Experience from Pilot Implementation of Substation Inspection Robots in Thailand’s High Voltage Substations

Each robot was equipped with high-resolution visual and thermal cameras, along with an image processing algorithm that could extract gauge values and status indicators from the images such as gas pressure readings or the open and closed positions of circuit breakers. The system could automatically read analog gauges, check mechanical status and LED states, and even detect simple anomalies, such as bird nests on transformers. After each run, the robot generated a report and triggered alarms according to given criteria. These included conditions like low gas pressure, a change in circuit breaker status, equipment hotspots and other user-defined criteria.

In one test, the robot successfully identified a mock bird nest placed on a transformer as an anomaly, and in another, it extracted a warning-level dial reading from a test SF₆ pressure gauge not connected to any system, confirming that it could both interpret and respond to visual thresholds.

Two substations were selected for this trial project. Ayutthaya 2 is an outdoor air insulated substation (AIS) operating at 115, 69, and 22 kV. The switchyard is a mix of concrete service roads and compacted gravel. A few paths had to be improved so the wheel-driven robot could navigate properly. Fencing around the transformers, installed to prevent animal access, also blocked the robot’s view of certain inspection points, such as the winding temperature gauges.

Bang Pa-in 1 is a compact indoor gas-insulated substation (GIS) at 115 and 22 kV. Being an indoor type, it offered a better environment for the robot, with flat floors, consistent lighting, and no weather-related variables. At both sites, the robot completed over 150 inspection rounds and covered more than 600 individual points, including transformers, circuit breakers, current and voltage transformers, and surge arresters. The result was higher inspection frequency, more detail captured at each point, and reduced routine workload for the operator. All data were stored as images and values extracted by the robot, which made it easier to verify issues, monitor trends, and plan maintenance with real evidence in hand.

Since thermal inspections are usually performed every six months, having the robot scan more frequently helped detect early signs of abnormal heating that might have otherwise gone unnoticed, such as a temperature imbalance found on a 22 kV disconnecting switch at

Ayutthaya 2, which later led to feeder maintenance. Although experienced personnel were still needed to confirm results, having regular thermal images gave more lead time to act before problems escalated.

The trial also brought up a few practical limitations in using the robot. A lot of its performance depended on the quality of the images it captured, which could vary depending on lighting, the angle of view, or how easily it could access the equipment. When the robot inspected equipment types it hadn’t been trained to recognize, the readings were inaccurate or it couldn’t read them at all. This highlighted the need for additional tuning or broader training data to help the system recognize those devices properly. Thermal imaging had its own set of challenges. The robot could scan and flag spots that looked unusual, but the results still needed a trained person to double-check and confirm what was really going on. As for movement, the robot handled most routes without trouble. It could adjust its path when alternate routes were available, but in tight areas, the robot couldn’t actually avoid the obstacle and had to take a detour. Clearly marked paths and well-arranged equipment helped both the robot and the operators carry out inspections more smoothly. At this point, the robot is a helpful support tool that can take care of regular inspection tasks under limited conditions and give an early warning when there are potential problems, but it still relies on human expertise to ensure and verify issues.

In conclusion, the trial project confirmed that substation inspection robots can effectively support routine tasks, enhance inspection frequency, and serve as an early warning system for potential equipment issues. While human expertise remains essential for final verification, these results demonstrate that robotic inspection is ready to move from pilot stage to practical use in daily substation maintenance