The structural design process is non-linear; it always requires evaluating different alternatives, iterating on designs, or reviewing modification requirements. This entire process is even more complex in foundation design, as foundations are often the last elements to be designed but the first to be built, meaning time is rarely on the engineer's side.
The emergence and widespread use of structural analysis software was a revolution in this regard, significantly reducing calculation times and increasing precision. However, everything still had to be modeled manually—a task that becomes extremely tedious when evaluating multiple solutions and searching for an optimal one. By combining a powerful and efficient calculation engine with Artificial Intelligence that automates the configuration of foundation systems, unprecedented productivity is achieved. That is Foundaxis.
Will ChatGPT configure my foundation system?
The short answer is no. For a more elaborate response, it is necessary to understand a few basic concepts of Artificial Intelligence. In principle, AI is nothing more than a tool that learns from vast datasets to identify complex patterns and then use them to perform tasks.
Think of AI image generators. These models have been trained on billions of images and text descriptions. By analyzing these images, the neural network learns visual patterns: what a dog looks like, how colors combine in a cubist style, or what textures define concrete on a construction site.
A similar logic applies to conversational AI, such as ChatGPT. These tools have read a massive amount of human text and, in doing so, have identified patterns in language: grammar, syntax, how we answer questions, how we explain concepts, or even how we write structural engineering articles. They do not truly "understand" structural engineering at a conceptual level; they replicate patterns of how humans explain these concepts.
In the same way, we can train a neural network specialized in the configuration of shallow foundation systems. To do this, we would need to start with a series of "raw" foundation systems, which are then modified by an experienced structural engineer. Thus, the neural network relates a "raw" configuration to a "refined" one, identifying patterns to replicate later.
Training the Neural Network
Training a neural network requires thousands of data points, so an optimal method for generating this data must be designed. We can generate a database of superstructures, with the coordinates of their basal nodes and reactions. But how do we define an initial foundation system? An intuitive, automatable approach that provides a wealth of information is to design an isolated footing for each load point. This initial design is not done with AI, but with a numerical optimization algorithm.
Clearly, this is not an admissible foundation configuration, but as an input, it is very useful for visualizing possible arrangements. An experienced structural engineer could make modifications to the foundation system just by looking at this image to make it feasible. What kind of modifications? For example, opting for a strip foundation instead of two very close isolated footings, or a mat foundation, or adding a foundation beam to support an eccentric footing.
By providing thousands of these input-output pairs to the neural network, it identifies patterns to achieve, to a certain extent, the replication of experienced engineering judgment. In this way, based on a new structure, it can propose various reasonable alternative solutions based on multiple parameters: bearing level depth, foundation proximity, allowable soil pressures, pressure distribution, isolated footing overlap, and many others.
Limitations of AI usage
The use of Artificial Intelligence has great potential in engineering, despite the fact that its adoption in structural engineering has been quite slow. This behavior is not arbitrary; it is a field where calculations are extremely delicate and we are accustomed to being conservative by nature. Therefore, it is no surprise that we are also cautious with the use of new technologies.
In Foundaxis, Artificial Intelligence is employed solely for foundation configuration proposals. The evaluation of foundation performance, as well as the design of steel reinforcement, is achieved through the automation of traditional design methods. Specifically, the Finite Element Method (FEM) is fully implemented in Foundaxis, considering non-linear interaction with the soil. In this way, you can be certain that the results are precise and reliable.
However, this could soon change. Artificial Intelligence has significant potential to be integrated directly into calculation engines, replacing the algorithms currently in use. This could include generating finite element meshes or even replacing the Finite Element Method itself. If trained with enough cases of input (geometry, forces) and output (stresses, deformations), an AI could be developed to predict pressures and settlements based on loading conditions without actually performing any explicit calculation.
Discover how AI-powered foundation systems configuration is integrated into the complete Foundaxis workflow:
Foundation Design Software: A Complete Guide for Structural Engineers 2026