The application of intelligent robot systems in industrial manufacturing and maintenance has become a cornerstone of modern automation. These systems, equipped with a suite of internal and external sensors, effectively grant machines capabilities such as vision, hearing, and touch. The defining characteristics of an intelligent robot—perception, reaction, and cognition—enable them to undertake complex tasks traditionally performed by humans, leading to significant improvements in efficiency, consistency, and safety. In the domain of railway freight car management, the marking of identifiers such as operational content, model, and maintenance status on vehicle bodies is essential for organization and traceability. Historically, this spraying task has been manual, a method plagued by low efficiency, positional inaccuracies, and inherent safety risks for workers who must operate at height. Consequently, the transition to employing intelligent robots for this precise and repetitive task represents a critical advancement. The core challenge in this automation lies in the generation of the spraying path, which directly dictates the final quality, clarity, and uniformity of the applied marking. An inaccurate or suboptimal path can lead to misaligned markings, uneven coating, or overspray. While various path-planning algorithms exist, many struggle with the specific requirements of multi-height, large-surface喷涂 common to freight cars. To address this, we have designed and implemented an Automatic Spraying Path Generation System specifically for intelligent robots tasked with railway freight car marking.
The overall architecture of our designed intelligent robot spraying system is composed of six integrated subsystems that work in concert. The holistic system architecture is designed to ensure a seamless workflow from command reception to task completion and data management. The synergy between these components is what enables the intelligent robot to perform its duties autonomously and reliably. The core subsystems and their primary functions are detailed in the table below:
| Subsystem | Primary Components | Core Function |
|---|---|---|
| Spraying Robot | Dual-arm manipulator, Electric lift, Control cabinet | Executes the physical spraying operation; one arm holds the stencil, the other operates the spray gun. |
| Automatic Guided Transport | Chassis, Driven steering wheels, Magnetic navigation & obstacle avoidance sensors, On-board battery & controller | Mobile platform that carries the entire system to designated spraying stations autonomously, avoiding obstacles. |
| Paint Supply | Air compressor, Automatic paint supply unit, Automated spray gun, PLC controller | Provides and regulates the paint and air pressure required for the spraying process. |
| Auxiliary Support | Cleaning equipment, Paint mist recovery unit, Vehicle-mounted pump | Supplies essential utilities and manages waste (e.g., overspray, cleaning fluids). |
| Stencil Library | Physical repository of acrylic stencils (e.g., 11 types, 60 pieces) | Stores all marking stencils; the robot’s gripper arm retrieves the required stencil based on the task command. |
| Electrical & Intelligent Control | Ground Control Module, On-board Control Module, Wireless Communication Module | The central “brain” of the intelligent robot. It receives tasks, generates paths, coordinates all subsystems, and manages data. |
The physical embodiment of the intelligent robot is a dual-arm structure mounted on an electric lifting platform. This configuration allows one arm to precisely grasp and position the required stencil from the library against the freight car surface, while the other arm maneuvers the spray gun. The electric lift provides crucial vertical mobility, enabling the intelligent robot to cover the entire range of marking heights found on different freight car models. This hardware design is fundamental to the system’s adaptability.
The intelligence and coordination of the entire operation are governed by the Electrical & Intelligent Control Subsystem. This subsystem operates on a client-server model facilitated by robust wireless communication. The Ground Control Module, featuring a server and Human-Machine Interface (HMI), connects to the railway’s vehicle management database. It receives vehicle information, generates the specific marking instructions (e.g., “spray identifier ‘X’ at position ‘Y’ on car ‘Z'”), and dispatches these tasks. The On-board Control Module, powered by a high-capacity lithium battery for sustained operation, is responsible for real-time execution. It utilizes a PC-based soft logic controller, which integrates the functionalities of a PC and a PLC, offering superior computational power for complex path planning while maintaining real-time control robustness. Communication between these modules and within the mobile platform relies on a network built with CANopen bus, wireless Ethernet bridges, and switches, ensuring reliable and low-latency data transmission for the intelligent robot’s commands and sensor feedback.
The most critical computational task performed by the On-board Control Module is the automatic generation of the spraying path or trajectory. The quality of the applied coating is highly sensitive to the spray gun’s path, which must maintain a constant orientation (typically perpendicular) and distance relative to the work surface. To achieve this, our system employs a point cloud technology-based algorithm. The logical flow for this automatic path generation is as follows:
First, point cloud data representing the target spraying surface area is acquired. For a freight car, this involves segmenting the broad surface into the specific, smaller areas designated for marking. Once the point cloud \(Q\) for a target area is obtained, it is processed. The slicing direction for the path is determined interactively, aligned with the intended spray gun movement direction. The slice thickness \(\delta\) is calculated based on the desired overlap and coating parameters. The point cloud is then mathematically sliced into layers. The core of the path generation involves calculating the precise offset points that define the gun’s trajectory. Let \(Q_i^j\) represent a sample point in the cloud, where \(j\) denotes the slice layer and \(i\) is the point index within that layer. For each point \(Q_i\), we compute its local surface normal vector \(\vec{N}_i\). This is achieved by analyzing its neighboring points \(P_k\) (where \(k=1,2,…,m\)). The unit normal vector \(\vec{n}_{\Delta_k}\) for each triangle formed by \(Q_i\) and its neighbors is calculated. A weighted average of these triangle normals gives the final estimated surface normal at \(Q_i\):
$$
\vec{N}_i = \frac{\sum_{k=1}^{m} \omega_k \cdot \vec{n}_{\Delta_k}}{\|\sum_{k=1}^{m} \omega_k \cdot \vec{n}_{\Delta_k}\|}
$$
where the weight \(\omega_k\) is often related to the area \(S_{\Delta_k}\) of the corresponding triangle, for instance, \(\omega_k = S_{\Delta_k}^\gamma\), with \(\gamma\) as a control parameter.
With the surface normal \(\vec{N}_i\) and a defined standoff distance \(H\) (the optimal distance between spray gun nozzle and surface), the offset point \(O_i\) for the gun tip position is calculated using a vector offset algorithm:
$$
O_i = Q_i + H \cdot \vec{N}_i
$$
The data structure for \(O_i\) contains both its coordinate and the associated unit normal vector \(\hat{N}_i\), which defines the gun’s orientation. By applying this process to all relevant sample points in the point cloud \(Q\), we generate a complete set of offset points \(O = \{O_1, O_2, …, O_n\}\). Finally, these discrete points are connected into a continuous, smooth path using interpolation techniques (e.g., B-spline interpolation). This resulting continuous trajectory is the optimized spraying path that the intelligent robot will follow, ensuring consistent gun orientation and distance throughout the operation.
To validate the performance of our intelligent robot spraying path automatic generation system, a comprehensive test platform was constructed. The core of this platform was a UR10 collaborative robot serving as the intelligent robot manipulator, mounted on a custom automatic guided transport vehicle. The stencil library was populated, and all control subsystems were integrated. Testing was conducted to evaluate three key aspects: mobility and obstacle avoidance, positioning accuracy, and final喷涂 quality.
The automatic guided transport subsystem, carrying the full intelligent robot system, was tasked with navigating to target spraying stations in two distinct environments: a natural outdoor setting and a structured indoor setting. The path taken by the intelligent robot system in both environments was logged and analyzed. In both cases, the system successfully navigated from its starting point to the designated target location. Crucially, it demonstrated effective real-time obstacle avoidance, dynamically adjusting its path to circumvent static obstacles and walls without collision. This confirms the reliability of the mobile platform in delivering the intelligent robot to its worksite under various conditions.
The accuracy of the喷涂 execution and the efficacy of the generated path were tested by commanding the intelligent robot to spray markings on vertical panels at specified heights. The target height was compared against the actual height of the喷涂 center point measured post-operation. The results, summarizing tests across multiple target heights, are presented below:
| Target Height (mm) | Actual Sprayed Height (mm) | Absolute Error (mm) |
|---|---|---|
| 500 | 500.2 | 0.2 |
| 1000 | 999.7 | 0.3 |
| 1500 | 1500.4 | 0.4 |
| 2000 | 1999.6 | 0.4 |
The data shows that the positioning error was consistently below 1 mm, with a mean error of approximately 0.3 mm. This high level of accuracy is a direct result of the precise trajectory generated by the point cloud algorithm and the accurate servo control of the intelligent robot.
Furthermore, the visual quality of the sprayed markings was excellent. The intelligent robot correctly retrieved different stencils from the library based on the task command and executed the spray. The resulting markings exhibited sharp edges, complete coverage, and no blurred areas. To quantify the uniformity of the coating, the spray overlap rate \(\zeta\) was calculated for several runs. The overlap rate is a critical metric indicating the consistency of the path and the uniformity of paint deposition. It is defined as:
$$
\zeta = \left(1 – \frac{D}{K}\right) \times 100\%
$$
where \(K\) is the spray fan width and \(D\) is the distance between successive parallel paths. For our tests, the system maintained an overlap rate very close to the theoretical target of 100%, with measured values consistently between 99.5% and 100%. This indicates no gaps or missed areas between adjacent spray paths, confirming the effectiveness of the automatically generated path in achieving complete and uniform coverage. The high overlap rate, combined with sub-millimeter positioning accuracy, validates that the path generation system successfully translates digital commands into flawless physical execution by the intelligent robot.
In conclusion, the transition from manual to automated spraying of railway freight car markings is essential for improving efficiency, standardization, and worker safety. The system we have presented addresses the core technical challenge in this automation: the intelligent generation of an optimal spraying path. By integrating a mobile platform, a dual-arm intelligent robot, and a sophisticated control system centered on a point-cloud-based path planning algorithm, the system achieves a high degree of autonomy and precision. Experimental validation confirms that the intelligent robot can reliably navigate to its workstation, avoid obstacles, and execute喷涂 tasks with exceptional positional accuracy and coating uniformity. This demonstrates the practical viability of using such an intelligent robot system for industrial marking applications, paving the way for wider adoption in railway maintenance and other similar domains requiring precise, automated surface treatment.