Helicopter composite material product polishing quality evaluation method and system
By combining optical tracking and AI vision with dynamic self-calibration and explosion-proof laser devices, along with multimodal sensors and edge computing, the accuracy and efficiency issues of composite material grinding quality inspection in dusty environments have been solved, achieving efficient and reliable grinding quality assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies cannot efficiently and safely detect the grinding quality of composite materials in dusty environments, and lack dynamic calibration and closed-loop feedback, resulting in unstable detection accuracy and low efficiency.
Dynamic self-calibration is achieved by using optical tracking and AI vision. Combined with explosion-proof laser devices and multimodal sensors, real-time roughness assessment is performed through edge computing, and process compensation is achieved using particle swarm optimization algorithms.
It achieves high-precision and high-efficiency grinding quality assessment in dusty environments, shortens inspection time, improves reprocessing qualification rate, and enhances process consistency and service reliability.
Smart Images
Figure CN121649863A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of industrial automation testing and intelligent manufacturing technology, and in particular relates to a method and system for evaluating the grinding quality of helicopter composite material products. Background Technology
[0002] With the widespread application of aerospace composite materials in main load-bearing structures such as helicopter rotor blades and fuselage panels, the surface grinding quality directly affects subsequent painting, bonding, assembly, and service performance. Traditional grinding quality inspection mainly relies on manual visual inspection or offline sampling measurement, which has the following prominent problems: 1. Low inspection efficiency: The inspection time for a single piece is usually >2 hours, which cannot meet the production line cycle time; 2. Inconsistent accuracy: Human judgment is highly subjective, and Ra error is often > ±0.5μm; 3. Poor environmental adaptability: High dust concentration in the grinding workshop (>30 g / m³), failure rate of ordinary sensors >15%; 4. Lack of closed-loop control: Detection and process are disconnected, with a feedback cycle of ≥2 weeks; 5. Difficulty in multi-device collaboration: There is a lack of a unified coordinate system among the robot, sensors, and grinding head, resulting in positioning drift > ±0.3mm.
[0003] In recent years, although there have been documents on laser scanning, visual recognition, force-controlled grinding, etc., none of them have simultaneously solved the problems of dynamic deformation compensation, dust explosion prevention, online closed-loop, and full-surface high-density detection.
[0004] In existing technologies, a multi-sensor fusion-based aerospace composite material grinding quality inspection system employs a combined measurement system of laser displacement sensors and vision cameras. It establishes a sensor coordinate system mapping relationship through static calibration and utilizes Kalman filtering to achieve multi-sensor data fusion. However, this system's calibration process relies on a pre-set calibration plate, making it unable to adapt to coordinate system drift in dynamic working scenarios (error > ±0.3mm). Roughness detection uses contact probes, which are susceptible to dust interference and have a measurement speed below 0.5m / s. Furthermore, it lacks an integrated explosion-proof design, failing to meet the safety inspection requirements of a dusty environment in helicopter rotor blade workshops.
[0005] A robotic adaptive grinding device and control method for rotor blades are disclosed. The method utilizes a six-axis robot equipped with a force-controlled grinding head, adjusts the grinding path based on real-time contact force feedback, and employs a point cloud matching algorithm for surface localization. However, this method relies on offline programmed trajectories for localization, resulting in a localization response delay >500ms in dynamic environments. Quality assessment depends solely on grinding force threshold determination, lacking quantitative analysis of surface morphology (such as roughness parameter Ra). The separate design of the detection device and grinding head leads to an increase in operation interruption rate exceeding 20%.
[0006] An online measurement device for helicopter rotor blade surface roughness based on line laser technology integrates a line laser sensor at the end effector of a robot. It generates surface contour data through scanning trajectories and calculates roughness parameters using a moving average filtering algorithm. However, this device's measurement algorithm does not comply with the GB / T 1031 standard, and its sampling length is fixed at 5mm, making it unsuitable for dynamic requirements with Ra 3.2~6.4μm. Furthermore, the sensor's protection rating is only IP54, resulting in a failure rate >15% in dusty environments. Data processing relies on host computer computation, with a single-point detection cycle >2s, failing to meet the real-time requirements of online detection.
[0007] Therefore, there is an urgent need for a grinding quality assessment system and method that is suitable for dusty environments, supports dynamic calibration, has closed-loop feedback, and can achieve high-efficiency online evaluation. Summary of the Invention
[0008] To address the limitations of existing technologies in evaluating grinding quality in dusty environments, supporting dynamic calibration, providing closed-loop feedback, and enabling high-efficiency online assessment, this invention provides a method and system for evaluating the grinding quality of helicopter composite materials. The system achieves dynamic self-calibration through optical tracking and AI vision, constructing a unified coordinate system at the ±0.05mm level. An Expxb IIIC T135 D explosion-proof laser device is used to acquire surface contours online in dusty environments. Combining a three-stage Ra algorithm and PSO optimization, high-speed evaluation and process closed-loop feedback are achieved at 0.8s / point. The technical solution is as follows: Firstly, a method for evaluating the grinding quality of helicopter composite material products is provided, including: S1: A unified coordinate system is established for the robot, line laser sensor and grinding head through a dynamic self-calibration system, so that the repeatability of positioning between multiple devices can reach ±0.05mm; S2: Based on the multimodal sensing resource scheduling module, the robot is driven to trigger detection along the grinding path using a preset equal-density sampling strategy; S3: Surface contour data is collected online using an explosion-proof embedded roughness detection device, with a single-point sampling period of ≤0.1s; S4: Through the edge computing unit, run the three-stage Ra calculation model, sequentially perform Gaussian filtering, least squares midline fitting and arithmetic mean deviation calculation based on GB / T1031, and output the surface roughness within ≤50ms; S5: If Ra exceeds the preset range of 3.2μm–6.4μm, the edge computing unit calls the particle swarm optimization algorithm to generate process compensation parameters within ≤20 iterations and feeds them back to the robot to perform local reprocessing; S6: Upload the detection and process compensation data to the MES system through the communication interface to form a closed loop of quality traceability.
[0009] Optionally, the multimodal sensing resource scheduling module uses a surface curvature κ > 0.05 mm. -1 As a threshold, the line laser sensor and vision camera are activated preferentially, and the single-point detection time is ≤0.8s.
[0010] Optionally, the filtering stage of the three-stage Ra calculation model adopts a Gaussian filter with a cutoff wavelength of λc=2.5mm, the midline fitting stage adopts a least squares algorithm with ≤5 iterations, and the arithmetic mean deviation formula adopted in the Ra calculation stage is Ra=(1 / n)Σ|zi|.
[0011] Optionally, the particle swarm optimization algorithm uses the grinding speed v and the contact force F as optimization variables, and compensates according to the mapping model Ra=k1·v-0.8+k2·F1.2 to make the reprocessing qualification rate ≥94%.
[0012] In a second aspect, a system is provided for evaluating the grinding quality of helicopter composite material products according to any of the methods described in the first aspect, comprising: The dynamic self-calibration module is used to establish a unified coordinate system for the robot, the line laser sensor, and the grinding head; An explosion-proof embedded roughness detection device is used to collect surface contour data online. The device is equipped with a line laser sensor, a Z-axis electric cylinder, an Ex explosion-proof housing and a pneumatic dustproof door, and the load of the whole machine is <500g. The multimodal sensing resource scheduling module is used to drive the robot to trigger detection along the grinding path using a preset equal-density sampling strategy and dynamically allocate sensor resources according to the curvature threshold. Edge computing units are used to run the three-stage Ra computation model and particle swarm optimization algorithm; The communication interface is used to feed back process compensation parameters to the robot and upload them to the MES system via the EtherCAT bus.
[0013] Optionally, the dynamic self-calibration system includes: An optical tracker is used to acquire the coordinates of target spheres arranged at the four corners of the fixture at a frequency of ≥200Hz. The video stream AI module is used to identify the feature holes of the robot flange in real time and calculate the initial rotation matrix R0 and translation matrix T0; A weighted least squares optimizer is used to converge to a calibration error of ≤0.02° and ≤0.05mm within ≤30s.
[0014] Optionally, the line laser sensor has a Z-axis range of 8 mm, a resolution of 0.8 μm, a sampling frequency of 2000 Hz, and IP67 protection is achieved through fiber optic isolation transmission.
[0015] Optionally, the Z-axis electric cylinder has a stroke of 50mm, a repeatability of ±0.01mm, and is connected to the robot end flange via an M4 interface.
[0016] Optionally, the total time taken by the edge computing unit to calculate Ra and PSO optimization is ≤80ms, satisfying the online closed-loop delay <0.5s.
[0017] Optionally, the Ex explosion-proof housing meets Ex pxb IIIC T135 D, and can operate continuously in an environment with a dust concentration >30g / m³ and a failure rate <0.5%.
[0018] The system supports high-density detection of 9 points per square meter, which is 3 times higher than the traditional single-point traversal, and the CPK is ≥1.67.
[0019] The communication interface supports the OPC UA protocol, enabling bidirectional interaction with the MES system to achieve quality data traceability and process parameter download.
[0020] The beneficial effects of this invention are at least as follows: This invention systematically solves the industry problem of quality control in grinding aerospace composite materials by combining intelligent dynamic calibration, highly reliable detection devices, efficient evaluation algorithms, and closed-loop process control, thus promoting the manufacturing of core helicopter components towards full-process unmanned and high-precision manufacturing.
[0021] 1. Fully Automated Dynamic Calibration: A Breakthrough in Both Accuracy and Efficiency: This invention employs a dynamic mutual calibration technology that integrates an optical sensing system with video stream AI. It constructs a global coordinate system using a large-scale measurement field (covering 50m³) and optimizes the rotation matrix (error ≤0.02°) and translation matrix (accuracy ±0.05mm) based on a weighted least squares algorithm, enabling real-time collaborative positioning of multiple devices. Compared to traditional static calibration techniques, calibration time is reduced from 2 hours to 5 minutes, improving efficiency by 400%. It also supports device drift compensation in dynamic environments, achieving a positioning accuracy of ±0.05mm (compared to ±0.3mm with traditional techniques), laying a spatial reference foundation for high-precision grinding and inspection.
[0022] 2. High-Reliability Detection in Dust-Proof Environments: Combining Safety and Performance. Addressing the harsh working conditions of helicopter composite rotor blade grinding workshops with high dust concentrations (>30g / m³), this invention designs an ExpxbIIIC T135 D explosion-proof certified embedded detection device. It integrates a micro-iridium laser sensor (0.8μm resolution), an IP67-rated dustproof housing, and a pneumatic dustproof door (opening / closing time <0.5s). Through fiber optic isolated signal transmission (withstanding 2500V AC) and redundant circuit design, the device reduces the failure rate to <0.5% in dusty environments (compared to >15% for traditional equipment), while maintaining an 8mm Z-axis range and 0.8μm resolution, ensuring high accuracy and reliability for complex curved surface detection.
[0023] 3. High-efficiency intelligent detection, revolutionizing density and speed: A multimodal sensor resource scheduling algorithm driven by a central processing unit enables coordinated control of a 6-axis robot, line laser sensor, and vision camera (response latency <80ms). A curvature threshold activation strategy prioritizes scanning high-curvature areas (>0.05 mm⁻¹). Combined with dynamic sampling length technology (intelligent switching from 2.5 to 12.5 mm), single-point detection speed reaches 0.8s (traditional technology 2s / point), improving detection efficiency by 150%. It also supports high-density sampling of 9 points per square meter (traditional technology ≤3 points / square meter), upgrading the quality monitoring mode from discrete sampling to full surface coverage.
[0024] 4. Real-time feedback in closed-loop process, leading to a significant improvement in quality controllability: based on a roughness-process parameter mapping model (Ra=k1v⁻). 0 · 8 Using the +k2F¹·² algorithm and the Particle Swarm Optimization (PSO) algorithm, this invention pioneers a closed-loop "detection-evaluation-compensation" mechanism. When the Ra value is detected to be out of tolerance, the system generates a trajectory compensation command (such as adjusting the grinding head feed speed by ±10%) within 0.5s, and optimizes the compensation path through the PSO algorithm within 20 iterations, thereby increasing the reprocessing pass rate by 40% and achieving a key quality indicator CPK≥1.67 (compared to ≤1.33 in traditional processes), significantly improving the process consistency and service reliability of helicopter rotor blades. Attached Figure Description
[0025] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of dynamic self-calibration (optical target ball + video stream AI mutual calibration). Figure 3 This is a flowchart of multimodal data acquisition and processing. Figure 4 This is a flowchart for process closed-loop feedback and quality traceability. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0027] The features and illustrative embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific setups and methods set forth below, but covers any improvements, substitutions, and modifications to structures, methods, and devices without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description to avoid unnecessarily obscuring the invention.
[0028] It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other, and the various embodiments can be referenced and cited from each other.
[0029] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.
[0030] This invention provides an online quality assessment method and system for automated grinding of components such as helicopter composite blades. It is applicable to surface roughness detection, shape and position accuracy assessment, and process closed-loop optimization of carbon fiber / glass fiber reinforced resin matrix composite parts and their coated curved surfaces.
[0031] This invention achieves dynamic self-calibration through optical tracking and AI vision, constructing a unified coordinate system at the ±0.05 mm level. It employs an Ex pxb IIIC T135 D explosion-proof laser device to acquire surface contours online in dusty environments. Combining a three-stage Ra algorithm and PSO optimization, it achieves high-speed evaluation and process closed-loop feedback at 0.8 s / point. This solves the problems of low efficiency, poor environmental adaptability, and lack of closed-loop control in traditional inspection methods, reducing single-piece inspection time by 80%, achieving Ra error ≤ ±0.15 μm, a reprocessing pass rate ≥94%, and CPK ≥1.67. It is suitable for intelligent grinding production lines for composite material components such as helicopter rotor blades and panels.
[0032] Combination Figure 1 Online grinding quality assessment solution: The online grinding quality assessment system consists of a line laser sensor, an industrial computer, and an explosion-proof housing. The probe itself is IP65 rated, ensuring it will not come into contact with dusty environments. It is treated with appropriate explosion-proof measures according to national or enterprise standards, achieving an explosion-proof rating of Ex pxb IIIC T135 D, making it suitable for dusty environments.
[0033] The system consists of a sensor and data acquisition module, a data transmission and processing module, a probe, and a telescopic electric cylinder, wherein: 1. In the sensor and data acquisition module, the sensor is a line laser sensor with a Z-axis range of 52.5mm from the start point to the midpoint, 56.5mm from the midpoint, and 60.5mm from the end point, and a range height of 8mm. The X-axis range has a start point of 9.4mm, a midpoint of 10mm, and an end point of 10.7mm, and a measurement frequency of 2000Hz.
[0034] 2. In the data transmission and processing module, the hardware part of the data processing module adopts an industrial control computer. This industrial control computer is equipped with a processor and graphics card with considerable processing power. The overall size is 135×115×36mm, and it is powered by a 19VDC 65W power supply.
[0035] 3. The bare weight of the probe is 440g (excluding cables). The probe has two aviation connector cables with a diameter of 6mm. One cable is connected to a 24V power supply and is a Φ6 air supply cable. The other cable transmits data via a network port and uses a trigger-based data acquisition method. The probe is manufactured in accordance with relevant explosion-proof requirements.
[0036] 4. The Z-axis telescopic electric cylinder has a stroke of 50mm and is used to extend the probe and explosion-proof housing downwards to the rated detection distance. The telescopic cylinder is used to open and close the dust door. The explosion-proof housing and probe are mounted on one side of the grinding actuator via a connecting plate, without interfering with the grinding head. Detection is performed by adjusting the robot position during operation. The Z-axis electric cylinder is connected to the outer plate of the grinding actuator with three M4×30 hex socket screws. The connecting plate is mounted on the slide of the Z-axis electric cylinder with six M4×8 hex socket screws and can slide up and down via the cylinder. The telescopic cylinder is connected to the mounting plate with M8 screws. The explosion-proof housing and probe are mounted on the connecting plate with four M8×8 hex socket screws and slide up and down with the connecting plate.
[0037] When the grinding head is working, the Z-axis electric cylinder must be in the retracted state to ensure an absolute safe distance. When online evaluation is required, the grinding head must be stopped. At this time, the grinding head will report the distance between itself and the machine surface. Simultaneously, the measurement software will calculate how far the Z-axis electric cylinder needs to descend and control the Z-axis cylinder to extend downwards to ensure that the probe is at the working distance and that the detection safety distance is maintained.
[0038] After the area polishing is completed, the evaluation task is initiated. The robot retraces the polishing trajectory, sampling nine points at equal intervals per square meter. During polishing, the robot samples nine points at equal intervals per square meter, starting from the polishing landing point, recording the distance between the polishing head and the robot body. This data is saved locally for later use in measurements. During the inspection process, the robot executes the polishing trajectory without changing its posture. After the probe reaches the sampling point based on the recorded data, the inspection task is completed according to the above process, and the inspection file is generated and transmitted to the central control unit. The central control unit transmits the distance information of the robot's end effector from the surface at the sampling point to the industrial control computer connected to the probe via a network cable. Simultaneously, the industrial control computer calculates the required downward distance and transmits it to the electric cylinder, which moves the probe. After the measurement is completed, all data is saved locally, and the central control unit can access the measurement data at any time.
[0039] Implementation of online inspection and evaluation of grinding quality: 1. The roughness evaluation algorithm is used to convert the profile undulation data into roughness information. The steps are as follows: Step 1: Specify the sampling length, which is a baseline for evaluating surface roughness. This can suppress and reduce the influence of surface ripples on the surface roughness measurement results. Step 2: Filter the collected contour information and determine a baseline for quantitatively evaluating the surface roughness contour. Usually, the least squares midline or the arithmetic mean midline of the contour can be selected. Step 3: Within a sampling length range, calculate the arithmetic mean of the absolute values of the ordinates of each point on the evaluated profile to the profile centerline.
[0040] Step 4: Following the above evaluation method, after the area grinding is completed, perform surface roughness testing and output a grinding pass report based on the sampling results. If the grinding result fails the test, and the surface roughness does not meet the range of Ra3.2μm to Ra6.4μm, the report will indicate the non-compliant sampling points.
[0041] Step 5: After the area grinding is completed, the robot blows and vacuums the ground surface, and the grinding inspection device is activated. The grinding quality assessment workflow is as follows: a) The grinding quality assessment system sends a command to the central control system, informing the system that it is ready and can begin roughness measurement; b) The central controller, based on the coordinates and the offline programmed trajectory, controls the robot to the current measurement point and sends the current measurement coordinates and the arrival command to the polishing evaluation system; c) The grinding quality assessment system opens the front end of the explosion-proof housing, the line laser probe starts measuring, and transmits the collected data values to the industrial control computer; d) After receiving the feedback data value, the industrial control computer inputs it into the relevant software modules, quickly calculates the roughness of the current area, compares it with the theoretical value of the aircraft surface after polishing, and gives the evaluation result; e) Save the current detection data, send a command to the main control program to perform the next measurement, and repeat steps a) to d). The measured roughness values are considered acceptable if they fall within the range of Ra3.2-Ra6.4, and unacceptable if they fall outside this range.
[0042] For example, the online evaluation process for the grinding quality of the helicopter main rotor blade root is as follows: Step 1: Deploy an online quality assessment system in the grinding operation environment. This system consists of a 6-axis robot, an optical tracker, a vision-assisted system, and a main control system. 6-axis robot: End flange integrated explosion-proof embedded roughness detection device, weight ≤480g.
[0043] Optical tracker: Four φ19 mm ceramic target spheres are arranged in a 1.2m×0.8m rectangle to construct a 50m³ global coordinate system with a volume accuracy of ±0.05mm.
[0044] Visual assistance: 4 sets, frame rate 200Hz, equipped with 850nm near-infrared stripe light source, used for AI mutual calibration and surface defect re-inspection.
[0045] Main control: cycle time ≤ 10ms; Ra calculation and PSO optimization are completed on the edge industrial computer, single point ≤ 50ms.
[0046] Step 2: Grind the blade root. After completion, execute the dynamic self-calibration process. See... Figure 2 ,include: ① Within 30 seconds of power-on, the optical tracker collects the coordinates of the four target spheres and establishes an absolute coordinate system W.
[0047] ②Real-time identification of feature holes on robot flanges, and calculation of initial R0 and T0.
[0048] ③Weighted least squares optimization (5 iterations) → Repeat positioning accuracy ±0.05 mm, calibration error ≤0.02°.
[0049] ④ Automatic re-inspection every 60 seconds, recalibration is triggered if drift >0.03mm, and no human intervention is required throughout the process.
[0050] Step 3, perform multimodal sensor resource scheduling, see Figure 3 The algorithm is designed as follows: Curvature threshold κT = 0.05 mm⁻¹; When the preview path κ > κT, the PLC prioritizes activating the line laser + camera dual mode, and only retains camera monitoring in the low curvature segment.
[0051] The priority allocation algorithm reduced the time for single-point measurement from 1.2s to 0.35s, improving overall efficiency by 300%.
[0052] Step 4: Execute the working cycle of the explosion-proof detection device. The procedure is as follows: a) The grinding head stops rotating → The robot reports the current TCP distance d from the surface.
[0053] b) The PLC calculates the extension amount ΔZ of the electric cylinder based on d, ensuring that the laser focus falls in the middle of the 8mm range.
[0054] c) Dustproof door cylinder opens at 0.4MPa → electric cylinder descends → linear laser sampling at 2000 Hz × 0.1s → original contour upload.
[0055] d) The electric cylinder retracts → the dust door closes → the robot continues to the next path.
[0056] The complete cycle for a single point is 0.8s (0.1s for sampling + 0.7s for motion and communication).
[0057] Step 5, three-stage real-time Ra calculation, see Figure 4 , ① Gaussian filtering (cutoff wavelength λc = 2.5 mm) eliminates ripple interference; ② Least squares midline dynamic fitting (iterations ≤ 5 times); ③Ra=(1 / n)Σ|zi|, with a single-point calculation cycle of 28ms (e.g., C++ parallel optimization, OpenCV+Eigen). The measured Ra = 4.1 μm, which is within the acceptable range of 3.2–6.4 μm, and the system judged it as "acceptable".
[0058] Step 6, closed-loop feedback and process compensation, see Figure 4 The algorithm is designed as follows: Mapping model: Ra = 0.72·v⁻ 0 · 8 +0.18·F¹·² (v: grinding speed mm / min; F: contact force N).
[0059] When Ra > 6.4 μm, the PSO algorithm (particle number 20, iteration limit 20) outputs the compensation amount as follows: v decreases by 10%, F decreases by 8%.
[0060] The compensation command is delivered to the robot via the EtherCAT bus in less than 0.5 seconds, and the next segment of the trajectory is corrected in real time.
[0061] After one compensation, Ra in the leaf root region decreased to 4.5 μm, and the first-pass yield increased from 92% to 99.2%. Step 7: The test is complete. Collect and record the data.
[0062] Verification has shown that, using the online quality assessment system and method designed in this invention, the single-piece inspection time is reduced from 25 min to 3.8 min compared to traditional offline sampling inspection, the Ra error is reduced from ±0.5 μm to ±0.15 μm, the process feedback cycle is reduced from 2 weeks to 8 hours, and the CPK is increased from 1.33 to 1.67.
[0063] The above description merely illustrates embodiments of the present invention and is quite specific and detailed; however, it should not be construed as limiting the scope of the patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Furthermore, any parts of the present invention not described in detail are conventional techniques.
Claims
1. A method for evaluating the grinding quality of helicopter composite material products, characterized in that, include: S1: A unified coordinate system is established for the robot, line laser sensor and grinding head through a dynamic self-calibration system, so that the repeatability of positioning between multiple devices can reach ±0.05mm; S2: Based on the multimodal sensing resource scheduling module, the robot is driven to trigger detection along the grinding path using a preset equal-density sampling strategy; S3: Surface contour data is collected online using an explosion-proof embedded roughness detection device, with a single-point sampling period of ≤0.1s; S4: Through the edge computing unit, run the three-stage Ra calculation model, sequentially perform Gaussian filtering, least squares midline fitting and arithmetic mean deviation calculation based on GB / T1031, and output the surface roughness within ≤50ms; S5: If Ra exceeds the preset range of 3.2μm–6.4μm, the edge computing unit calls the particle swarm optimization algorithm to generate process compensation parameters within ≤20 iterations and feeds them back to the robot to perform local reprocessing; S6: Upload the detection and process compensation data to the MES system through the communication interface to form a closed loop of quality traceability.
2. The method according to claim 1, characterized in that, The multimodal sensing resource scheduling module prioritizes the activation of line laser sensors and vision cameras with a surface curvature κ > 0.05 mm⁻¹ as the threshold, and ensures that the single-point detection time is ≤ 0.8s.
3. The method according to claim 1, characterized in that, The filtering stage of the three-stage Ra calculation model uses a Gaussian filter with a cutoff wavelength of λc=2.5mm, the midline fitting stage uses a least squares algorithm with ≤5 iterations, and the arithmetic mean deviation formula used in the Ra calculation stage is Ra=(1 / n)Σ|zi|.
4. The method according to claim 1, characterized in that, The particle swarm optimization algorithm uses grinding speed v and contact force F as optimization variables, and compensates according to the mapping model Ra=k1·v-0.8+k2·F1.2 to make the reprocessing qualification rate ≥94%.
5. A system for evaluating the grinding quality of helicopter composite material products according to any one of claims 1 to 4, characterized in that, include: The dynamic self-calibration module is used to establish a unified coordinate system for the robot, the line laser sensor, and the grinding head; An explosion-proof embedded roughness detection device is used to collect surface contour data online. The device is equipped with a line laser sensor, a Z-axis electric cylinder, an Ex explosion-proof housing and a pneumatic dustproof door, and the load of the whole machine is <500g. The multimodal sensing resource scheduling module is used to drive the robot to trigger detection along the grinding path using a preset equal-density sampling strategy and dynamically allocate sensor resources according to the curvature threshold. Edge computing units are used to run the three-stage Ra computation model and particle swarm optimization algorithm; The communication interface is used to feed back process compensation parameters to the robot and upload them to the MES system via the EtherCAT bus.
6. The system according to claim 5, characterized in that, The dynamic self-calibration system includes: An optical tracker is used to acquire the coordinates of target spheres arranged at the four corners of the fixture at a frequency of ≥200Hz. The video stream AI module is used to identify the feature holes of the robot flange in real time and calculate the initial rotation matrix R0 and translation matrix T0; A weighted least squares optimizer is used to converge to a calibration error of ≤0.02° and ≤0.05mm within ≤30s.
7. The system according to claim 5, characterized in that, The line laser sensor has a Z-axis range of 8mm, a resolution of 0.8μm, a sampling frequency of 2000Hz, and IP67 protection through fiber optic isolation transmission.
8. The system according to claim 5, characterized in that, The Z-axis electric cylinder has a stroke of 50mm, a repeatability of ±0.01mm, and is connected to the robot end flange via an M4 interface.
9. The system according to claim 5, characterized in that, The total time for the edge computing unit to calculate Ra and PSO optimization is ≤80ms, satisfying the online closed-loop delay <0.5s.
10. The system according to claim 5, characterized in that, The Ex explosion-proof housing meets Ex pxb IIIC T135D standards and can operate continuously in environments with dust concentrations >30g / m³ with a failure rate <0.5%.