NNAISENSE moves into AI-based city digital twins

NNAISENSE moves into AI-based city digital twins

Technology News |
By Nick Flaherty

Industrial AI and sensor startup NNAISENSE is to create a digital twin of the city of Lugano in Switzerland.

The digital twin, created using 3rd Generation AI with evolutionary reinforcement learning, will enable the municipality to establish a virtual “situation room” to solve real world issues such as pollution and congestion, as well as test potential policies in a digital “sandbox” before enacting them in the real world.

NNAISENSE is a 2015 spinout of the Swiss IDSIA (Istituto Dalle Molle di Studi sull’Intelligenza Artificiale) research lab at Lugano, developing recurrent neural networks (RNNs), long short-term networks (LSTM) algorithms and HighwayNets (see below). It has so far focussed on optimising industrial processes using massive sensor networks and third generation machine learning to create sophisticated custom models without the need for training. This is now being applied to city-wide digital twin networks.

An RNN is a network of neurons with feedback connections that can learn many behaviours not achievable by traditional machine learning methods. Artificial RNNs can learn algorithms that map inputs to outputs with or without teaching and so are computationally more powerful than other adaptive approaches.

“We have a focus on the industrial space for monitoring, process control and fully autonomous control that involved the two main technologies of neural networks on GPUs and modelling,” said Faustino Gomez, CEO and one of the founders of NNAISENSE.

“The digital twin comes in as a special case of the process modelling effort,” he said.  “Because of what you can do with supervised learning and DNN what you end up learning is a process model that learns the dynamics of the system you are modelling and learns the behaviour direct from the data. We have a third generation digital twin – you may augment it with data for predictive maintenance or as a simulator to improve control but what you typically don’t do is use the data to create the model, you work from first principles or physics models. We have done things that you can’t do that way, direct from the data,” he said.

The company has been working with a number of process technology companies, one making the glass vials for Covid-19 vaccines for example, another operating wind turbines and a third using additive manufacturing 3D printing machines. The company uses the data to build bespoke models of the manufacturing process or operation so that it can be optimised.

“You need sufficient sensors that can characterise the performance so you have a high dimensional time series, and rather than predict a category, you want to predict the future inputs. You would use walk forward testing with out of sample data,” he said. “You are predicting future values of the inputs and from that you can get a model.”

“We have a model that learns the physics and the chemistry that specific to a machine, things that are difficult to model and code into a simulator. That was a really cool achievement as it was created entirely with the data,” he said.

The company raised an undisclosed Series B round of funding in 2019. Led by Samsung Ventures along with Repsol Energy Ventures and glass-making customer Schott, part of Zeiss.

Jaan Tallinn, a founder of Skype and Kazaa increased his investment through his Metaplanet fund.

Back in 2015, the team invented Highway networks, the first deep networks capable of being trained with over 100 layers. Very deep networks can allow machines to better understand context for a city-based digital twin.

Other articles on eeNews Europe


If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles