Scientific information is often difficult to understand, full of technical terms and and complicated formulas. We have talked to civil engineer Jan Onne Backhaus about evolutionary algorithms for our current article in a way that everyone – from geneticists to computer scientists to elementary school children – can understand it. Promise.

*********

When the Queen of Hearts said to little Alice “In this country you have to run as fast as you can if you want to stay in the same place”, she wasn’t thinking of evolution – but author Lewis Caroll was. In 1973 Biologist Leigh Van Valen ultimately created an evolutionary biological theory from the statement, according to which organisms are in a constant “arms race” with one another. It is time for the construction industry to move with the times as well, thought Onne and developed an algorithm that creates more efficient processes and more precise forecasts for the construction sector.

Evolution is actually quite simply explained a development, the change of characteristics of a population over the course of generations, where, in the optimal case, they get better and better through selection, recombination and mutation. The principle is mostly understood in a biological sense, but can also be applied to other areas, e.g. the construction site. “Evolutionary algorithms are about solving a problem the way evolution does it,” explains Onne using an easy-to-understand example: “Assuming that a dog breeder would want to breed dogs with very long legs. Then you would choose from the dogs you have those who already have very long legs and cross them with one another. From the puppies of this parent generation, the next step would be to select those individuals who have the longest legs and cross them again. “

“The process approaches the optimal solution with an increasing number of iterations. It is important to understand that in the end we usually get very good solutions, but often they are not optimal solutions in the mathematical sense. In civil engineering, however, they often don’t have to be. It is usually enough if they are almost optimal”, emphasizes Onne.

**Game, set, victory**

One technique that is used again and again in construction is the so-called Monte Carlo simulation, named after the infamous casinos of Monaco. The method was developed in the 1940s by scientist Stanislaw Ulam, who got the idea while playing a round of solitaires and wondering what the odds of a successful Canfield solitaire (a casino game with very low odds) at 52 Cards were. “After spending a lot of time trying to estimate them by pure combinatorial calculations, I wondered whether a more practical method than “abstract thinking” might not be to lay it out say one hundred times and simply observe and count the number of successful plays. This was already possible to envisage with the beginning of the new era of fast computers”, Ulam is quoted in” Stan Ulam, John von Neumann, and the Monte Carlo method “(Eckhardt, Roger (1987)).

This technique solves problems that cannot be solved analytically or only with a very large expenditure of time, with a large number of random experiments. “Assuming we have a construction site,” Onne gives another example, “where a lot of different activities are carried out. However, these activities or processes do not always take the same length. Sometimes it rains and a task takes longer, other times the team is very motivated and the construction process is shortened. These processes are often very interdependent. For example, the roof of a hall can only be built when the walls are already in place. This makes it very difficult to estimate the average expected total construction time, even if you know the stochastic distribution of the individual processes. This is where the Monte Carlo simulation comes into play. As part of the simulation, we calculate a large number of total construction times, whereby in each run we draw the random duration of each individual process from a set of possible process times. The result is then a large number of theoretically possible total construction times. With the Monte Carlo simulation, we assume that such total construction times, which we have calculated very often, will also be realized with an increased probability. In order to calculate the expected, mean total construction time, we only have to calculate the mean value of the calculated solutions. “

**From theory to practice**

Testing these theoretical considerations in practice is not that easy, says Onne. For the simulation, input data has to be collected and digitized. “So it is necessary to either collect it yourself or to find someone who is already doing this – and is also willing to share their data. In my very specific case, I was lucky enough to run into a Renesco GmbH project manager, Sewerin Sabew, at a conference. He told me about his injection project in the Feuerbach tunnel and the collaboration with eguana. “

As part of the Stuttgart 21 construction project, a two-tube tunnel leads from Stuttgart’s main train station to the Feuerbach train station. The tunnel is partially located in the anhydrite-bearing mountains, which made several thousand sealing injections necessary to prevent water ingress. For everyone of these injections, eguana collected process parameters such as flow rate and pressure via SCALES and automatically derived the corresponding manufacturing and secondary processes. Together with various metadata, these were daily made available to Onne in a corresponding format in a separate cloud. “All in all, it was documented down to the second what each pump had done for every second of the project.” It was clear to Onne that a real treasure trove of data had been recovered, and he immediately set about utilizing the potential of the digitized data.

“First of all, I programmed a model that is able to predict the construction times and construction costs of the more than 300,000 injections. The model works with statistical distributions for the duration of the individual processes (injecting, relocating the pumps, etc.) and also takes maintenance cycles and downtimes into account, for example due to technical problems.” An extended version of the model even takes into account the geometry of the two-tube tunnel one – for example, the injection units could get in each other’s way in the narrowness of the tunnel.

**More accurate than the construction schedule**

Onne’s model was then compared with the real construction site data to determine whether the prediction was correct and the model provided the right results. “It was really exciting for the site managers,” said Onne, delighted about the response that his work had received. “Until then, they mainly used the data to document the many thousands of boreholes, to ensure their quality and roughly extrapolate construction progress into the future.” To formulate statements about the future (“With a probability of 95 percent we are done on day X and that costs us Y euros ”) was not yet possible.

“After the model delivered good results, we started optimizing. A big question was what the optimal number of machines and their optimal period of use would be with a view to the total construction time and construction costs. This is where the evolutionary algorithm and the Monte Carlo simulation came into play. ”With his model, Onne was able to show that the number of machines was well chosen, but that there were still a few among the millions of hidden parameter combinations that “ the project result would have further improved. Finding these solutions without simulation would have been next to impossible.”

It was too late for the Feuerbach construction site. But the gain in knowledge of what can be salvaged from data treasures and what more flexible models allow in calculation and construction contract can be an asset for future projects.

*********

Our thanks for today’s blog post go to Jan Onne Backhaus on the one hand for his expertise and his ability to put the most complicated processes into simple words – on the other hand, our thanks also go to Ella, his short-legged dog lady, who made the search for descriptive examples so easy!

**About Jan Onne Backhaus:**

Born in Hanover, Onne loves culture clash – maybe one of the reasons why he spent a semester abroad in Vienna in 2004? The civil engineer, MBA and doctoral candidate from the Technical University of Hamburg not only likes to expand his horizons in geographical terms, but also in cultural terms, which is why he likes to attend Spanish courses in his free time, meditate or win prizes at the International Toastmasters (which, admittedly, has got nothing to do with slices of toasted white bread but is the name of an NPO promoting the art of public speaking).

As a senior consultant at Drees and Sommer, he has recently been optimizing construction sites with lean construction – another of his heartfelt topics.

**About Ella:**** **

When she’s not chilling on the sofa, she prefers to spend her free time in nature. Excavators are one of her greatest passions.