MAC-Penalised Fisher Information Criterion for Optimal Sensor Placement in Structural Health Monitoring of Lattice Tower Structure

Structural health monitoring is the practice of keeping a constant, listening ear on the world's largest civil structures, from suspension bridges to the towering lattice masts that carry power lines and telecommunications signals. These structures are constantly vibrating, even when the wind is calm, and those vibrations hold a secret code about the structure's internal condition. By measuring how a building sways or a tower twists, engineers can identify its natural frequencies and the specific shapes it takes when it moves, known as mode shapes. If a part of the structure weakens or cracks, these vibration patterns change. To read this code, engineers attach sensors to the structure, but they face a difficult physical constraint: they cannot place a sensor at every single point on a massive tower. They must choose a limited number of locations that capture the most useful information. If they choose poorly, the sensors might end up listening to the same vibration pattern twice, or worse, they might miss the subtle differences between two very similar movements, making it impossible to tell if the structure is healthy or in danger.

For decades, the standard method for choosing these locations has been to maximize the amount of information gathered, a strategy that works well in theory but often fails in practice on complex, symmetric structures. This traditional approach tends to cluster sensors in areas where the strongest vibrations occur, ignoring the fact that some vibration patterns look almost identical to the sensors. When two different vibration patterns produce nearly the same signal, the system becomes confused, a problem known as modal aliasing. In a recent study, researchers at Tongji University in Shanghai tackled this issue by developing a new way to place sensors on a 60-meter lattice tower. Instead of just chasing the most information, their new method actively penalizes the sensors for being too similar to one another. They created a system that balances the need for data with the need for distinct, clear signals, ensuring that every sensor adds a unique voice to the chorus of the structure's movement.

The researchers tested their idea on a detailed computer model of a steel lattice tower, a structure resembling a giant, tapered X-braced ladder standing 60 meters tall. This tower has 44 nodes where sensors could potentially be placed, and the researchers were tasked with selecting just 12 of them to monitor eight specific vibration modes. The standard method, known as Effective Independence, would typically spread these sensors out to capture the strongest movements, often placing them high up on the tower where the sway is most dramatic. However, this traditional approach resulted in a critical flaw: two specific vibration modes, which occur at very similar frequencies, became indistinguishable. The sensors were so clustered in their listening habits that the system could not tell these two modes apart, with a correlation so high it was nearly perfect.

The new method, which the author calls MAC-penalised Fisher Information, changes the rules of the game. It introduces a penalty for placing sensors in a way that makes different vibration modes look alike. By running a step-by-step selection process that constantly checks for this confusion, the algorithm forces the sensors to spread out in a way that keeps the modes separate, even if it means sacrificing some raw data volume. In the simulation, this approach successfully reduced the confusion between the two problematic modes by about 24 percent. The trade-off was that the sensors ended up clustered in the lower 40 percent of the tower, leaving the top half unmonitored. While this might seem counterintuitive for a structure that sways most at the top, the researchers found that this specific arrangement was necessary to untangle the confusing signals of the lower, more complex vibration modes.

When the researchers simulated real-world conditions by adding random noise to the measurements, the results revealed a nuanced picture of reliability. In environments with high levels of noise, such as a windy day, the new method performed just as well as the traditional one, and in some cases, slightly better. However, in very quiet conditions where the signal is clear, the traditional method produced a slightly more accurate average result because it had gathered more raw information. The true advantage of the new method appeared in the consistency of the results. While the traditional method occasionally produced large, erratic errors in its worst-case scenarios, the new method remained steady and reliable. The range of errors was much tighter, meaning that engineers could trust the system to perform consistently, even when the conditions were not perfect. This reliability is crucial for safety, as it ensures that the system does not occasionally fail to detect a problem just because of a rare, unlucky combination of noise and sensor placement.

The study concludes that for critical infrastructure like lattice towers, where the cost of missing a structural flaw is high, the ability to consistently distinguish between similar vibration patterns is more valuable than squeezing out every last bit of raw data. The new method offers a way to design sensor networks that are robust against confusion, providing a clearer, more reliable picture of a structure's health. While the approach requires a compromise in spatial coverage, concentrating sensors in the lower sections of the tower, the gain in signal clarity and consistency suggests a significant step forward for structural health monitoring. The researchers note that future work will involve testing these ideas on physical towers and refining the method to handle structures that change over time, but the simulation results already offer a compelling argument for rethinking how we listen to the world's tallest structures.