Homotopy surrogate model assisted double-loop approach for efficient structural reliability-based design optimization

Engineers have long faced a difficult trade-off when designing bridges, aircraft, and skyscrapers. On one hand, they must ensure these structures are safe enough to withstand the unpredictable nature of the real world, where materials vary slightly, loads shift unexpectedly, and dimensions are never perfectly exact. On the other hand, they must avoid making these structures so heavy and expensive that they become impractical. The traditional way to design for safety often ignored these small uncertainties, leading to designs that were either dangerously fragile or unnecessarily bulky. To solve this, a field called reliability-based design optimization emerged. This approach treats uncertainty not as a nuisance to be ignored, but as a central factor to be calculated, aiming to find the lightest possible structure that still has a mathematically proven chance of surviving its worst-case scenarios.

The challenge with this method is that calculating safety is incredibly expensive in terms of computer time. To know if a design is safe, engineers must simulate thousands of different possible realities, checking how the structure behaves under every combination of slight variations. Doing this for every single design tweak during the optimization process can take weeks or even months of computing power, making the method too slow for complex, real-world projects. Researchers at Wuhan University of Technology have developed a new way to speed this up without sacrificing accuracy. They created a method that acts like a highly skilled shortcut, allowing engineers to predict how a structure will behave under uncertainty without having to run the full, heavy simulations every single time.

The core of their new method is a mathematical tool called a static homotopy surrogate model. In the past, engineers trying to speed up these calculations often relied on data-driven models, which are like trying to guess the weather by looking at a few past photos. These models need a massive amount of training data to learn the patterns, and if the data is messy or incomplete, the predictions can be wrong. The new tool takes a different path. Instead of learning from examples, it builds a direct, explicit map between the uncertain inputs and the structural response using a specific mathematical expansion. This means it can generate accurate predictions almost instantly once the initial setup is done, without needing a large library of past data to teach it. It is a more fundamental approach that understands the physics of the problem directly rather than just memorizing the results.

To make this tool even more efficient, the researchers added a clever reuse strategy. Imagine an engineer designing a new bridge. As the computer tests different versions of the bridge, it often encounters designs that are very similar to ones it has already analyzed. Instead of building a brand new prediction model for every single new design, the new method checks its history. If the new design is close enough to a previous one, the system simply reuses the old prediction model. This avoids the need to rebuild the complex mathematical map from scratch, saving a tremendous amount of time. The researchers set a strict rule for when this reuse is allowed, ensuring that the shortcut is only taken when the designs are similar enough that the old map remains accurate.

The team tested this approach on four different problems, ranging from simple mathematical puzzles to complex real-world structures. They started with a highly nonlinear problem, which is a type of difficult equation where small changes can lead to huge, unpredictable results. They also tested it on two different truss structures, one with ten bars and another with seventy-two bars, which are common frameworks used in bridges and towers. Finally, they applied it to a detailed model of a fighter jet wing box, a structure made of thousands of individual parts. In every case, the new method found designs that were just as safe and efficient as those found by older, slower methods.

The results showed a dramatic reduction in the time required to find the best design. For the ten-bar truss, the new method reduced the number of heavy computer simulations needed by more than 99 percent compared to some existing techniques. For the seventy-two-bar truss, which involved twenty different uncertain factors, the savings were even more pronounced, cutting the computational work by nearly 99 percent. Even for the complex wing box, which involved a massive number of parts, the method managed to find a lighter, safer design while using only a tiny fraction of the computing power usually required. The researchers found that by reusing their prediction models, they could skip building new ones for hundreds of design iterations, effectively doing the work of a much larger team in a fraction of the time.

This work suggests that the bottleneck of high computational cost in safety-critical design is not an unsolvable problem. By combining a physics-based prediction tool with a smart strategy for reusing past work, engineers can now explore a much wider range of design possibilities. This means that in the future, structures could be designed to be lighter, more efficient, and safer, without the prohibitive cost of running endless computer simulations. The method does not replace the need for rigorous safety checks, but it removes the barrier that has kept these advanced techniques out of reach for many complex engineering challenges.