To address issues such as low efficiency in manual loading and unloading and poor sorting accuracy in sheet metal laser cutting, this paper presents a design for an automated loading, unloading, and sorting system based on machine vision and PLC control.
The system consists of a three-dimensional storage rack, a transport and positioning mechanism, a vision recognition system, a six-axis robot, and a sorting and palletizing module, thereby automating the entire manufacturing process from raw materials to finished products.
Testing has confirmed that the system operates stably and achieves a recognition accuracy rate of over 98.5 percent, with a maximum sorting capacity of 240 pieces per hour.
Before this, the level of intelligence and production performance of the laser cutting unit had been further improved.
Overall Design and Workflow of the System
Overall Structural Design
This automated loading, unloading, and sorting system employs a modular design, integrating the loading, conveying, vision recognition, robotic gripping, sorting units, and control system into a single unit.
It features a relatively compact structure and strong adaptability, making it suitable for a wide range of industrial applications.
The loading unit employs a double-tiered three-dimensional storage rack with 15 storage levels, capable of holding sheet materials ranging from 3000 mm × 1500 mm to 6000 mm × 2000 mm.
It integrates a hydraulic push rod, servo drive, and double-sheet detection to achieve automatic separation and loading of sheet materials.
The conveying unit consists of a motor-driven roller conveyor, with pneumatic cylinders pushing V-blocks and baffles for precise positioning, achieving a repeatability error of ≤±0.5 mm.
The vision recognition unit is located in the middle section of the conveyor line.
It uses a 5-megapixel industrial camera and annular lighting, relying on the Halcon library for contour recognition, position correction, and defect detection. The robotic gripping unit utilizes a Nacho LR Mate 200iD six-axis robot with a repeatability of ±0.02 mm.
Equipped with a vacuum suction cup and an electromagnetic gripper at the end effector, it enables flexible gripping of workpieces made of various materials and in various sizes.
The sorting unit employs a diverging conveyor line with double-decker interchangeable carts.
Based on vision system results, finished products are directed to the corresponding bins, with a maximum stacking weight of 3 metric tons.
Workflow Design
The plant’s workflow is based on the principle of cyclic operation, ensuring seamless coordination between each process step. The specific workflow is as follows.
(1) Loading Stage:
The automated storage and retrieval system (AS/RS) automatically dispenses sheet material according to production instructions, and a hydraulic pusher feeds the sheet onto the roller conveyor line.
(2) Conveying and Positioning:
The roller conveyor line transports the sheet to the cutting area at a speed of 15 m/min, where the positioning mechanism activates to precisely position the sheet.
(3) Vision Recognition:
An industrial camera captures images of the positioned sheet.
Using edge detection and template matching algorithms, it identifies the sheet’s position and model, and feeds back any positional deviations to the robot control system.
(4) Robot Gripping and Cutting:
Based on the coordinate information provided by the vision system, the robot grips the sheet and places it on the laser cutting worktable. During the cutting process, the robot returns to the standby position to prepare for the next cycle.
(5) Unloading and Sorting:
After cutting is complete, the robot picks up the finished parts;
The vision system performs secondary identification and classification of the finished parts; and the sorting unit conveys the finished parts to the corresponding exit stack according to instructions.
The entire process employs a parallel strategy: while one sheet is being cut, the robot is already prepared to pick up the next sheet, which significantly improves robot utilization.
This “process while preparing” assembly line model increases equipment utilization from 60% in the traditional serial mode to over 85%.
Calculations show that the system’s standard cycle time is 260–300 seconds, with a maximum sorting capacity of 240 pieces per hour—more than three times the efficiency of manual operations.
The system’s cycle time can be dynamically adjusted via the MES interface;
It automatically extends the cycle when processing large sheet metal parts and reduces auxiliary time during high-speed cutting of small parts, demonstrating excellent flexibility and adaptability.
Analysis of Key Module Design and Selection
Loading and Unloading Mechanism Design
As the core component of the material handling system, the loading and unloading mechanism employs a double-girder truss structure and consists of a frame, a drive system, crossbeams, and actuators.
The drive system is equipped with servo motors and high-precision ball screws, with a horizontal travel speed of 40 m/min, a vertical speed of 20 m/min, and an acceleration of ≤0.5 g, ensuring smooth operation.
The actuator is equipped with 36 vacuum suction cups, each with a diameter of Ø150 mm; the lifting capacity of a single suction cup is >30 kg, and the total lifting capacity reaches 720 kg (see Table 1).
The suction cups are made of wear-resistant polyurethane, suitable for gripping sheet materials of various types, and are equipped with pneumatic pressure monitoring and automatic compensation functions to ensure gripping safety and system reliability in real time.
| Parameter Category | Technical Specification | Remarks |
|---|---|---|
| Maximum Sheet Size (mm × mm) | 6,000 × 2,000 | Customizable |
| Maximum Sheet Weight (kg) | 1,000 | Carbon Steel |
| Horizontal Travel Speed (m/min) | 40 | Servo-Controlled |
| Vertical Travel Speed (m/min) | 20 | Servo-Controlled |
| Positioning Accuracy (mm) | ±0.5 | Optical Encoder |
| Number of Vacuum Suction Cups | 36 | Diameter Ø150 mm |
| Suction Capacity per Cup (kg) | 30 | Under Normal Operating Conditions |
Table 1. Key Technical Specifications of the Loading and Unloading Mechanism
Visual Inspection System Design
The visual inspection system is key to intelligent sorting. Its primary function is to accurately identify the position, shape, and defects of sheet metal parts using image processing technology.
It consists of three main components: image acquisition, image processing and analysis, and result output.
Image Acquisition Hardware and Lighting System
For the hardware, the Basler CA5000-500um industrial camera was selected, with a resolution of 5 million pixels and a frame rate set to 15 fps, along with a ring-shaped LED for diffuse reflection to suppress glare from metal surfaces.
The camera lens is a Computar M1614-MP2 fixed-focus lens with a focal length of 16 mm and an aperture range of F1.4–16C, ensuring clear images at working distances between 300 and 800 mm.
The lighting system uses red ring-shaped LEDs (wavelength 625 nm) in conjunction with a polarizing filter to effectively suppress specular reflections from metal surfaces, improving image contrast by more than 40%.
It connects to an industrial PC via Gigabit Ethernet to enable real-time transmission of image data.
Image Processing and Feature Extraction Algorithms
The software algorithm is based on a multi-feature fusion strategy and primarily includes the following processing steps:
(1) Image preprocessing: Gaussian filtering and histogram equalization are used to eliminate noise and the effects of uneven lighting.
To address the reflective properties of metal surfaces, an adaptive bilateral filtering algorithm was designed to smooth out noise while preserving edge details, resulting in a 30% increase in processing speed.
(2) Edge Extraction: The Canny operator is used to detect workpiece contours, combined with the Hough transform to identify straight line and arc features.
For complex curved boundaries, a Bézier curve fitting algorithm is introduced, achieving sub-pixel-level fitting accuracy.
(3) Template Matching: The system matches shape-based feature descriptors against preset templates to calculate positional offsets and rotation angles.
It comes with 50 built-in standard templates and supports online learning to add new templates, with matching times kept under 50 ms.
(4) Defect Detection: Surface defects such as scratches and deformations are identified through grayscale threshold segmentation and morphological operations.
To address the challenge of detecting minute defects, a semantic segmentation model based on convolutional neural networks is employed, achieving a detection rate of 95% for scratches 0.5 mm or larger.
Deep Learning Optimization and Recognition Performance
Testing has shown that the system achieves a recognition success rate of over 98.5% for sheet metal parts ranging in size from 100 to 1,500 mm, with an average processing time of less than 200 ms, meeting real-time sorting requirements.
To address workpieces with complex geometries, the system incorporates deep learning algorithms; training via convolutional neural networks has enhanced its ability to recognize small-sample and irregularly shaped parts.
Specifically, an optimized, lightweight YOLOv5s model was adopted, which reduces the model’s parameter size to 7.2 MB while maintaining accuracy, enabling deployment on edge computing devices.
The training dataset consists of 5,000 annotated images, which were expanded to 30,000 using data augmentation techniques;
Transfer learning improved the recognition accuracy by 12% in low-sample-count scenarios.
For aluminum alloy parts with severe glare, a multi-exposure fusion algorithm was developed, which increased the dynamic range by 60% and effectively resolved the issue of edge loss caused by overexposure.
Robotic Gripping Unit Design
The robotic gripping unit utilizes a Fanuc LR Mate200iD six-axis industrial robot, which features a 717 mm working radius and a repeatability of ±0.02 mm, meeting the requirements for high-cycle-rate sorting (see Table 2).
The end-effector features a modular design, integrating a vacuum suction cup and an electromagnetic gripper.
These are automatically switched via a PLC and are suitable for handling flat sheet metal and workpieces with holes or curved surfaces, respectively.
It is equipped with a force-control sensor that monitors gripping force in real time, effectively preventing workpiece deformation or dislodgement.
Trajectory planning employs a combination of linear and circular interpolation; after simulation optimization, this significantly reduces idle travel time and effectively prevents interference between equipment.
| Parameter Category | Technical Specification | Remarks |
|---|---|---|
| Robot Model | FAIRINO LR Mate 200iD | Six-Axis Articulated Robot |
| Maximum Payload (kg) | 7 | Including End Effector |
| Repeatability (mm) | ±0.02 | ISO Standard |
| Working Radius (mm) | 717 | — |
| Control Mode | Point-to-Point / Continuous Path | — |
| Communication Interface | PROFINET | Integrated with PLC |
Control System Design
As the core of the system, the control system employs a “PLC + host computer” architecture to achieve integrated control.
The lower-level controller uses a Siemens S7-1200 series PLC, which is responsible for signal processing, servo drives, and safety interlocks;
The host computer integrates the HMI interface and vision processing software, and handles task scheduling, status monitoring, and data management functions.
The network architecture is based on Industrial Ethernet and uses the PROFINET protocol to enable real-time communication between the PLC, the robot controller, and the vision system.
Safety Protection Strategy
The safety system is equipped with light curtain sensors, safety door locks, and emergency stop buttons;
When personnel intrusion or equipment abnormalities are detected, a safety shutdown is immediately initiated to ensure safe and reliable operation.
The safety controller adopts Siemens SIMATIC Safety’s modular design, achieving a Safety Integrity Level (SIL) of 2 with a response time of <20 ms.
The system incorporates a three-tier safety strategy: Level 1 consists of zone protection, using safety light curtains to monitor the robot’s work area;
Level 2 consists of access control, where safety door locks ensure the equipment is de-energized during maintenance; Level 3 consists of emergency stop, where triggering the emergency stop button immediately cuts off the servo power supply and activates the brake.
The control process includes the following core functions:
(1) Task Scheduling: Receives production orders issued by the MES system and automatically generates a work sequence.
(2) Motion Control: Coordinates the timing of movements between the loading/unloading mechanisms, the robot, and the conveyor line.
(3) Vision Guidance: Receives coordinate deviations from the vision system and corrects the robot’s grasping position.
(4) Data Traceability: Records sorting quantities, labor hours, and fault information, and generates production reports.
The system supports both automatic and manual operating modes; in automatic mode, it can operate unmanned, while manual mode is used for equipment debugging and maintenance.
In addition, the system integrates predictive maintenance capabilities.
By monitoring parameters such as servo motor current and temperature, and utilizing an LSTM neural network to predict potential failures, it provides early warnings 72 hours in advance with an accuracy rate of 85%, significantly reducing unplanned downtime.
System Integration and Performance Validation
System Integration Solution
This automated loading, unloading, and sorting system is the result of the deep integration of three major systems: mechanical, electrical, and software.
On the mechanical side, high-strength bolts and positioning pins ensure precise alignment of all modules;
On the electrical side, a distributed I/O layout simplifies wiring operations; at the software level, data exchange with the MES system is facilitated through the OPC UA protocol.
This system innovatively applies digital twin technology to map the operational status of physical entities via virtual models, thereby enabling fault prediction and optimization of parameter settings, Combined with AI vision recognition technology similar to TRUMPF’s Sort Master Vision, the system can independently identify separated components and automatically plan robot motion paths, enabling automated operations without the need for specialized programming for the sorting process.
Performance Testing and Validation
To verify the correctness of the device design and its performance metrics, a physical test platform was constructed to conduct various functional tests.
Under environmental conditions of 25 ± 5 °C and 60% ± 10% relative humidity, experiments were conducted on materials such as carbon steel, stainless steel, and aluminum alloy, with sample dimensions ranging from 30 mm × 80 mm to 1,000 mm × 1,500 mm.
Positioning Accuracy and Repeatability Tests
For positioning accuracy testing, a laser tracker was used to measure the repeatability of the robot end-effector’s positioning.
At maximum operating speed, the measured positioning error was less than 0.1 mm, meeting the requirements for picking up and placing sheet metal parts. The positioning accuracy test data are shown in Table 3.
Unit: mm
| Test Number | X-Axis Error (mm) | Y-Axis Error (mm) | Z-Axis Error (mm) |
|---|---|---|---|
| 1 | 0.08 | 0.06 | 0.05 |
| 2 | 0.09 | 0.07 | 0.06 |
| 3 | 0.07 | 0.05 | 0.04 |
| 4 | 0.08 | 0.06 | 0.05 |
| 5 | 0.06 | 0.08 | 0.05 |
| Average | 0.076 | 0.064 | 0.050 |
Table 3. Positioning Accuracy Test Data
From a statistical perspective, the standard deviation of positioning errors in all directions is less than 0.012 mm, indicating that the system exhibits good repeatability.
In particular, stability in the Z direction is optimal, primarily due to the precise braking of the servo motor and the rigid design of the mechanical structure.
Visual Recognition Accuracy Tests
Visual recognition rate tests were conducted under various lighting and background conditions.
Recognition tests were performed on 100 sheet metal parts of different specifications.
The results showed a recognition success rate of over 98.5%; recognition errors occurred only for a small number of parts with severe glare or blurred edges, and overall performance was stable (see Table 4).
| Workpiece Type | Number of Test Samples | Successfully Recognized | Recognition Rate (%) |
|---|---|---|---|
| Regular Rectangular Parts | 30 | 30 | 100 |
| Parts with Holes | 25 | 25 | 100 |
| Irregularly Shaped Parts | 25 | 24 | 96 |
| Small Parts (＜50 mm) | 20 | 19 | 95 |
| Total | 100 | 98 | 98.5 |
Table 4. Visual Recognition Rate Test Results
Further analysis indicates that the primary reason for the relatively low recognition rate of irregularly shaped parts is that some workpieces have complex curved surfaces, resulting in insufficient features in the two-dimensional images.
To address this issue, structured light 3D vision technology will be introduced in subsequent work to improve recognition accuracy by acquiring depth information.
The recognition rate for small workpieces is limited by the camera’s resolution;
when the workpiece size is less than 30 mm, the insufficient number of edge pixels can easily lead to misclassification.
This can be improved by incorporating a local magnification step during imaging.
Continuous Operation Stability and Sorting Efficiency
An 8-hour continuous stability test demonstrated that the device performed excellently, with an average cycle time of 15 seconds per part and a maximum sorting capacity of 240 parts per hour—approximately three times the efficiency of manual operation— and no malfunctions or downtime were observed.
A comprehensive performance comparison shows that it far surpasses traditional manual methods in both speed and accuracy.
It is particularly well-suited for flexible production lines handling a wide variety of items in small batches.
Furthermore, compared to other models in the same category, it offers certain economic advantages when processing small-to-medium-sized items (In terms of sorting small- and medium-sized objects, its flexible recognition algorithm offers distinct advantages over the Huagong Laser LCK series, making it irreplaceable and a standout feature).
Application Analysis and Outlook
The application prospects for automated loading, unloading, and sorting systems for laser-cut sheet metal parts are extremely promising.
In industries such as automotive manufacturing, electrical equipment, and aerospace, these systems enable highly efficient sorting and non-destructive handling, offering significant economic benefits.
A single machine can replace 2 to 3 operators, saving approximately 450,000 yuan in annual labor costs, while increasing equipment utilization from 60% to over 85% and boosting production capacity by about 30%, with a payback period of less than two years.
The direction of technological development is toward intelligence, flexibility, and sustainability.
Deep learning algorithms enhance the ability to recognize complex workpieces; digital twin technology enables predictive maintenance and parameter optimization; 5G networks combined with edge computing ensure real-time data transmission and analysis;
energy consumption optimization and lightweight design have significantly reduced carbon emissions throughout the entire life cycle, among other benefits.
The convergence and development of these technologies have provided strong support for the digital transformation of the manufacturing industry.
Conclusion
This paper describes the design of an automated loading, unloading, and sorting system for laser-cut sheet metal parts.
Integrating machine vision, robotics, and PLC control, the system accomplishes a fully automated process from loading to positioning to sorting.
This modular system features a compact design, facilitating rapid integration into existing equipment.
Experiments have demonstrated that the robot’s repeatability error is less than 0.1 mm, the vision recognition rate reaches 98.5%, and the sorting efficiency of 240 parts per hour is three times higher than that of manual labor, meeting the flexible production requirements for parts of various materials and dimensions.
Moving forward, we will refine the vision algorithms for complex backgrounds to enhance the recognition of irregularly shaped parts and further strengthen integration with the MES system to achieve intelligent management of production data, thereby providing the industry with a more comprehensive intelligent solution.