High-Dimensional Digital Twins

If you’ve ever done any sort of structural inspection, you know that we collect a LOT of data over the life-cycle of a system, whether it’s a bridge, a pipeline, or a ship. The problem is, we tend to look at each information source in a vacuum, and most of our analysis isn’t particularly quantitative. What we propose it to integrate all of this data into a digital twin of a given system. A digital twin is a living model, one that can be used for numerical simulations, visualizations, or just complex data management. Inspection data gets layered onto this model and then analyzed.

Conceptual HD Twin for a ship structure

Things start getting interesting once you’ve formed the model. We’re exploring methods of mathematically fusing these data together, using a combination of pattern extraction and high-dimensional machine learning techniques. Basically, for each data type, we start by describing critical data (say, a flaw) as a set of numerical parameters, sometimes referred to as a feature vector. This helps to reduce data size and makes for more repeatable analyses. We’ve already done this for image and image-like information. If different observations of the same flaw exist, these can be stacked into one larger vector and then tracked through high-dimensional geometric analysis.

More to come!