Logistics robot system capable of sorting steel body material defects

Abstract

The invention relates to a logistics robot system capable of sorting steel body material defects, which builds a sorting-logistics-endurance integrated whole-course autonomous system architecture, and integrates a heterogeneous dual-core processor architecture with deep hardware. The two key technologies of a YOLOv11-SG lightweight defect detection model and a lightweight LCC-LCC topology wireless charging system are organically combined, a state machine driven full-process cooperative control logic is matched, a complete autonomous sorting logistics solution is formed, the lightweight defect detection model guarantees independent perception of a terminal, the endurance problem is solved through the wireless charging technology, and the system is simple in structure and convenient to operate. And by combining the multi-module collaborative hardware and process design, the robot can autonomously operate in narrow and small scenes such as a 400mm narrow channel, the dependence on cloud computing and manual intervention is thoroughly eliminated, and a low-cost intelligent upgrading path is provided for medium and small-scale logistics scenes.

Description

Technical Field

[0001] This invention relates to the field of automated control for material sorting and transfer, and specifically to a logistics robot system capable of sorting defects in steel materials. Background Technology

[0002] With the deepening of intelligent manufacturing, the automation and intelligentization of logistics sorting and material transfer have become key links in improving production efficiency. Among them, logistics robots that can autonomously identify and sort defective materials are the core equipment for realizing the intelligent upgrading of warehousing and production lines. The industry generally expects such systems to achieve deep integration of the three major functions of "logistics transfer, defect sorting, and continuous operation" to complete the unmanned operation of the entire process from "perception-decision-execution-recharge".

[0003] However, despite numerous technological explorations in this field in China, the long-term isolated development of various functional modules has resulted in significant technological fragmentation and collaboration barriers. This has made it difficult to truly implement a fully autonomous robot system integrating "logistics + sorting + endurance." This system-level bottleneck is specifically manifested in the following ways: At the system architecture level, sorting (perception) and logistics (execution) functions are disconnected from each other, making it difficult to form an autonomous closed loop. Existing mainstream solutions place complex defect detection algorithms in the cloud or on fixed computing power platforms, with the robot only serving as an execution terminal. This results in low intelligence levels in the robot itself, heavy reliance on network stability, and an inability to achieve integrated "detection and sorting" operations at the terminal. Network latency exists during the identification process, making it difficult to meet the real-time requirements of dynamic sorting. Furthermore, once the network or cloud fails, the entire sorting function becomes ineffective, the robot cannot complete the operation independently, and the system has poor robustness.

[0004] At the core perception level, there is a gap in lightweight detection models suitable for terminal deployment. Existing high-precision algorithm models are large and computationally complex, making them unsuitable for the robot's limited local computing power; while overly simplified models struggle to meet the accuracy requirements for identifying defects in small targets, resulting in significant challenges in achieving high-precision sorting independently at the terminal.

[0005] At the level of continuous operation, traditional energy replenishment methods are seriously mismatched with the needs of autonomous operation. Relying on manual battery swapping or charging at fixed base stations not only interrupts the operation process and increases costs, but also has poor adaptability in application scenarios with limited space, making it impossible to support the robot to achieve uninterrupted continuous operation, which has become a key weakness in the fully unmanned process.

[0006] In summary, the lack of system-level design in existing technologies leads to mutual constraints between logistics, sorting, and battery life, making effective integration into a single robot platform impossible. Therefore, there is an urgent need for an integrated and innovative solution to overcome these collaborative bottlenecks and promote the practical application of truly autonomous logistics sorting robot systems. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a robotic system capable of sorting defects in steel materials. It aims to construct an integrated system with deep collaboration between "intelligent perception, precise execution, and autonomous operation" to achieve fully unmanned operation from task acquisition to autonomous charging, thus breaking through the technological bottlenecks of integration and autonomy.

[0008] To achieve the above objectives, the technical solution adopted by this invention is: a logistics robot system capable of sorting defects in steel materials. The robot system establishes a fully autonomous system architecture integrating sorting, logistics, and endurance, including a perception layer, a decision-making layer, and an execution layer. The perception layer includes a vision subsystem, a gyroscope, and an encoder, responsible for collecting information to provide input for sorting decisions and path planning. The decision-making layer, with a microcontroller at its core, receives data from the perception layer, runs a PID control algorithm and inverse kinematics analysis, and generates robotic arm grasping instructions, chassis movement instructions, and wireless charging control instructions to achieve intelligent decision-making and precise control. The execution layer includes a chassis, a three-degree-of-freedom robotic gripper, and a wireless charging receiver circuit, responsible for executing specific physical actions and feeding back the execution status to the decision-making layer in real time through sensors. The vision subsystem is equipped with a lightweight YOLOv11-SG defect detection model, which can identify and locate materials in real time and perform defect detection. Based on the detection results of the vision subsystem, the decision layer controls the robotic arm to grab qualified materials and then transfers the materials to the rough processing area or temporary storage area according to the preset path. The wireless charging receiver circuit is integrated under the chassis and adopts a single-coil architecture and energy storage capacitor structure. When the decision layer detects that the energy storage capacitor charge is ≤10%, it immediately triggers the autonomous guidance program. The robot autonomously returns to the wireless charging transmitter and completes the alignment. The wireless charging system adopts magnetic coupling resonant technology. The transmitter converts DC into high-frequency alternating current through a full-bridge inverter to drive the transmitting coil to generate an electromagnetic field. After the receiving coil couples the energy, it outputs a stable DC through a rectifier bridge and filter capacitor to charge the energy storage capacitor.

[0009] Furthermore, the YOLOv11-SG defect detection model embeds a C3K2-SG module into the backbone network, and the network has a three-level structure: Backbone network: Feature extraction is achieved through C3K2-SG and C2PSA modules. The C3K2-SG module replaces the original bottleneck Bottleneck to reduce parameter redundancy. Neck network: Employs SPF module and upsampling to achieve multi-scale feature fusion, enhancing the feature representation of defects in small targets; Inspection head: The inspection head outputs the defect category and bounding box coordinates to complete end-to-end inspection.

[0010] Furthermore, the C3K2-SG module replaces the standard Bottleneck in the C3K2 framework. By fusing self-moving-point convolution SMPConv and convolutionally gated linear unit CGLU, it achieves adaptive information filtering and discriminative feature capture. SMPConv shares point positions and independent weight parameters in each channel, and achieves continuous convolution through interpolation. CGLU adds a 3×3 depthwise convolution before the gated branch activation function to build a channel attention mechanism, enhance the interaction between adjacent features, and improve the robustness of the model.

[0011] Furthermore, the transmitter circuit uses a full-bridge H-inverter topology to convert DC power into a high-frequency square wave, and uses an LCC resonant cavity as a high-quality factor bandpass filter. At the same time, the circuit integrates an R32 alloy sampling resistor, a TP181A1 amplifier, and an RC filter network. The power is displayed in real time through a digital tube, and a fan module ensures heat dissipation.

[0012] Furthermore, the receiver circuit integrates the LCC compensation network, synchronous rectifier bridge, and filter capacitor using small-sized 1206 surface-mount components. At the same time, an intelligent protection circuit is integrated on the output side. It senses the charging status in real time through a resistor divider network and uses a Zener diode D3 to form a hardware voltage clamping circuit to strictly limit the sampling voltage to within 3.3V, providing the MCU with absolute overvoltage protection with nanosecond-level response.

[0013] Furthermore, the wireless charging system adopts magnetic coupling resonant technology. The transmitter converts DC into high-frequency alternating current through a full-bridge inverter to drive the transmitting coil to generate an electromagnetic field. After the receiving coil couples the energy, it outputs stable DC through a rectifier bridge and filter capacitor to charge the supercapacitor. The electrical energy output by the wireless charging receiver is stored in the supercapacitor bank and then powered by boost and buck modules to supply power to the drive unit and computing unit respectively.

[0014] Furthermore, the vision subsystem is a Raspberry Pi 5 vision subsystem, including a camera and a QR code scanning module. The QR code scanning module is used to acquire the task code. The microcontroller is an Arduino Portenta H7. The Arduino and the Raspberry Pi communicate bidirectionally via UART serial port. The Raspberry Pi outputs visual positioning data, and the Arduino integrates gyroscope angular position information and dynamically corrects the chassis motion trajectory by optimizing PID parameters.

[0015] Furthermore, the chassis adopts the Mecanum four-wheel system, with the angle between the wheel roller axis and the wheel axis being γ. Omnidirectional movement is achieved through forward and inverse kinematic equations, and "oblique driving + fine-tuning steering" is completed through multi-parameter coordinated adjustment.

[0016] Furthermore, the mechanical gripper is 3D printed using carbon fiber composite material, with polyurethane stripes embedded in the gripper, enabling it to perform combined radial contraction and axial rotation movements; it integrates a PVDF thin film sensor to provide tactile feedback, and dual cameras are set at the center and outer sides to simultaneously complete defect identification and gripping positioning.

[0017] Furthermore, the robotic arm adopts a human-simulated three-degree-of-freedom structure, establishes a kinematic model based on the DH parameter method, smooths the trajectory through cubic spline interpolation, and compensates for position errors through PID closed-loop control to ensure grasping and positioning accuracy.

[0018] Beneficial effects: This invention constructs a fully autonomous system integrating sorting and logistics through a multi-module collaborative hardware architecture, a fully autonomous operation process design, and an autonomous endurance linkage mechanism, enabling autonomous operation throughout the entire process from task acquisition to autonomous charging.

[0019] This invention designs a lightweight defect detection model by embedding a C3K2-SG module into the backbone network of YOLOv11–SG, which integrates self-moving point convolution (SMPConv) and convolutional gated linear unit (CGLU). SMPConv ensures the flexibility of feature extraction with low computational overhead, while CGLU enhances the robustness of the model. The two work together to improve the defect detection capability of small targets, which can be adapted to small robot terminals, ensure independent perception of the terminal, and improve the real-time performance and autonomy of defect detection.

[0020] This invention proposes a high-efficiency, short-time, lightweight wireless charging technology adapted to confined spaces. Based on an LCC-LCC topology, the wireless charging system combines a single-coil receiving architecture with a CC-CV switching strategy to enable robots to autonomously and quickly replenish power in confined spaces, reducing reliance on manual labor.

[0021] This invention forms a complete autonomous sorting and logistics solution through the deep coupling of three core technologies: integrated hardware architecture, lightweight defect detection model, and wireless charging technology. The lightweight defect detection model ensures independent perception at the terminal, and the wireless charging technology solves the problem of battery life. Combined with the multi-module collaborative hardware and process design, the robot can operate autonomously in narrow scenarios such as 400mm narrow channels, completely eliminating the dependence on cloud computing and human intervention, and providing a low-cost intelligent upgrade path for small and medium-sized logistics scenarios. Attached Figure Description

[0022] Figure 1 System architecture diagram of the logistics robot system of this invention; Figure 2 The Mecanum four-wheel system and wheel structure diagram in this invention; Figure 3 Connection diagram of the electronic control system of this invention; Figure 4 Schematic diagram of the wireless charging system of this invention; Figure 5 Overall circuit design of the transmitter of the wireless charging system of this invention; Figure 6 Overall circuit design of the receiver end of the wireless charging system of this invention; Figure 7 This is a schematic diagram of the YOLOv11-SG network structure; Figure 8 Here is a structural diagram of the C3K2-SG model; Figure 9 The logic diagram of the state machine of this invention; Figure 10 This is a schematic diagram of a simulated factory site. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0024] This embodiment provides a logistics robot system capable of sorting defects in steel materials. Through a heterogeneous dual-core processor architecture and deep hardware integration, it organically combines two key technologies: the YOLOv11-SG lightweight defect detection model, specifically designed to address the high computational complexity of traditional models, and the lightweight LCC-LCC topology wireless charging system, aimed at overcoming the integration challenges of traditional charging systems. It is further equipped with state machine-driven full-process collaborative control logic. The YOLOv11-SG model solves the bottleneck of deploying complex algorithms at the terminal, while the lightweight LCC-LCC topology wireless charging system overcomes the integration barriers of the battery life module. Through the integration of these technologies, a unified "perception-decision-execution-battery life" logistics robot system solution is formed, capable of stable operation in real-world industrial scenarios.

[0025] I. System Overall Architecture Integrated collaborative system architecture for practical application: To address the system redundancy caused by the loose coupling of perception, decision-making, and execution modules in traditional solutions, this invention adopts a heterogeneous dual-core processor solution (Arduino Portenta H7 + Raspberry Pi 5) and a customized PCB power network. Through deep integration at the hardware level, the originally independent and complex functions are integrated into a unified low-power, high-real-time control link, providing a stable and efficient physical foundation for the deployment of advanced algorithms on the terminal.

[0026] like Figure 1 As shown, the logistics robot system is designed based on the principle of "layered execution + closed-loop collaboration". It adopts a hardware architecture of "perception-decision-execution" three-layer closed-loop collaboration to ensure deep coupling of logistics, sorting and endurance functions.

[0027] The perception layer consists of a Raspberry Pi 5 vision subsystem (including a camera lens, a QR code scanning module), a gyroscope, an encoder, etc. It is responsible for image acquisition (material defects, QR code task codes), position and attitude perception, and outputs defect detection results and coordinate positioning data to provide information input for sorting decisions and path planning. The vision subsystem is equipped with a lightweight YOLOv11-SG defect detection model, which can identify and locate materials in real time and perform defect detection.

[0028] Decision layer: With Arduino Portenta H7 as the main control core, after receiving data from the perception layer, it generates robotic arm grasping instructions, chassis movement instructions and charging control instructions through PID algorithm and inverse kinematics equation analysis. It is the central hub for realizing intelligent decision-making and precise control.

[0029] The execution layer includes a Mecanum wheel chassis, a mechanical gripper driven by an MG996R servo motor, and a wireless charging receiver. Under the command of the decision-making layer, it performs actions such as path movement, material grabbing / placement, and autonomous charging. At the same time, it provides real-time feedback on the execution status through sensors, forming a closed-loop data link of "perception-decision-execution-feedback" to ensure the spatiotemporal synchronization and coordination of each functional module.

[0030] II. Mechanical System Design 1. Gear train structure: such as Figure 2 As shown, the wheel system adopts the Mecanum four-wheel system. The angle between the axis of the wheel roller and the axis of the wheel is γ. The vehicle coordinate system (XOY) and the wheel coordinate system (xoy) are established. R is the wheel radius and ω is the angular velocity. Omnidirectional movement is achieved through forward and inverse kinematic equations. Multi-parameter coordinated adjustment completes "oblique driving + fine-tuning steering" and adapts to multi-area path switching (such as raw material area → rough processing area).

[0031] 2. System Structure Mechanical claw structure: It is 3D printed using carbon fiber composite material (preferably with an elastic modulus of 150GPa), with polyurethane stripes embedded in the claw, and has the ability to perform combined radial contraction and axial rotation. It integrates a PVDF thin film sensor (sensitivity 10mV / N) to realize tactile feedback, and dual cameras (arranged in the center and outer side of the mechanical claw) to simultaneously complete defect identification and clamping positioning. Robotic arm control: It adopts a human-simulated three-degree-of-freedom structure (bottom rotation joint + upper and lower arm pitch joints), establishes a kinematic model based on the DH parameter method, smooths the trajectory through cubic spline interpolation, and compensates for position errors with PID closed-loop control to ensure grasping and positioning accuracy of ±0.5mm.

[0032] III. Electronic Control System and Wireless Charging System 1. Electronic control system architecture like Figure 3 As shown, the electronic control system achieves unified management of energy, data, and motion control.

[0033] Energy management chain: The electrical energy output from the wireless charging receiver is stored in a supercapacitor bank (12V / 75F, 4 series and 1 parallel structure in this embodiment). One path powers the stepper motor through a 15V boost module, and the other path powers the Raspberry Pi, Arduino and servo motors through a 5V buck module, forming a hierarchical power supply network.

[0034] Data interaction link: Arduino and Raspberry Pi communicate bidirectionally via UART serial port (115200bps). Raspberry Pi outputs visual positioning data, and Arduino integrates gyroscope angular position information and dynamically corrects the chassis motion trajectory by optimizing PID parameters.

[0035] Motion control link: The encoder and stepper motor form a closed loop, and the duty cycle is adjusted by PWM wave drive. Multi-directional movement is achieved in conjunction with the Mecanum wheel, with a response delay of ≤100ms.

[0036] 2. Wireless charging technology To overcome the problems of traditional charging solutions disrupting operational continuity due to their bulky size and stringent alignment requirements, this invention designs a lightweight wireless charging system based on an LCC-LCC topology. Through a simplified design using a single-coil receiving architecture and a relay-type CC-CV fast switching circuit, the system size and complexity are reduced to a level suitable for embedding in robots, achieving an energy conversion efficiency of 88% and a rapid recharging time of 15.4 seconds, making fully unattended operation possible.

[0037] like Figure 4As shown, the overall architecture of the wireless charging system is as follows: It adopts magnetically coupled resonant (MCR-WPT) technology. The transmitter converts DC into high-frequency alternating current through a full-bridge inverter to drive the transmitting coil to generate an electromagnetic field. After the receiving coil couples the energy, it outputs stable DC through a rectifier bridge and filter capacitor to charge the supercapacitor.

[0038] Compensation Topology: Both the transmitter and receiver adopt LCC topology. The resonant parameters (transmitter coil Lp, receiver coil Ls, compensation capacitor Cp / Cf1) are optimized through offline parameter tuning to achieve constant current-constant voltage (CC-CV) output at 350W power. The receiver completes CC-CV mode switching within 1ms through a relay and provides overcharge protection. This lightweight wireless charging design enables the robot to quickly replenish power during work breaks and is perfectly adapted to narrow spaces such as 400mm narrow channels.

[0039] The circuit design of the wireless charging transmitter and receiver is as follows: Figure 5 and Figure 6 As shown: Transmitter: A full-bridge inverter H-bridge is constructed using HYG180N1OLS1P MOSFETs. Voltage and current signals are acquired through R32 alloy sampling resistors and TP181A1 amplifiers. The power is displayed in real time on a digital tube, and a fan module ensures heat dissipation.

[0040] The core of the transmitter circuit of this invention lies in its integrated design of high-frequency inverter and resonant filter, as well as its integrated real-time monitoring mechanism: The circuit uses a full-bridge H-inverter topology to convert DC power into a high-frequency square wave, and then uses a carefully tuned LCC resonant cavity as a high-quality factor bandpass filter to effectively attenuate switching harmonics based on the Fourier analysis principle, thereby obtaining a near-pure fundamental frequency sinusoidal current on the transmitting coil. This "harmonic purification" process is the key to achieving high-efficiency single-frequency power transmission. At the same time, the circuit integrates a high-bandwidth, low-noise current sensing circuit composed of an R32 alloy sampling resistor, a TP181A1 amplifier, and an RC filter network, realizing real-time digital monitoring of transmission power and resonance state, and providing a guarantee for system stability and safety.

[0041] Receiver: An LCC resonant network, an MBR30100CT rectifier bridge, a 470μF filter capacitor, and an LMV321M5 / TR voltage comparator trigger an HK4100F-DC5V-SHG relay to achieve CC-CV mode switching (constant current stage: LCC topology; constant voltage stage: parallel capacitors form a T-type network). The overcharge protection response time is approximately 1ms.

[0042] When the mutual inductance coefficient is 11.5uH, the energy conversion efficiency reaches 88% and the charging time is 15.4s, which is suitable for the rapid energy replenishment needs of robots during work breaks.

[0043] The innovation of the receiver circuit in this invention lies in its extreme compactness, integrated layout, and built-in hardware-level intelligent protection mechanism: To adapt to the extreme space constraints under the robot chassis, the receiver innovatively adopts a highly integrated "single board" design using small-sized 1206 surface-mount components for the LCC compensation network, synchronous rectifier bridge, and filter capacitor, achieving extremely light weight and small size of the receiver module, perfectly solving the physical problem of achieving high-power reception in a confined space; at the same time, a refined and efficient intelligent protection circuit is integrated on the output side, which senses the charging status in real time through a resistor voltage divider network, and innovatively uses a Zener diode D3 to form a hardware voltage clamping circuit, strictly limiting the sampling voltage to within 3.3V, providing the MCU with absolute overvoltage protection with nanosecond-level response, significantly improving the robustness of the system.

[0044] IV. Visual Defect Detection Algorithm To address the bottleneck of mainstream detection models' redundant parameters and high computational complexity, which prevent them from running in real-time on small robots, this invention provides an enhanced YOLOv11-SG model architecture. By embedding a newly designed C3K2-SG module into the overall framework, it overcomes the problems of lost object motion features, low detection accuracy of small defects on metal surfaces, and false positives and false negatives in material sorting. This design compresses the number of parameters to 2.5 × 10⁻⁶. 6 The computational complexity was reduced to 6.1 GFLOPs, and high-precision defect detection with an accuracy of 98.2% was achieved locally on the robot without relying on cloud computing power, overcoming the core challenge of implementing perception algorithms at the terminal.

[0045] This model introduces a point-to-point shift mechanism to achieve dynamic feature enhancement in the backbone network while maintaining model compactness, effectively balancing representation flexibility and computational efficiency. By optimizing the feature flow path, the model architecture significantly improves the reliable identification capability of subtle defects in chaotic industrial environments, supports the stable and continuous operation of robot sorting tasks, and combines lightweight design with robust detection performance, solving the feature extraction bottleneck under dynamic working conditions.

[0046] 1. Model Architecture like Figure 7 The visual defect detection model of this invention is based on an improvement of YOLOv11n, embedding a C3K2-SG module in the Backbone layer. The network has a three-level structure: Backbone: Feature extraction is achieved through the C3K2-SG module and the C2PSA module. The C3K2-SG module replaces the original Bottleneck module to reduce parameter redundancy. Neck: Employs SPF module and Upsample upsampling to achieve multi-scale feature fusion, enhancing the feature representation of small target defects (such as microcracks in steel); Head: The Detect head outputs the defect category (6 categories including cracks, inclusions, and patches) and bounding box coordinates to complete end-to-end detection.

[0047] 2. Core Module C3K2-SG Principle The structural principle of the C3K2-SG module is as follows: Figure 8 As shown, it integrates the self-moving-point convolution SMPConv and the convolution-gated linear unit CGLU. SMPConv achieves dynamic weight adaptation by sharing spatial sampling points and channel-specific weight parameters, combined with interpolation techniques, and establishes a continuous convolution that does not rely on neural networks, thus dealing with heterogeneous data distributions with minimal computational cost. CGLU, as an efficient channel mixer, integrates 3×3 depthwise convolution and gating activation mechanisms, deriving contextual channel relationships from spatial neighborhoods to enhance robustness.

[0048] SMPConv self-moving point convolution: Shares point positions and independent weight parameters in each channel, and achieves continuous convolution through interpolation, improving the flexibility of feature extraction with low computational overhead (20% less GFLOPs than traditional 3×3 convolution). CGLU Convolutional Gated Unit: A 3×3 depthwise convolution is added before the activation function of the gated branch to build a channel attention mechanism, enhance the interaction between adjacent features, and improve the robustness of the model (on the NEU-DET dataset, the crack defect detection precision is improved by 1.4% compared with YOLOv11n).

[0049] like Figure 7 As shown, this module replaces the standard Bottleneck in the C3K2 framework. By combining SMPConv and CGLU to form a collaborative structure, it achieves adaptive information filtering and discriminative feature capture. It innovatively solves the feature occlusion problem caused by variable geometry, motion blur and computational redundancy, thereby improving the defect detection accuracy and robot sorting operation stability. While ensuring the flexibility of feature extraction with low computational overhead, it enhances the robustness of the model to minor defects.

[0050] Performance metrics: The model has 2.5 × 10⁻⁶ parameters. 6 With a computational complexity of 6.1 GFLOPs, it achieves a defect identification accuracy of 98.2% and a false negative rate of 1.8% in sorting scenarios, meeting the lightweight and high-precision requirements of terminal deployment and fundamentally solving the decision-making level's dependence on cloud computing.

[0051] V. Fully Autonomous Operation Sequence Based on the above overall architecture, this invention achieves fully autonomous collaborative control driven by a state machine, such as... Figure 9 The system uses a state machine to achieve autonomous task flow in the factory design. The key timing nodes are as follows: Initial startup: The system performs a power-on self-test (including the wireless charging module, drive unit, and vision module), and the robot is in standby mode in the start-stop zone; Task Acquisition: Move along the specified path to the QR code area, and the vision module decodes the task code (material type, handling priority); Defect sorting: The material is moved to the sorting area and the YOLOv11-SG defect detection model is used to detect material defects. After screening qualified materials, the robotic arm uses the posture analysis algorithm to achieve ±0.5mm precision gripping. Precise transfer: The material moves along the path of "raw material area → preliminary processing area → temporary storage area". The trajectory is corrected by dual positioning of gyroscope and vision to ensure precise placement of materials. Autonomous operation: Real-time monitoring of the energy storage capacitor's charge level. When the charge level is ≤10%, the autonomous guidance program is triggered, the position of the transmitting coil is visually identified, and the system returns to the start / stop zone to recharge. Once fully charged, the operation process is restarted until the mission is closed.

[0052] This invention addresses the industry dilemma that existing high-precision defect detection algorithms and efficient wireless charging technologies are difficult to deploy in resource-constrained integrated robot terminals due to their large computational load and complex structure. Through a series of lightweight and integrated innovations, it constructs a practical and feasible fully autonomous solution for "logistics + sorting + battery life".

[0053] This invention fully demonstrates the feasibility of the technical solution through triple verification of industrial scenario simulation, core module testing, and full-process experiment. All core performance indicators have achieved the design goals. The specific verification process and results are as follows.

[0054] I. Industrial scenario simulation verification.

[0055] 1. Simulation Site Design Construct a 2400mm×2400mm simulated factory testing area, such as... Figure 10 As shown, it includes: a 400-450mm variable width narrow channel (suitable for small robots), functional zones (start / stop area, sorting area, rough processing area, temporary storage area), and six steel materials (three types of shapes: triangular prism / cylinder / pentagonal prism, each type containing one intact product + one defective product), simulating the real logistics "sorting-processing-storage" closed-loop process.

[0056] Functional zones: Blue start / stop zone (integrated wireless charging station), gray sorting zone (6 materials: 1 intact + 1 defective of each of 3 shapes), yellow preliminary processing zone, and green temporary storage zone.

[0057] Environmental constraints: 400-450mm variable width gray lanes (including 400mm narrow lanes), non-passage areas use matte white / yellow background to form visual boundaries, simulating the spatial constraints of real logistics sites.

[0058] 2. System Function Verification according to Figure 9 The state machine logic diagram in the diagram was tested 10 times to verify the reliability of the entire process.

[0059] Path adaptability: The robot moves along the planned path in a closed loop of "start and stop area → QR code area → sorting area → rough processing area → temporary storage area" without collision in a 400mm narrow passage. The trajectory tracking error is ≤±0.8mm (when there is a load of 1.8kg), which meets the requirements of operation in confined spaces.

[0060] Task completion rate: In 10 repeated simulations, the robot was able to autonomously complete the entire process of "task acquisition → defect sorting → precise transfer → autonomous charging" without any task interruption, achieving a 100% success rate in closed-loop operation.

[0061] Charging coordination: When the battery is low (≤10%), it will automatically return to the homing charging station with an alignment accuracy of ±1mm. After being fully charged, it will automatically restart the operation and achieve a 100% task completion rate without human intervention.

[0062] Module coordination: Visual defect detection, motion control, and wireless charging modules respond in a coordinated manner, with a connection time of ≤30s from "low battery trigger charging" to "full charge restart operation", without module lag or data delay.

[0063] II. Core Module Testing and Verification: Key Technology Performance Meets Standards 1. Lightweight Defect Detection Model (YOLOv11-SG) Testing Test conditions: Based on the NEU-DET dataset (6 defect classes, 1800 images, training:validation:test = 7:2:1), Windows 10 system + RTX4070 GPU, 300 rounds of training, input image size 200×200; compared with the commonly used YOLOv8n and YOLOv11n, the performance results are shown in Table 1.

[0064] Table 1 Test results of the lightweight defect detection model Model <![CDATA[Parameter quantity (×10 6 ).]]> Computational complexity (GFLOPs) Average Precision (%) Average Recall (%) mAP50 (%) Sorting accuracy rate (%) Missed detection rate in sorting scenarios (%) YOLOv8n 3.2 8.1 70.45 73.37 74.82 Calculation too slow Calculation too slow YOLOv11n 2.6 6.3 69.42 72.55 73.63 95.5 4.2 YOLOv11-SG 2.5 6.1 71.03 73.23 74.50 98.2 1.8 Conclusion: The lightweight characteristics (number of parameters and computational complexity) of the model used in this invention are superior to the comparative models. The detection accuracy (especially for small target defects) and practicality in sorting scenarios meet the standards and can be deployed in robot terminals.

[0065] 2. Lightweight Wireless Charging System Test Test conditions: Transmitter / receiver coil inductance 32uH, 24V DC power supply, transmission power 350W, testing charging efficiency and time under different mutual inductance coefficients.

[0066] The performance results are shown in Table 2.

[0067] Table 2 Test Results of Wireless Charging System Mutual inductance coefficient (uH) Emitter resonant capacitance (nF) Resonant capacitance at the receiver (nF) Charging time (s) Energy conversion efficiency (%) 9.5 38.1 443.4 20 84 10.88 38.7 386.8 18.5 85 11.5 38.9 366.3 15.4 88 12 39.1 351.0 17.5 87 12.5 39.2 336.9 17 86 Conclusion: When the mutual inductance coefficient is 11.5uH, the system reaches its optimal state with an energy conversion efficiency of 88% (close to the average level of 88%-90% in existing literature), a charging time of 15.4s, and overcharging prevention is achieved through relay-type CC-CV switching, thus meeting both safety and rapid energy replenishment requirements.

[0068] 3. Mechanical and motion control testing Wheel system motion accuracy test: under no load / with load (1.8kg), parameters such as longitudinal speed, lateral speed, and rotational speed are tested. Through error compensation algorithm (load coefficient to correct wheel speed, PID parameter adaptive), the lateral speed error rate is reduced from 4.0% (with load but no compensation) to below 1.5%, and the linear positioning accuracy reaches ±0.5mm.

[0069] Robotic arm grasping test: 10 repeated grasping of 3 types of materials with positioning error ≤ ±0.5mm, no material falling off or posture deviation, grasping success rate 100%.

[0070] III. Full-process usage verification: Achieving autonomy and practicality standards 1. Use Case Design In the simulated factory environment, the robot can start with a single click from the start-stop area without human intervention, and continuously perform 10 batches of material sorting tasks (6 materials per batch), during which it will trigger two autonomous charging cycles (when the battery level is ≤10%).

[0071] 2. Results of use Level of autonomy: The entire process is completed automatically without human intervention. Task acquisition (QR code decoding), defect detection (YOLOv11-SG), precise transfer (PID positioning), and autonomous charging (visual alignment) are all completed automatically, completely eliminating reliance on cloud computing and human intervention. Key performance indicators achieved: Material defect identification accuracy of 98.2%, gripping and positioning accuracy of ±0.5mm, and wireless charging efficiency of 88%; all three key indicators met the design goals. Scenario adaptability: The hardware structure is compact and can be moved flexibly in a 400mm narrow channel. It can be deployed without modifying the existing site and is suitable for the space and cost requirements of small and medium-sized logistics scenarios.

[0072] 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 logistics robot system capable of sorting steel body material defects, characterized by, The robot system builds a sorting-logistics-endurance integrated whole-process autonomous system architecture, including a perception layer, a decision layer and an execution layer, the perception layer includes a vision subsystem, a gyroscope and an encoder, responsible for collecting information to provide information input for sorting decision and path planning; the decision layer takes a microcontroller as the core, after receiving the data of the perception layer, runs a PID control algorithm and inverse kinematics analysis, generates mechanical arm grabbing instructions, chassis motion instructions and wireless charging control instructions, realizes intelligent decision and precise control; the execution layer includes a chassis, a three-degree-of-freedom mechanical claw and a wireless charging receiving end circuit, responsible for executing specific physical actions, and feeding back the execution state to the decision layer in real time through sensors; Among them, a lightweight YOLOv11-SG defect detection model is deployed in the vision subsystem, which can realize real-time identification and positioning of materials and perform defect detection; the decision layer controls the mechanical arm to grab qualified materials according to the detection results of the vision subsystem, and then transports the materials to the rough machining area or the temporary storage area according to the preset path; The wireless charging receiving end circuit is integrated below the chassis, adopts a single coil architecture and an energy storage capacitor structure, when the decision layer detects that the energy storage capacitor capacity is less than or equal to 10%, an autonomous guidance program is triggered immediately, the robot returns to the wireless charging transmitting end automatically and completes the alignment, the wireless charging system adopts magnetic coupling resonance technology, the transmitting end converts direct current into high-frequency alternating current through a full-bridge inverter, drives the transmitting coil to generate an electromagnetic field, and after the receiving coil couples the energy, a stable direct current is output through a rectifier bridge and a filter capacitor to charge the energy storage capacitor.

2. The logistics robot system capable of sorting defects of steel body materials according to claim 1, characterized in that, The YOLOv11-SG defect detection model embeds a C3K2-SG module in the backbone network, and the network is divided into three levels of structure: The backbone network: realizes feature extraction through the C3K2-SG module and the C2PSA module, and the C3K2-SG module replaces the original bottleneck Bottleneck to reduce parameter redundancy; The neck network: adopts an SPF module and up-sampling to realize multi-scale feature fusion and enhance the feature expression of small target defects; The detection head: outputs defect categories and bounding box coordinates through the detection head to complete end-to-end detection.

3. A logistics robot system capable of sorting steel body material defects according to claim 2, characterized in that, The C3K2-SG module replaces the standard Bottleneck in the C3K2 framework, realizes adaptive information filtering and discriminative feature capture by fusing the self-moving point convolution SMPConv and the convolution gating linear unit CGLU, wherein the SMPConv shares point positions in each channel and has independent weight parameters, and realizes continuous convolution through interpolation; the CGLU adds a 3x3 deep convolution before the activation function of the gating branch to construct a channel attention mechanism, enhance the interaction of adjacent features, and improve the robustness of the model.

4. The logistics robot system capable of sorting defects of steel bulk material according to claim 1, characterized in that, The transmitter circuit adopts a full-bridge H inverter topology to convert direct current into a high-frequency square wave, and an LCC resonance cavity as a high-quality factor band-pass filter, at the same time, a R32 alloy sampling resistor, a TP181A1 amplifier and an RC filter network are integrated in the loop, and the power is displayed in real time through a digital tube, and the fan module ensures heat dissipation.

5. The logistics robot system capable of sorting defects of steel body materials according to claim 1, characterized in that, The receiver circuit highly integrates the LCC compensation network, the synchronous rectification bridge and the filter capacitor by using small-size 1206 patch elements; meanwhile, the intelligent protection circuit is integrated on the output side, the charging state is sensed in real time through the resistance voltage division network, and the zener diode D3 is used to form a hardware voltage clamping circuit, so that the sampling voltage is strictly limited within 3.3 V, thereby providing the MCU with nanosecond-level response absolute overvoltage protection.

6. The logistics robot system capable of sorting defects of steel bulk material according to claim 1, characterized in that, The wireless charging system adopts the magnetic coupling resonance technology, the transmitting end converts the direct current into high-frequency alternating current through the full-bridge inverter, drives the transmitting coil to generate an electromagnetic field, the receiving coil couples energy, and outputs stable direct current through the rectifier bridge and the filter capacitor to charge the super capacitor; the energy output by the wireless charging receiving end is stored through the super capacitor group, and the boost and buck modules are used to supply power to the driving unit and the computing unit respectively.

7. The logistics robot system capable of sorting defects of steel bulk material according to claim 1, characterized in that, The visual subsystem is a Raspberry Pi 5 visual subsystem, including a camera and a two-dimensional code scanning module, the two-dimensional code scanning module is used for acquiring a task code, the microcontroller is an Arduino Portenta H7, the Arduino and the Raspberry Pi communicate with each other through a UART serial port, the Raspberry Pi outputs visual positioning data, the Arduino fuses gyro angular position information, and dynamically corrects the chassis motion trajectory by optimizing PID parameters.

8. The logistics robot system capable of sorting defects of steel bulk material according to claim 1, characterized in that, The chassis adopts a Mecanum four-wheel system, the angle between the roller axis and the wheel axis is gamma, omnidirectional movement is realized through forward and inverse kinematics equations, and the functions of "oblique driving + fine steering" are realized through multi-parameter cooperative adjustment.

9. The logistics robot system capable of sorting defects of steel bulk material according to claim 1, characterized in that, The mechanical claw is made of carbon fiber composite material and is 3D printed, the claw part is embedded with polyurethane stripes, has the ability of radial contraction and axial rotation composite motion, integrates a PVDF film sensor to realize tactile feedback, and sets double cameras on the center and the outside to realize synchronous defect identification and clamping positioning.

10. The logistics robot system capable of sorting defects of steel body materials according to claim 1, characterized in that, The mechanical arm adopts a human-simulated three-degree-of-freedom structure, a kinematics model is established based on the D-H parameter method, a cubic spline interpolation is used to smooth the trajectory, a PID closed-loop control is used to compensate the position error, and the grasping positioning accuracy is ensured.

Images