In the corrosion protection workshop of the Steel Structure Maintenance Center at the Three Gorges Project, two intelligent spray-painting robots are at work. Moving steadily across large sluice gates, they apply coatings with uniform coverage and strong adhesion. When working on closely spaced trash racks, the robots retract their arms with precision, spraying deep into narrow gaps For the R&D team, scenes like this once seemed out of reach.

The Steel Structure Maintenance Center at the Three Gorges Project is responsible for corrosion protection work covering approximately 35,000 square meters of metal structures each year. Massive sluice gates, each weighing tens of tonnes, and densely arranged trash racks will undergo a series of processes after dismantling, including abrasive blasting for rust removal, zinc spraying, and multilayer coating, before being reinstalled with a protective coating system. In the past, however, this work was done almost entirely by hand. The process involved dust-heavy conditions, highly repetitive and physically demanding tasks, and inconsistent efficiency. More importantly, the quality of corrosion protection depended heavily on the experience of individual operators.

The project team at the Overhaul and Maintenance Factory (OMF) kept asking themselves: could machines take over this demanding work and do it better than humans?

To identify a suitable robotic solution, the project team conducted a systematic feasibility analysis of replacing manual operations, considering the characteristics of components such as sluice gates and trash racks at the Three Gorges Project. Their goal extended beyond automated spraying: the system needed to adapt to a wide range of complex structures, including large flat surfaces, irregular edges and corners, and densely arranged grid elements.

After defining the development direction, the team conducted on-site investigations at several shipbuilding yards and steel structure corrosion protection facilities. They examined different types of spray-painting robots, focusing on structural design, motion systems, trajectory control, spray-gun compatibility, and performance in real-world applications on complex steel surfaces. The team then compared various solutions in terms of service scope, flexibility, process maturity, and potential for upgrading. After repeated evaluations, the team ultimately decided to introduce a spray-painting robot commonly used in the shipbuilding industry.

When the robot was first tested on complex structures, expectations were high. What followed, however, was disappointing: its trajectory adaptation lacked sufficient flexibility, making it difficult to handle the narrow, closely spaced gaps in trash racks. In confined areas, coating coverage was incomplete, leaving sections untreated. Coating thickness was also uneven—too thick in some areas and too thin in others.

The team quickly established a dedicated task force, transforming the workshop into a testing ground. With no purpose-built control system tailored to hydropower metal structures, they began developing their own-building models and writing code from the ground up. The first round of parameter tuning did not produce the desired results. A second round of adjustments also proved inadequate. Ultimately, the team made a decisive move: they abandoned the existing approach and rebuilt the algorithmic model from scratch.

Trash racks proved to be the greatest obstacle. With their extremely dense structure and narrow gaps, conventional spraying methods simply could not penetrate effectively.

The team introduced real components as test specimens and built a full-scale experimental platform. They conducted multi-factor orthogonal testing, adjusting parameters such as spray angle, particle size, and process sequence. The process involved repeatedly stopping the system, reviewing results, and refining the code in iterative cycles. Working closely around the robot, the team monitored coating performance while continuously optimizing the control program. After more than 100 rounds of testing, they achieved millimeter-level, point-by-point calibration and successfully developed a "spot spraying" mode.

Another persistent challenge was coating sagging. In the early stages, repeated adjustments to coating formulations produced limited results. Efforts to coordinate spray-gun pressure with travel speed also proved difficult, making it challenging to balance operational efficiency with coating quality.

The team adopted an intensive trial-and-error approach, conducting test sprays, reworking results, and comparing outcomes in successive iterations. After dozens of cycles, the accumulated data filled stacks of notebooks.

After dozens of rounds of program optimization and real-world refinement, the robot developed an intelligent control system capable of adapting to the spray-coating requirements of hydropower metal structures. It moves steadily across large sluice gates, applying coatings with uniform coverage and strong adhesion. When handling complex components, it adjusts flexibly to deliver precise and consistent coverage across intricate surfaces.

In the autumn of 2025, with the final section of trash racks coated and approved, the R&D program was successfully completed. A total corrosion protection area of 5,740 square meters met all specified standards, with surface preparation achieving near-bare-metal cleanliness. Automated coating coverage reached 60% for sluice gates and up to 80% for trash racks.

More importantly, the deployment of the robot has significantly reduced workers' exposure to high-dust environments, while enhancing the continuity and stability of spray-coating operations on complex components. It marks a key step toward standardized anti-corrosion practices in hydropower facilities.

Through sustained innovation, the maintenance team behind the world's largest clean energy corridor has transformed robotic spraying from an experimental concept into a precise and reliable system. Their work is equipping one of the Key National Projects with a new layer of intelligent protection, and this is just the beginning.