Realtime – living models & digital twins
By not just presenting the space – but by driving real-time modelling with live data it offers incremental and perpetual system capable of fusing data and providing actionable and qualifiable intelligence and situational awareness.
Simulation allows a range of what if scenarios and to utilise simulation to suggest a maximum likelihood model using partial or synthetic data – augmenting response and enabling training and planning.
This work involved us creating a range of materials and objects within the simulation that were able to interact with Radiation.
We were able to show that a radiation source could paint other areas in the scene as well as profile the points that a ray might strike – or assist in understanding how a measurement might be used to infer or model the 3D perspective of a source.
Whether derived from Mesh, Point Cloud or Vector data – our system breaks down the world into 3D points (presented as Voxels) and stored in an ultra-fast spatial database.
This permits us to scan and update models – and begin to add semantic data to drive increasingly realistic models augmenting them with real-time data and agents, providing insight and situational awareness.
In effect creating a data driven sandbox for interactions, planning and a living representation of different events in larger systems – capable of comparing disparate datasets to extract knowledge and insight by leveraging the volumetric interplays between them.
We began by creating a synthetic environment – showing two cut away old style water pressure reactors inside of a building – to allow us to factor in the propagation of a source against different types of material – using the Density of the different values to allow the realtime sim to then spread out – with different materials attenuating different strengths and types of emitter.
This allowed us to then expand to produce a whole port scene facility – with a view that we could then generate a vareity of scenarios and factor in damage or change to both systems and the scene based on potential events. Eg what might happen if power was suddenly cut, or a vessel ruptured.
Working at Cm scale – this gives us approx 60 Billion Voxels to work with inside of an 8k Cube (typical – on i9 & 4080ti) 8k-x-axis 8k-y-axis x2k-z-axis 1080 HD Volume is 8,000 – 30,000 voxels – so more than enough to perform the realtime calculations for a live sim, profile behaviour and write back to the DB on any changes or updates.
We then applied this a real environment using a scan from the M-Shed Museum – a large concrete structure with many different machine workspaces – collected as part of our work on 5G-Victori
We also experimented on the 200GB scan of the Berlin HBH – simply to trial penetration and modelling of other types of EM – e.g RF penetration and propagation – that behaves differently, but still follows similar rules (more wave like behaviour).
This modelling approach allows us to then consider the movement of people inside of a 3D or 2D space – as well as their access to connectivity and data. Which can be modelled in real-time in the case of an event or change – which is very similar to the 5G-Nomadic node work done within 5G-Victori’s trials.
2D, 3D, & 5D models
The above image was showing the potential risk mapping of a contamination event – using the Terrain engine to model flow and potential spread.
Every point in space has a value and via our forthcoming scripting engine and low code tool – can be interrogated, updated and fed with real data. This allows for complex simulations to be chained to provide useful simulations that can grow with complexity over time – and interface into other tools.
Many of the logistical challenges that Polaron was designed to address are common to Nuclear – such as interchanges to other transport modalities, modernization and changes in policy and approach.
Mapping and modelling also helps to permit faster proposals and helps to address increasing financial pressures and long term planning. This approach also lets us interface into other external models – whether that’s plume models from local sensors or larger MET data models.
Use case – Australian Caesium-137 mining sample loss applied to Modelling contamination on railway lines
https://www.bbc.co.uk/news/world-australia-64483271
In Jan 2023 – a sample of Caesium 137 was lost in Australia in a mining facility. This prompted a huge search across 1400km.
Emergency services had “literally found the needle in the haystack”, they said. A huge search was triggered when the object was lost while being transported along a 1,400km (870 mile) route – between between 10-16 January (6 day window – 12th now likely).
Detected lost on the 25th Jan.
Discovered / Recovered 1st Feb
A 20 day window between likely loss and recovery
The search area “…roughly equivalent to the distance by road from John O’Groats in northern Scotland to Land’s End in south-west England … searching for a “…Capsule (is) 6mm (0.24 inches) in diameter and 8mm long and contains a small quantity of Caesium-137, which could cause skin damage, burns or radiation sickness…”
Fortunately the sample was eventually found – however it posed an interesting example and use case of how simulation might augment the response.
We used this as the basis of a demo focussed on the railway line around Sellafield NPP – to then approximate where a similar loss might occur and how we might augment the process of detection.
Open Street Map data used to model rail and pathways around Sellafield NPP to model potential propagation along those paths – and focus on the likely area of contamination / most probable location.
We used a model of the Railway system surrounding Sellafield to estimate the location of a source or contaminant – and thus can infer and predict what areas are likely to have been contaminated and thus focus limited resource or prioritise them by augmenting decisions with simulation.
Polaron – driven by sensor data and human input – is not just a 3D tool – but an interface into a big data engine capable of running reports, models, simulations and create a tangible interface and data driven twin of the immediate and wider area of interest. Which in real terms means being able to feed in live sensor data, statistics and other biases – to allow our engine to calculate the maximum likelihood of a given location or contamination – based on a variety of factors – and even allow users to “paint in data” as reports or human insights allow for rapid modelling (e.g. where are the truck stops, anecdotal data on the condition of the road or railway surface, potential breaking or accelerating events).
Scaling up the search or query area
Small issues – can quickly grow to affect larger areas.
We then looked to model a larger area – focussing on a City – in this case NYC – where an issue or incident might quickly expand to not just effect a a building, but grow to affect multiple km2 and the surrounding areas.
At 30km2 and into 100km2 and potentially consider the impact of systems across whole continents.
Polaron is designed to scale and permit different scales of simulation to interact at different layers – ingest data from other tools – such as the MET office weather data – or large transit and situational awareness C2 functions to augment decision making and facilitate planning.
A key benefit of Polaron is to allow these different scales of simulation to interact. Whether that’s 3D – Realtime into a terrain model or pushing data from a site scale model > county > region > national and continental scales.
At the time of writing we have unpacked both the OSM DB of Europe as well as the North American OSM DB – essentially creating an 1m2 interactive semantic raster of the OSM nodes.
Bristol & Bristol Airport test – presented at 3 x different scales and resolutions within a single simulation – they can also be connected into a federated system via ports or over Web-RTC or similar connection to stream data and present the interface.
Paris @1m2 resolution with a basic journey map plan presented.
Simulation offers insight in the face of uncertainty
There are limited sensors to directly detect contamination and these are clustered around key sites and key location.
What happens if a contamination event occurs in transit ?
We also believe that smaller next generation reactors will likely require the greater movement of material to and from different sites and the risk of an accident is increased. As well as potentially other considerations with more sites spread around – and their supporting infrastructure.
Modelling not only the primary effect but also secondary effects is key to focusing resources and evaluating risk.
We also wanted to demonstrate how a Voxelised model can be used to consider other types of contamination, such as water based contamination and to show how the Voxelised model – allows for the “flow” and spread of a contaminant – could very quickly then move to cause issues into other systems, water courses or other impacts.
It is possible to consider low fidelity models interfacing with higher fidelity simulations and to begin to combine them.
This allows for vary Levels of Detail or can factor in the particular known behaviours of given materials. Polaron is not a flood modelling tool – but can interface with them – offering a single platform that can interact with lots of other systems. This allows for the comparison of different values to predict likely problem areas or possible risks. Novel zero-trust network systems present this sort of federated approach as well as blending en-prem and cloud based solutions for greater resilience.
In this instance – painting in a flooded area – which could then allow for a comprehension of the scene – with more data being able to be added as a response was formulated.
Realtime feedback and input over mobile devices allowing edge or mobile inputs and interfaces into other tools.
Part of our work with 5G-Victori is the ability to pull data from a variety of sources as well as inputs. And factor this into our tool in real-time – essentially having access to a real-time GIS simulation tool from your mobile device.
In this more simple example – we were experimenting with road closures and route mapping – allowing the end user to add and then calculate how their journey might be planned to avoid or re-direct around a given area based on semantic data they had added.