Multi-point monitoring of PM10, PM2.5, and PM1.0 at a combined-cycle power plant: field agreement between four co-deployed electrostatic monitors

Air is rarely empty. Even on a clear day, it carries a suspension of tiny solid and liquid particles, ranging from coarse dust to microscopic soot. These particles, known as aerosols, are not just a nuisance; they are a health hazard and a mechanical threat. When inhaled, the smallest fractions can penetrate deep into the lungs, while larger ones can clog the delicate machinery of industrial engines. For workers in power plants, where massive turbines draw in vast quantities of air to generate electricity, knowing exactly how much of this dust is present, and how big the particles are, is a matter of safety and efficiency. If the air is too dirty, the machines foul up, losing power and burning more fuel. If the air is too dirty for the people working around them, their health is at risk. Yet, for decades, measuring this dust in real-time at multiple spots inside a single industrial site has been difficult, expensive, and often limited to just one location.

In a combined-cycle power plant in Samut Prakan Province, Thailand, a researcher set out to change that. They deployed four identical, custom-built monitors at different locations across the facility to watch the air continuously for over four months. Their goal was twofold: to see if these new, affordable devices could track the dust levels that threaten the plant's turbines, and to see if four identical machines, sitting in the same general area, would agree with one another. The answer they found was nuanced. The machines worked well enough to show the big picture of how the air changed over time, but they did not agree perfectly on the exact numbers from day to day. This discovery suggests that while a network of these sensors can tell a plant manager when the air is generally getting dirty, it cannot yet be trusted to tell them precisely how much dirtier one corner of the plant is compared to another without careful cross-checking.

The researcher used a specific type of sensor that relies on electricity rather than light or weight to count particles. In these devices, air is drawn in and passed through a chamber where the particles are given a tiny electric charge, much like how a balloon rubbed on hair picks up a static charge. Once charged, the particles are separated by size using a simple physical principle: heavier particles hit a wall and stop, while lighter ones keep going. The remaining charged particles are then caught on a filter, and the electric charge they deposit is measured to calculate their total mass. This method is fast, robust, and can be built into a compact unit that can run unattended for months. The researcher adapted an existing design, which had previously been tested in a coal mine, to measure three specific sizes of dust: the larger particles that settle quickly, the medium-sized ones that stay suspended longer, and the finest, sub-micron particles that are hardest to catch but most likely to reach the deepest parts of the lung.

Four of these units were installed in a shelter at the power plant, each positioned to sample air from a different part of the site. One stood near the main turbine hall, another by the air intake ducts, a third near the fuel transfer pier, and the fourth at the electrical substation. They ran continuously from February to June 2020, logging data every hour. This period covered the tail end of Thailand's dry season, a time when regional haze from agricultural burning often drifts over the country, before the rains arrived to wash the air clean. The monitors recorded data on 92 to 100 percent of the days, providing a rare, high-resolution look at the air quality inside the plant boundaries.

The results showed a clear seasonal pattern. The dust levels were highest in February, with the larger particles reaching daily averages of up to 160 micrograms per cubic meter, and the fine particles peaking at over 110 micrograms per cubic meter. As the months passed and the wet season approached, the levels generally dropped, though the fine particles showed a slight, temporary rise in May before falling again. This pattern matched the broader regional trends of the dry season haze, suggesting that the air inside the plant was largely influenced by the weather and the surrounding landscape rather than just the plant's own operations. The researcher also noted that the dust levels did not change significantly between weekdays and weekends, further indicating that the air was being shaped by regional factors rather than the daily rhythm of the plant's work shifts.

However, the most significant finding of the study concerned the agreement between the four machines. Because all four units were identical and sampled the same general air mass, the researcher expected them to report nearly the same numbers every day. Instead, they found that the machines often disagreed. On any given day, one unit might report a dust level that was more than three times higher, or less than one-third, of what a neighboring unit reported. This disagreement was not caused by a single faulty machine; all four units showed similar variations. It was also not because the air was different at each spot; the differences were too large and too random to be explained by small-scale variations in the wind or local sources.

The study concluded that the disagreement was due to the instruments themselves behaving slightly differently over time. While the machines were excellent at tracking the general trend—showing when the air was getting cleaner or dirtier—they were not reliable enough to compare the exact concentration at one specific point against another on the same day. If a plant manager looked at the data from just one machine, they might get a poor estimate of the true air quality at that spot. But if they averaged the data from all four machines, the estimate became much more reliable, capturing the true state of the air with a high degree of confidence.

This distinction is crucial for how such technology is used. The study proves that a network of these low-cost, electrostatic monitors can provide a continuous, multi-point view of air quality that was previously impossible to get without expensive, complex equipment. They can successfully track the seasonal cycles of pollution and alert operators to periods of high dust loading that could damage turbines or harm workers. However, the study also rules out the idea that these machines can be used to pinpoint exact differences between two locations without a rigorous calibration process. The researcher emphasized that before these devices can be used to make critical decisions about filter maintenance or worker safety, they need to be tested side-by-side against standard reference instruments and checked against each other regularly to correct for their individual quirks.

The data also highlighted a gap in current safety standards. The average dust levels recorded at the plant, particularly the fine particles, frequently exceeded the health guidelines set by the World Health Organization and the Thai government. While these levels were far below the limits for industrial dust that would cause immediate harm to workers' lungs, they were high enough to suggest that outdoor workers at the plant were breathing air that was unhealthy by modern standards. The study did not measure how much dust individual workers actually inhaled, but it provided the first continuous record of the environment they worked in, showing that for much of the campaign, the air was consistently above the recommended safety limits.

Ultimately, this research demonstrates a practical step forward in industrial hygiene and environmental monitoring. It shows that it is possible to build a network of sensors that can watch the air at a large facility around the clock, capturing the complex dance of dust and weather. But it also serves as a reminder that in science, having more data points does not automatically mean having more agreement. The machines worked, but they did not speak with one voice. For the data to be truly useful, the human operators must understand that the value lies in the collective average, not in the specific reading of any single device. This insight will guide future deployments, ensuring that when these sensors are used to protect workers and machines, they are used in a way that respects their limitations and leverages their strengths.