FIRE’s Security Correspondent Dr Dave Sloggett explores ways in which wider benefits may be gained from a collective effort in studying the wider benefits of a systemic approach to AI in the Fire and Rescue Service

The field of Artificial Intelligence (AI) has gone through a number of waves of development over time. But one enduring message about its potential is the ability of the technology to power robots to work autonomously in dangerous environments, such as nuclear facilities.

The profession of the Fire and Rescue Service is based on the simple premise that its staff will work in dangerous environments. Be they areas at risk from flooding, forest or heathland fires, domestic and industrial fires (sometimes involving the release of toxic chemicals) and, more recently from an historical viewpoint, terrorism incidents.

Given this obvious match between the research aims of AI developers and the dangerous environments in which firefighters have to operate, it is worthwhile exploring the degree to which AI can play a role in assisting firefighters in conducting their duties.

In the companion paper to this an exploration of one specific facet of the potential for AI and Machine Learning (ML) techniques to benefit firefighters using electro-optical imagery (from hand-held/body-worn and drone-based sources) to build a picture of the centre of heat of a fire is discussed (see pg 32).

The point about data being fused from various sources, with the locations of each sensor system known with some precision into a common operating picture is clear. The sensor systems help build a 3D image of the fire. This aids decision making. It is also a great application that may also help save lives, both of firefighters and members of the public. This form of integrated analysis is based on a high degree of precision on where the individual cameras are based. Building a fused picture from which accurate decisions can be taken is not straightforward.

That high degree of accuracy may well involve establishing a base line from which Differential GPS (DGPS) signal analysis techniques can be deployed. The accuracy of this form of analysis is measured in centimetres and not the levels that are usually available, which are of the order of tens of metres depending on the situation. Bearing in mind that any approach has to fully function in both urban and rural environments.

It is important to remember that the fusion of this data into a 3D image is not an application of AI or ML techniques. That comes in the analysis of the fused picture, where the electro-optical imagery will be processed to find features in the thermal structures derived.

Once derived, such as through gradient-analysis techniques, these features might provide alerts as to when flash-over might be imminent, or how drafts are affecting the development of a fire. They will also help identify potential hot spots in the fire or a trajectory along which the fire is developing, allowing prompt decision making as to plans for evacuation of people potentially at risk.

While this is very exciting from a real-time analysis viewpoint, there are other benefits from a more integrated approach to this type of research. Perhaps one of the most significant outcomes is an integrated dataset from which all firefighters can benefit in their training programmes.

It is perhaps one of the huge ironies of the current world that due to the excellent work of firefighters in the work of preventing fires that their ability to respond to real-world events has been undermined. But as recent events, such as Grenfell Tower, have shown, the potential for a specific cocktail of extraordinary circumstances to present themselves has not diminished.

Systemic Learning

Learning from such events is crucial. If that learning is simply created by word of mouth, its effectiveness quickly diminishes. What is needed is systemic learning, where experiences are not just passed on by word of mouth at a conference or workshop but where the approaches and ideas become embedded in the training programmes, through sharing a common library of datasets derived from events.

Whereas in the past the ability to carry out such sharing has been limited, recent technical developments in 3D vision and synthetic environments for training have created an opportunity that needs to be grasped. Such a sharing environment also creates an important building block for the kind of national resilience ideas that this publication has been promoting.

One example is the Buncefield fire on December 11, 2005. In a few months this will be an event that is 15 years old. And yet it remains one of the greatest challenges the UK Fire and Rescue Service has collectively faced. Who is to say that an event of its size may not happen again?

In this Covid-19 world of remote working, what risks are we running on detaching people from industrial plants by making people work at home? What is the potential for some of the atmospheric indicators of problems at an industrial plant that would be sensed by those who work on site being missed by the benefits of home working?

Anyone that has worked at a major chemical facility will say they are aware of the normal noises that appear from its day-to-day operations. These signatures of normal operations become part of a routine pattern of life. They are also not something that the kind of contemporary industrial digital process control systems, can ever be aware of in any depth. They simply measure a parameter, like temperature or pressure.

For someone who works on site at a plant any changes in this background noise, to which their sensor systems (audio and visual) becomes tuned, will be instantaneously obvious. These cues create an opportunity for decisions to be made that can avoid the development of a major hazard. Placing operators remote from a plant, giving them access to the normal data they see on their control panels, does not enable them to understand the ‘atmospherics’ of a plant.

This argument contends that it is possible that we may yet see, as a result of the new norms emerging from Covid-19 behaviours and ways of working, a new significant industrial fire. If it were to happen, how might current commanders tap into the knowledge of the Buncefield event?

Some of those involved are probably retired or soon to leave the service. Others may have suffered Post Traumatic Stress Disorder outcomes from being in the event. This may incapacitate their ability to recall what happened, hampering the development of a swift and effective response. Arguably another building block of national resilience.

Building National Resilience

It is possible to argue that as a result of Covid-19 we need a national response to resilience like never before. Approaching innovations at a brigade level will never create the kind of mass that is needed for success to be achieved. What is required here is mass, a collective effort to achieve a national resilient capability.

Part of that mass is for brigades to pool their experiences of major events into a national library that all training suites across the country can access. This also becomes the collective corporate memory of major events like Buncefield and avoids any loss of corporate memory. Technology has enabled the window for such a central archive to be created, with imagery, videos, decision logs and personal anecdotes that will provide opportunities for systemic, long-term exploitation of real-world events.

This library could also be offered to universities for research efforts. Data collected at incidents can be made available encouraging universities to step forward and develop AI and ML algorithms that can process the data and help derive important features in the data that when exploited can help firefighters. A framework where algorithms can be plugged and played into a central training suite that each brigade provides allows for competition amongst researchers to develop the best processing algorithm that offers operational firefighters the next generation of capabilities.

Data standards will be an essential element of this, defined by those who operate the library. Rather than this being handed to one brigade it is suggested that a small consortium of brigades work collectively to establish the data standards for exchange. Brigades, such as Cleveland Fire and Rescue Service, with its need to support responses to a range of Control of Major Accident Hazards sites could be an important element. Other obvious brigades include those with responsibilities for ports, nuclear facilities, and other elements of Critical National Infrastructure. Once established this library would be hugely beneficial for Local Resilience Forums to use to plan exercises, involving partners to the Fire and Rescue Service.

One perennial problem with the development of new and exciting technologies, such as AI and ML, is that people sometimes get carried away with their potential before establishing some basic building blocks upon which their application in an operational sense can be built and where they show genuine benefits. The altar of those who think that AI and ML are the answers to many people’s needs is littered with failed projects and well-meaning efforts to make progress.

Recent history shows that it is axiomatic that robust and operationally beneficial applications of these exciting technologies only comes where the basic building blocks of success are also put in place. That way, following a process where a national effort is coordinated, applications of AI and ML techniques can be developed that help foster a systemic approach to resilience.