle types include at least: photovoltaic module frames, junction boxes, dust accumulation bumps, bird droppings, and broken or warped fragments of the module; when identified as a flexible obstacle, the calculation weight of the collision risk coefficient is reduced; when identified as a rigid protruding obstacle, the calculation weight of the collision risk coefficient is increased.

[0027] Specifically, the collision risk coefficient is determined by a combination of the following factors: The difference between the actual height of the obstacle and the elastic compression margin of the brush roller, the ratio between the length of the obstacle along the direction of travel of the sweeping robot and the current braking distance of the sweeping robot, and the instantaneous rate of change of the brush roller torque current per unit time. The above factors are assigned different evaluation levels, and a collision risk coefficient between 0 and 1 is output through fuzzy logic rules. Among them, the instantaneous change rate of torque current is used as an input of fuzzy logic. The larger the value, the higher the output collision risk coefficient.

[0028] Specifically, the variable speed lifting curve is an S-shaped speed curve, which includes a uniform acceleration segment, a uniform speed segment, and a uniform deceleration segment. The acceleration of the uniform acceleration segment is dynamically adjusted according to the change of the obstacle slope: when the obstacle slope is less than a first slope threshold, a first acceleration value is used; when the obstacle slope is greater than a second slope threshold, a second acceleration value less than the first acceleration value is used.

[0029] Specifically, the target offset of the lateral offset curve is at least equal to the sum of the width of the projection range of the obstacle on the brush roller axis and the safety margin reserved at both ends of the brush roller, and the offset direction is selected as the direction of the nearest axial free area; if the width of the axial free area on both sides of the brush roller is less than the preset minimum value, only lifting to avoid the obstacle is performed and no axial offset is performed.

[0030] Specifically, the control brush roller performs lifting and axial offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory, including: dynamically selecting the execution order of lifting and axial offset according to the shape characteristics of the obstacle; When the ratio of the height of the obstacle to the diameter of the brush roller is less than the first threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is greater than the second threshold, the brush roller is controlled to first perform axial offset and then lift. Conversely, when the ratio of the height of the obstacle to the diameter of the brush roller is greater than the third threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is less than the fourth threshold, the brush roller is controlled to first perform lifting and then axial offset. For obstacles whose height-to-diameter ratio and axial projection width-to-brush roller length ratio are both within the intermediate threshold range, the lifting and axial offset are activated simultaneously.

[0031] Specifically, the control of resetting the brush roller to the initial working height and axial position includes: using segmented reset, firstly restoring the height to a preset ratio of the initial working height at a first reset speed, pausing the reset action and visually confirming again that the surface flatness behind the obstacle and the absence of residual obstacles in the axial projection range, and then fully resetting the brush roller to the initial working height and axial working position at a second reset speed; wherein, the first reset speed is greater than the second reset speed; and simultaneously, the axial reset and height reset are synchronized in time.

[0032] Specifically, when the length of the obstacle in the direction of the cleaning robot's travel exceeds a preset length threshold, and there are flat areas on both sides of the obstacle that allow the cleaning robot to pass, while controlling the brush roller to lift and axially deflect to avoid the obstacle, the travel path of the cleaning robot is further adjusted. The adjustment of the travel path includes: planning a local arc trajectory that bypasses the obstacle, and while the cleaning robot is walking on this trajectory, the brush roller remains in a lifted and axially deflected state, and the original straight travel path is restored after the cleaning robot has completely passed the obstacle.

[0033] Specifically, the local knowledge base stores multiple obstacle feature templates and their corresponding optimal three-dimensional obstacle avoidance trajectory parameters. The feature templates include the type of obstacle, typical geometric size range, and common position distribution patterns. When it is found that the matching degree between the current obstacle and a certain feature template exceeds a preset threshold, the optimal lifting start time, target lifting height, axial offset direction, and offset amount stored in the template are directly called as initial parameters. During the execution process, online fine-tuning is performed based on the actual detected brush roller torque current and contact feedback. The fine-tuned parameters are used to update the template.

[0034] Specifically, when visual detection identifies multiple obstacles on the same photovoltaic module surface, the collision risk coefficient of each obstacle is calculated sequentially according to the direction of travel of the cleaning robot. If the distance between adjacent obstacles is less than the minimum distance required for the brush roller to reset its axial offset, the brush roller is controlled not to perform axial reset after passing the first obstacle, but to maintain the current axial offset and continuously pass subsequent obstacles. If the distance between adjacent obstacles is greater than or equal to the minimum distance, the brush roller is controlled to reset after passing the current obstacle, and then a three-dimensional obstacle avoidance trajectory is regenerated for the next obstacle.

[0035] Specifically, it also includes an anomaly handling process: if any of the following anomalies are detected during the lifting or axial offset process: the brush roller torque current exceeds the safety threshold, visual detection feedback indicates that the brush roller is still in obvious contact with the obstacle, or the actual displacement of the brush roller is detected to deviate from the target three-dimensional obstacle avoidance trajectory beyond the allowable range, an emergency stop is immediately executed. At the same time, the brush roller is controlled to reset to the initial position in reverse order at an emergency reset speed, and an alarm message containing the obstacle location, anomaly type, and timestamp is sent to the remote terminal. After the emergency stop, the cleaning robot pauses its movement and waits for remote instructions or manual intervention.

[0036] Specifically, during the lifting and axial offset obstacle avoidance of the brush roller, the rotational speed of the brush roller is simultaneously reduced to a preset ratio of the normal rotational speed; at the same time, the traveling speed of the cleaning robot is increased according to the length of the obstacle avoidance path; after obstacle avoidance is completed, visual inspection confirms that the brush roller has passed the obstacle and the rear surface is flat, and then the rotational speed of the brush roller and the traveling speed of the cleaning robot are gradually restored to normal working parameters. The speed restoration process adopts a slow start method, wherein the acceleration of the slow start is a preset fixed value or a curve that increases linearly with time.

[0037] Example I. Application Scenarios and Hardware Configuration Taking a large-scale photovoltaic power station as an example, the photovoltaic modules are conventional polycrystalline silicon modules, measuring 1650mm × 992mm. There are approximately 5mm high borders between the modules, and each module has a junction box approximately 15mm high in the center. The cleaning area consists of multiple consecutive modules in the module array. The cleaning robot uses a tracked chassis, with a brush roller mounted at the front end. The brush roller's length matches the module width (approximately 1000mm), its diameter is 120mm, and the bristles are made of nylon with a 10mm elastic compression allowance.

[0038] The robot is equipped with the following vision inspection modules: an RGB camera (1920×1080 resolution, 30fps) and a depth camera (ToF principle, measurement range 0.5~3m, accuracy ±5mm), both mounted together at the front of the robot with a top-down angle of approximately 30°, covering the surface of components within a range of 0.2~1.5m in front of the robot. The robot also includes: a brush roller drive motor (with torque and current detection), a travel drive motor (with speed encoder), a brush roller actuator (two-degree-of-freedom parallel mechanism: two electric push rods, stroke 0~50mm, thrust 100N; one axial slide, stroke ±30mm, driven by a stepper motor), an industrial control computer (running a Linux system, deploying a deep learning inference engine), and local storage (for a knowledge base).

[0039] II. Typical Execution Process Scenario 1: Single junction box (rigid protrusion, height 15mm, axial projection width approximately 80mm) Step 1: Visual Recognition The robot moves at a speed of 0.2 m / s. An RGB camera captures images of the front, a deep learning model (such as U-Net semantic segmentation) outputs the obstacle category as "junction box," and a depth camera outputs the depth information of the area. After fusion by the industrial control computer, the following results are obtained: obstacle type is rigid protrusion, geometric dimensions (length 100 mm, width 80 mm, height 15 mm), spatial location (0.6 m from the front of the robot, located in the central area of ​​the brush roller axial projection), and projection range on the brush roller axial direction is [460 mm, 540 mm] (with the left end of the brush roller as the 0 point, total length 1000 mm).

[0040] Step 2: Calculate the collision risk factor The brush roller is currently rotating at 300 rpm, with a torque current of 0.8A (0.5A under normal no-load conditions), and the robot's travel speed is 0.2 m / s. The industrial computer calculates: Obstacle height 15mm vs brush roller elastic compression allowance 10mm → difference 5mm (positive value, indicating exceeding the limit); Obstacle travel length 100mm vs robot braking distance (approximately 50mm based on current speed) → Ratio 2.0; The instantaneous rate of change of torque current (from 0.5A to 0.8A in the last 0.1s, a change rate of 3A / s). Using predefined fuzzy logic rules (the membership function is triangular, and the rule table is set empirically), the output collision risk coefficient is 0.85 (the threshold is set to 0.6), indicating that obstacle avoidance is required.

[0041] Step 3: Generate 3D obstacle avoidance trajectory The obstacle is 15mm high, and its slope (approximately 90° to the side of the junction box) exceeds the second slope threshold (60°). Therefore, a smaller second acceleration value (0.5m / s²) is used. 2 This is used for the uniform acceleration segment of the lifting curve. S-shaped velocity curve parameters: Target lifting height = obstacle height + safety margin 5mm = 20mm; Lifting start time = 0.2m from the obstacle (based on the current speed of 0.2m / s, it needs to be initiated 1 second in advance). Axial offset: Target offset = projection interval width 80mm + safety margin on each side 10mm = 100mm; Detecting axial free areas on both sides: 460mm free on the left, 440mm free on the right, selecting a 100mm offset in the left direction. The generated offset curve is uniform acceleration-uniform speed-uniform deceleration, with a maximum speed of 20mm / s.

[0042] Step 4: Timing Execution Calculate obstacle shape parameters: The ratio of obstacle height to brush roller diameter = 15mm / 120mm = 0.125; The ratio of the axial projection width of the obstacle to the total length of the brush roller = 80mm / 1000mm = 0.08; Preset threshold: First threshold (upper limit of elongated height ratio): 0.2; Second threshold (lower limit of width ratio of flat and elongated shape): 0.3; Third threshold (lower limit of height ratio for tall structures): 0.4; Fourth threshold (upper limit of height-to-width ratio): 0.1; Judgment logic: The height ratio is 0.125 < 0.2, but the width ratio is 0.08 < 0.3 (not meeting the "greater than the second threshold"), so it does not belong to the elongated shape.

[0043] The height ratio is 0.125 < 0.4 (which does not meet the requirement of "greater than the third threshold"), therefore it does not belong to the tall shape.

[0044] Therefore, this obstacle is in an intermediate state, and the control lifting and axial offset are activated simultaneously.

[0045] When the robot moves to a distance of 0.2m from the obstacle, the brush roller begins to rise in an S-shaped lifting curve, while the axial slide shifts to the left. The acceleration during the uniform acceleration phase of the lift is set to the second acceleration value of 0.5m / s². 2 (Due to the obstacle's slope being greater than 60°), the target was raised by 20mm and deviated by 100mm. After 1 second, the brush roller height reached 20mm and the axial offset reached 100mm. At this point, the brush roller's axial projection range changed from the original [460mm, 540mm] to [360mm, 440mm], which did not overlap with the junction box's projection range [460mm, 540mm]. Furthermore, the bottom surface of the brush roller exceeded the top of the junction box, successfully achieving three-dimensional obstacle avoidance.

[0046] Step 5: Override and Reset The robot continues moving forward, and visual inspection confirms that the brush roller has passed the junction box (the obstacle appears behind the camera's field of view). The surface flatness behind is checked: flat, with no remaining obstacles; axial projection area clearance: ample clearance on the left. A segmented reset is performed: first, the height is restored to 80% of the initial working height (0mm), i.e., 16mm, at a first reset speed (30mm / s), taking 0.53s; a 0.2s pause is taken, and visual inspection is performed again to confirm no abnormalities; then, a second reset speed (10mm / s) is used to completely reset to 0mm, while simultaneously resetting axially to 0mm at 20mm / s, both synchronously. The total reset time is approximately 1.5s.

[0047] Step 6: Knowledge Base Recording The characteristics of the junction box (type, size range [10-20mm high, 70-90mm wide], position mode "component center") and the 3D obstacle avoidance trajectory parameters used in this case (raise the target by 20mm, offset by 100mm, and start simultaneously) are stored in the local knowledge base. When encountering similar junction boxes in the future, this template can be directly called, and the offset can be fine-tuned online (based on the actual projection width).

[0048] Scenario 2: Multiple consecutive low protrusions (such as component borders, 5mm high, continuous axially) When the robot encounters the edge of a component joint, the edge extends beyond a preset length threshold (e.g., 300mm) in the direction of travel, but its height is only 5mm (less than the 10mm elastic compression allowance of the brush roller). Collision risk factor calculation: height difference - 5mm (negative value). The risk factor is below the threshold, obstacle avoidance is not triggered, and the brush roller directly crushes through (the compression allowance is sufficient, and there will be no damage).

[0049] Scenario 3: Multiple adjacent junction boxes (150mm spacing) The two junction boxes are 150mm apart. After the brush roller passes the first junction box, the distance between adjacent obstacles (150mm) is compared with the minimum distance required for axial offset reset (preset to 200mm). 150 < 200, therefore, no axial reset is performed after passing the first junction box, maintaining the current left offset of 100mm. Continuing forward, the axial projection area of ​​the second junction box still does not overlap with the current brush roller position (because it has already been offset), so no further offset is needed, and it passes directly. Reset is performed only after passing two junction boxes consecutively and there are no new obstacles behind. This mechanism avoids two resets, saving approximately 2 seconds.

[0050] Scenario 4: Handling Abnormal Situations Suppose that during obstacle avoidance, the depth camera generates noise due to strong light reflection, resulting in insufficient lifting height (actually only 12mm, while the target is 20mm). The position sensor detects that the actual displacement deviates from the target trajectory beyond the allowable range (±3mm), triggering anomaly handling: immediate emergency stop, the brush roller reverses to its initial position at an emergency reset speed (50mm / s), the robot pauses its movement, and the industrial control computer sends an alarm message (including GPS location, anomaly type "insufficient lifting," and timestamp) to the remote maintenance platform. After remote inspection, maintenance personnel can command the robot to continue cleaning or intervene manually.

[0051] Scenario 5: Adaptive switching of cleaning modes During obstacle avoidance (from the start of lifting to the end of complete reset), the industrial control computer simultaneously reduces the brush roller speed from 300 rpm to 180 rpm (60%), while increasing the robot's travel speed from 0.2 m / s to 0.28 m / s (40%) to shorten the total obstacle avoidance time. After obstacle avoidance is completed, visual confirmation confirms that the obstacle has been cleared and the area behind is flat, and the brush roller speed is gradually restored: the acceleration is preset at 200 rpm / s, linearly increasing to 300 rpm; the travel speed is similarly restored gradually. This process avoids current surges.

[0052] III. Parameter Preset Examples To facilitate implementation, a set of typical preset parameters are provided (which can be adjusted according to actual conditions): First slope threshold: 30°, second slope threshold: 60°; first acceleration value: 1.0 m / s² 2 Second acceleration value: 0.5 m / s² 2 .

[0053] First threshold (height / diameter ratio): 0.2, second threshold (axial projection width / brush roller length): 0.3, third threshold: 0.4, fourth threshold: 0.1.

[0054] Safety margin: 5~10mm.

[0055] Minimum distance for axial offset reset: 200mm.

[0056] Segmented reset ratio: 80%, first reset speed: 30mm / s, second reset speed: 10mm / s.

[0057] During obstacle avoidance, the brush roller rotation speed is 50% to 70% of the normal rotation speed, and the speed increment is linearly adjusted according to the length of the obstacle avoidance path, with a maximum not exceeding 150% of the normal speed.

[0058] Soft start acceleration: 200 rpm / s (speed), 0.05 m / s² 2 (Speed ​​of travel).

[0059] IV. Implementation Results Verification Using the configuration and method described above, the robot ran continuously for 100 cycles on a test track containing a junction box (15mm high), a frame (5mm high), and simulated raised debris (25mm high, rigid). Statistical results showed that brush roller wear was significantly reduced compared to the no-obstacle-avoidance solution, with zero jamming failures. The average obstacle avoidance time (including reset) was 2.1 seconds, and the cleaning coverage was close to 100% (with minimal temporary blind spots caused by lifting). After 10 runs of the knowledge base, subsequent obstacle avoidance response time was reduced by nearly 30% (no need to repeatedly calculate the trajectory; templates were directly called for fine-tuning).

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A brush roller adaptive lifting control method based on obstacle recognition on photovoltaic module surface, characterized in that, include: Identify the type, geometry, spatial location, and projection range of obstacles along the brush roller axis; Obtain the rotational speed, torque current, and travel speed of the sweeping robot to calculate the collision risk coefficient; When the collision risk coefficient exceeds the threshold, a three-dimensional obstacle avoidance trajectory is dynamically generated based on the height and slope of the obstacle, the projection range, and the elastic compression margin of the brush roller; the three-dimensional obstacle avoidance trajectory includes a variable speed lifting curve in the lifting direction and a lateral offset curve in the brush roller axis. Before contacting the obstacle, the brush roller is controlled to lift and axially offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory. The height of the brush roller exceeds the height of the obstacle and the axial position of the brush roller moves out of the axial projection range of the obstacle. After confirming that the obstacle has been passed, the brush roller is controlled to return to the initial working height and axial position according to the surface flatness behind the obstacle and the empty space of the axial projection range. The feature parameters and 3D obstacle avoidance trajectory parameters of the current obstacle are recorded in the local knowledge base. Based on the local knowledge base, when similar obstacles are subsequently identified, the corresponding obstacle avoidance trajectory parameters are directly called for prediction and strategy adjustment.

2. The method according to claim 1, characterized in that, The obstacle identification process includes: using at least one RGB camera and one depth camera, based on the fusion of image semantic segmentation results and depth information, to identify the type, geometric size, spatial location, and projection range of the obstacle along the brush roller axis; the obstacle types include at least: photovoltaic module frames, junction boxes, dust protrusions, bird droppings, and broken or warped fragments of the module; when identified as a flexible obstacle, the calculation weight of the collision risk coefficient is reduced; when identified as a rigid protruding obstacle, the calculation weight of the collision risk coefficient is increased.

3. The method according to claim 1, characterized in that, The collision risk factor is determined by a comprehensive assessment of the following factors: The difference between the actual height of the obstacle and the elastic compression margin of the brush roller, the ratio between the length of the obstacle along the direction of travel of the sweeping robot and the current braking distance of the sweeping robot, and the instantaneous rate of change of the brush roller torque current per unit time. The above factors are assigned different evaluation levels, and a collision risk coefficient between 0 and 1 is output through fuzzy logic rules. Among them, the instantaneous change rate of torque current is used as an input of fuzzy logic. The larger the value, the higher the output collision risk coefficient.

4. The method according to claim 1, characterized in that, The variable speed lifting curve is an S-shaped velocity curve, which includes a uniform acceleration segment, a uniform speed segment, and a uniform deceleration segment. The acceleration of the uniform acceleration segment is dynamically adjusted according to the change of the obstacle slope: when the obstacle slope is less than the first slope threshold, the first acceleration value is used; when the obstacle slope is greater than the second slope threshold, the second acceleration value, which is less than the first acceleration value, is used.

5. The method according to claim 1, characterized in that, The target offset of the lateral offset curve is at least equal to the sum of the width of the projection range of the obstacle on the brush roller axis and the safety margin reserved at both ends of the brush roller, and the offset direction is selected as the direction of the nearest axial free area; if the width of the axial free area on both sides of the brush roller is less than the preset minimum value, only lifting to avoid the obstacle is performed and no axial offset is performed.

6. The method according to claim 1, characterized in that, The control brush roller performs lifting and axial offset simultaneously or sequentially according to the three-dimensional obstacle avoidance trajectory, including: dynamically selecting the execution order of lifting and axial offset according to the shape characteristics of the obstacle; When the ratio of the height of the obstacle to the diameter of the brush roller is less than the first threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is greater than the second threshold, the brush roller is controlled to first perform axial offset and then lift. Conversely, when the ratio of the height of the obstacle to the diameter of the brush roller is greater than the third threshold, and the ratio of the projected width of the obstacle in the axial direction to the total length of the brush roller is less than the fourth threshold, the brush roller is controlled to first perform lifting and then axial offset. For obstacles whose height-to-diameter ratio and axial projection width-to-brush roller length ratio are both within the intermediate threshold range, the lifting and axial offset are activated simultaneously.

7. The method according to claim 1, characterized in that, The control of resetting the brush roller to its initial working height and axial position includes: using segmented reset, firstly restoring the height to a preset ratio of the initial working height at a first reset speed, pausing the reset action and visually confirming again that the surface flatness behind the obstacle and the absence of residual obstacles within the axial projection range are achieved, and then fully resetting the brush roller to its initial working height and axial working position at a second reset speed; wherein, the first reset speed is greater than the second reset speed; and simultaneously, the axial reset and height reset are synchronized in time.

8. The method according to claim 1, characterized in that, When the length of the obstacle in the direction of the cleaning robot's travel exceeds a preset length threshold, and there are flat areas on both sides of the obstacle that allow the cleaning robot to pass, the robot's travel path is further adjusted while the brush rollers are controlled to lift and axially deflect to avoid the obstacle. The adjustment of the travel path includes: planning a local arc trajectory that bypasses the obstacle. While the cleaning robot is walking on this trajectory, the brush rollers remain in a lifted and axially deflected state. The original straight travel path is restored after the cleaning robot has completely passed the obstacle.

9. The method according to claim 1, characterized in that, The local knowledge base stores multiple obstacle feature templates and their corresponding optimal three-dimensional obstacle avoidance trajectory parameters. The feature templates include the type of obstacle, typical geometric size range, and common position distribution patterns. When it is found that the matching degree between the current obstacle and a certain feature template exceeds a preset threshold, the optimal lifting start time, target lifting height, axial offset direction, and offset amount stored in the template are directly called as initial parameters. During the execution process, online fine-tuning is performed based on the actual detected brush roller torque current and contact feedback. The fine-tuned parameters are used to update the template.

10. The method according to claim 1, characterized in that, When visual detection identifies multiple obstacles on the same photovoltaic module surface, the collision risk coefficient of each obstacle is calculated sequentially according to the direction of travel of the cleaning robot. If the distance between adjacent obstacles is less than the minimum distance required for the brush roller to reset its axial offset, the brush roller is controlled not to perform axial reset after passing the first obstacle, but to maintain the current axial offset and continuously pass subsequent obstacles. If the distance between adjacent obstacles is greater than or equal to the minimum distance, the brush roller is controlled to reset after passing the current obstacle, and then a three-dimensional obstacle avoidance trajectory is regenerated for the next obstacle.

11. The method according to claim 1, characterized in that, It also includes an abnormal handling process: if any of the following abnormalities are detected during the lifting or axial offset process: the brush roller torque current exceeds the safety threshold, visual detection feedback indicates that the brush roller is still in obvious contact with the obstacle, or the actual displacement of the brush roller is detected to deviate from the target three-dimensional obstacle avoidance trajectory beyond the allowable range, an emergency stop is immediately executed, and the brush roller is controlled to reset to the initial position in reverse order at an emergency reset speed, and an alarm message containing the obstacle position, abnormality type and timestamp is sent to the remote terminal; After an emergency stop, the cleaning robot pauses its movement and awaits remote instructions or human intervention.

12. The method according to claim 1, characterized in that, During the lifting and axial offset obstacle avoidance of the brush roller, the rotational speed of the brush roller is simultaneously reduced to a preset ratio of the normal rotational speed. At the same time, the traveling speed of the cleaning robot is increased according to the length of the obstacle avoidance path. After obstacle avoidance is completed, visual inspection confirms that the brush roller has passed the obstacle and the rear surface is flat. Then, the rotational speed of the brush roller and the traveling speed of the cleaning robot are gradually restored to normal operating parameters. The speed restoration process adopts a slow start method, where the acceleration of the slow start is a preset fixed value or a curve that increases linearly with time.

Images