Fire prevention and detection

Sensors In Fire Detection

Fire detection and simulation systems are growing more sophisticated and span larger distances.

The last 10 years or so have produced some colossal and deadly fire events that have destroyed whole towns, burned a record amount of acreage, and polluted skies for weeks. And wildfires are not just happening in the Western United States but have burnt out of control in Europe, the Amazon, and Australia.

Early wildfire detection and forest management via controlled burns are two ways to prevent wildfires from getting out of control. Sensor systems, networks, artificial intelligence, simulation, and modeling all have a role to play. Governments and governmental agencies have acted as catalysts to spur on development of models, sensors, and networks. Commercial offerings continue to tackle problems of power, cost, automation, and network accessibility for remote coverage areas.

Detection, inside and out
Gas and particulate sensors, cameras, lidar, and microphones all play a part in wildfire detection. An indoor fire detection system might have optical sensors that look for smoke, thermal sensors that detect rising temperatures, and sensors for CO and CO2 gas. Photoelectric smoke detectors use a light beam to detect smoke from smoldering fires, while ionization smoke detectors use electrically charged particles, or ions, to find smoke from flaming fires. Dual-sensor smoke detectors have photoelectric and ionization detectors.

More types of sensors are being added to indoor fire detection systems, such as motion detectors and audio systems the detect breaking glass. The basics of an indoor automatic fire alarm system include a receiver, automatic sensors, manual transmitters, sound devices, fire doors, fire shutters, smoke shutters, and network devices to which they are connected.

“Our customers are integrating IAQ (indoor air quality) sensors into fire and carbon monoxide detectors to add additional value,” said Dave Simpson, director of marketing for industrial sensing at Renesas Electronics. “We are seeing metal oxide for IAQ and humidity/temperature sensors” being used and developed into systems.

Fig. 1: An indoor smoke detector uses light to detect smoke particles. Source: Renesas

Fig. 1: An indoor smoke detector uses light to detect smoke particles. Source: Renesas

A wildfire detection system might have the same types of sensors, but it’s weather-proofed, lower power, and it typically is part of a network of many devices transmitting over a large area. The sensor systems have to be inexpensive enough and independent enough to stay running for 10 to 20 years without maintenance.

Visible data and long-wave infrared (LWIR) detecting cameras, often called forward-looking infrared (FLIR), play a large role in wildfire detection, looking for two things — smoke and heat.

In addition, heat-sensing cameras are being used in early fire detection (EFD) systems, mostly in industrial situations but that have a large coverage area.

MoviTHERM, for example, has developed an EFD system that uses FLIR cameras that detect heat before smoke and flames exist. A gateway connects the cameras to other sensors and that gateway connects to cloud software that is trained to recognize heat before smoke and fire emerge. Used in indoor industrial settings or outdoor areas to monitor coal or biomaterial piles that are susceptible to spontaneous combustion, those FLIR camera systems can detect the heat early and sometimes measure the temperature.

“MoviTHERM integrates fixed thermal imaging or infrared camera systems. It can also connect to other detector and sensing devices for fire detection, like a fire detector or maybe even a sprinkler-type system,” said David Bursell, the company’s vice president of business development. “The main applications that we’re focusing on is more for fire prevention, or very early fire warning fire detection.”

AI is used to train a system to disregard some heat signatures, such as exhaust from a vehicle. “Anytime we use thermal imaging, and anytime that piece of equipment comes into the scene, it’s going to indicate a hotspot, so we utilize the artificial intelligence to recognize a vehicle and actually remove it from the alarming on that particular camera at that particular time,” said Bursell. The company or its clients can train the AI models using internally developed software and a FLIR camera to build the training database. Multiple sensor data is aggregated to support condition awareness.

Fig. 2: Equipment used to detect the progression of an industrial fire from hot spot to flames. Source: MoviTHERM

Fig. 2: Equipment used to detect the progression of an industrial fire from hot spot to flames. Source: MoviTHERM

AmpliCam, meanwhile, uses readily available surveillance cameras mounted on towers to detect smoke, via video content analysis in the cameras. The system has fire location algorithms that pinpoint the fire’s location. The data analyzed on the edge (in video cameras) and sent to cloud software to create maps and routes for fire fighters. A purpose-built system from Vigilys (formerly called Ambient Control Systems) has an IR camera that monitors 360° looking for a distinct signature made by heat-vaporized fuel. The camera uses a narrow band spectral filter to detect the signature, crunching the data on board the camera. Algorithms can refine the detection to reduce false alarms. The system requires a camera placed every square mile to generate an alarm if a fire starts.

Gas and smoke detectors are available commercially for the wildfire use cases. Some of these systems were inspired by the U.S. Environmental Protection Agency’s (EPA’s) “Wildfire Sensors Challenge,” which encourages low cost, low power, portable, easy-to-use sensor designs. Thingy AQ won the 2021 challenge with a sensor system that detects wildfire smoke in real time and monitors air quality, collecting telemetry that can be sent wirelessly. Onboard are sensors detecting fine particulate matter PM 1.0/2.5/4.0/10 and gases, such as carbon monoxide, carbon dioxide, ozone, NO2, NOx, SO2, H2S, and TVOC. It takes relative humidity and air temp measurements. The system stores data on a microSD card but also transmits data in almost real time.

Fire maps, models, tools
The U.S. has research institutions and government agencies working together on fire detection, prediction, and plans for reducing fuel with prescribed burns. University of California San Diego (UCSD) has been working on the problems of both wild fire detection and prediction/risk mitigation maps. With its WIREFIRE Lab, which has funding from the National Science Foundation (NSF), UCSD collaborated with the Los Angeles Fire Department and Orange County Fire Authority to created Firemap, a web-based map that tracks current fires in real time from satellites and shows historical fire data. The maps shows where the Alert Wildfire PTZ cameras and the HPWREN cameras are, what area their view covers, and a live picture from the cameras.

This is just one of many applications of technology to prevent fires, or to spot them as early as possible when fewer resources are needed to bring them under control. Another is QUIC-Fire, a fire-atmosphere model developed and maintained by Los Alamos National Laboratory. “This QUIC-Fire model is the first fast-running coupled simulation fire behavior model that captures the influences of the 3D fuel structure at meter scales, and the coupled interaction between fire, dynamic atmosphere, vegetation structure, and topography, integrating data of varying scales and incorporating the internal processes of the fire dynamics,” according to WIREFIRE, which tracks next generation models and fire research.

The Insurance Institute for Business & Home Safety (IBHS) has a fire test chamber that is being used to study fire behavior near structures. IBHS, a non-profit association with funding from the insurance industry, simulates fire and weather (wind, hail, rain) events to test how they affect a house structure in its test chamber. The chamber is a huge building equipped with 102 fans that can approximate wind gusts from 12 to 120 mph.

A complex problem to simulate is the behavior of embers carried on the wind. Any method that tries to predict where a fire is moving would benefit from understanding where embers are traveling ahead of the fire. Embers make wild fires hard to predict. They can be blown miles from an active fire and the size of an ember affects how it behaves. Unlike particles in smoke that follow the wind pattern, embers are larger and do not always follow the pattern of the wind. Simulating the behavior in a lab requires simulating wind of different types and from different directions, and fire.

“We have a distribution of embers that are closer to real life,“ said Faraz Hedayati, a research engineer with IBHS, in a video seminar. IBHS adapted some ember generators from NIST that are designed to understand which embers move with the wind, which go up high in the atmosphere, and which fall short. Using machine vision, IBHS can see the vortices that are affecting the behavior of embers near a building, so they can create some models.

NIST, along with U.S. Forest Service fire labs such as the Missoula Fire Sciences Laboratory, and other organizations are researching how to predict fire behavior in different wind conditions.

Prescribed and illegal burns
The U.S. announced a 10-year implementation plan to treat more than 20 million acres using fire to fight fire in January of this year. The goal is to prevent large, out-of-control wildfires by reducing the fuel. The amount of dry fuel — built up by the shifting climate exacerbated by prolonged drought in the Western U.S. and 100 years of no-burn policy — will be controlled by prescribed burning, also known as Rx burning. A tool that native cultures have used probably for millennia, Rx burning is the setting of flow intensity fires in controllable conditions, such as on a wet day with no wind. The 10-year plan establishes 250,000-acre firesheds in the West at first, but all sections of the U.S. will get attention under the 10-year plan. Eventually 20 million to 50 million acres will be managed with prescribed fires and innovations.

Athens Prescribed Fire Lab uses lidar and the computationally efficient QUIC-Fire model to suggest Rx burns. “There are several realms of fire in the U.S. and around the world —wildfires, agricultural burning, and prescribed fire,” said Joe O’Brien, from the Athens Prescribed Fire Technology Hub (based in the U.S. state of Georgia) in a webinar. He said the southern U.S. has a burn culture in which it is part of the culture to reduce fuel through prescribed fires. “A burn boss, when he or she decides to light a fire, is making a very profound decision. He or she is apply fire to a landscape that is not on fire, and he or she is responsible for the consequences. There is not a lot of scientific support for making that decision, so we want to rectify that.”

The lab will change how it inventories fuels, and will use 3D landscape models to understand the wind fields. The goal will be to have fast and accurate scenario testing.

Detecting illegal activity is always a concern. Infineon’s Silicon Valley Innovation Center has been working with the San Francisco-based non-profit Rainforest Connection (RFCx) to help detect illegal burning or logging of the rainforest, or poaching of forest animals. The RFCx makes open acoustic monitoring systems that can detect the sound of chainsaws over the natural forest cacophony. Using old smart phones in protective boxes with solar panels on trees, the tree-mounted phones listen for the sounds and send alerts via cellular networks to cell phones on the ground when an event is detected. The systems also can be used for measuring and monitoring bio-diversity, according to the organization’s website.

RFCx also has been evaluating Infineon’s next-generation multi-gas sensor, which detects CO2, as a way of tracking various gases related to forest health monitoring and wildfire detection. “These modules do not have to be on top of a tree,” said Adrian Mikolajczak, vice president of Infineon’s Silicon Valley Innovation Center and Applied System Research, Power, Sensors, RF. The installations will start in Thailand, followed by Brazil.

Fig. 4: A prototype gas-detection module Infineon and Rainforest Connection are working on for wildfire detection in a rainforest. Source: Infineon

Fig. 5: The solar array attached to box that may hold a cell phone with audio or gas sensors. The system is mounted to a tree. Source: Rainforest Connection

Fig. 5: The solar array attached to box that may hold a cell phone with audio or gas sensors. The system is mounted to a tree. Source: Rainforest Connection

Powering a remote system
No one wants to change the batteries on 100,000 sensors in remote locations. Low power use and free energy sources are key design parameters.

Solar is a good option for remote systems. “We had one installation where we did a solar. We had to go remote. There was no way to power the cameras, or the modems or the gateway locally, so we used a solar-battery–based system to run things. It was a very remote installation,” said MoviTHERM’s Bursell. “The power draw on these devices is fairly low.” Most MoviTHERM cameras use Power over Ethernet (PoE), where the power is delivered over a single Ethernet cable to the camera, says Bursell.

Other camera systems also use solar. Vigilys’ cameras have a built-in solar panel, but no battery. Energy is stored in a “non-battery-based super capacitors for 20 years life,” according to the company’s product page, and the camera reports its own health 24/7. Again, the Rainforest Connection system captured audio data powered by a solar panel array and a cellular network — which had strong coverage and was the only option in that particular rainforest — to send audio data. The Thingy AQ sensor box uses solar or battery only configurations, to feed its less than 1-watt power habit.

Even energy harvesting is being used. A prototype detector with temperature and carbon monoxide sensors from Michigan State University uses a triboelectric generator to harvest energy from breezes as it hangs in tree branches. Two differently-sized and weighted cylinders, one coated with copper, the other Teflon, interlock, one over the other and are connected with a rubber band. The two cylinders create energy as they rub against each other, which a triboelectric generator.

The longevity of the system and the power use has to do with how much data is being sent, which eats battery power. A system that just sends a few bits of data a day versus a video or audio system that sends more data, means a different type of network has to be used. Or networks have to be combined.

Using a low-power, long-range (LoRA) network keeps the power use down. “Dryad extensively uses LoRaWAN in our Silvanet system, which provides a standards-based end-to-end solution for ultra-early detection of wildfires,” said Carsten Brinkschulte, co-founder & CEO of Dryad. “Silvanet uses solar-powered gas sensors detecting wildfires with on-board AI, detecting fires even during the smoldering stage and providing a crucial time-advantage to firefighters. As our target environment is the forest, traditional IoT connectivity solutions (in particular NB-IOT) were not a viable option as mobile network coverage in the depth of the forest is virtually non-existent because water-filled trees and plants are blocking radio wave propagation. We overcame this challenge by implementing LoRaWAN, extending the reach of the network into the depth of the forest.”

Networks
Networks used for fire detection systems include everything from long-range, low-power implementations suitable for small, less frequent payloads, to high-bandwidth wireless backbones that are sending video and audio live streams. One network does not fit all use cases, but networks can be combined in a use case.

The LoRa Alliance’s, for instance, says it’s LoRaWAN plays well with other network types. A fire detection system may have two types of networks, one for video, another for very sensors that only send one bit a day. “Sometimes the use case needs videos, sometimes it needs low power. It is about market need, and no technology is going to hit them all,” said Donna Moore, CEO and chairwoman of the LoRa Alliance. “We are a pillar. There’s Wi Fi, there’s cellular, and there’s LoRaWAN.”

The Thingy AQ also uses LoRa to transmit after the data is compressed, but it can now also transmit over LTE and satellite to any cloud application.

Companies making detection systems are being flexible also. Many offer LoRA, Wi-Fi, and cellular access for their systems. It depends what system works best for the location. Indoor systems may “We see Bluetooth,” said Renesas’ Simpson.

UCSD’s High Performance Wireless Research and Education Network (HPWREN), which started in 2000 with NSF funding, is connecting hundreds of video cameras used for wildfire detection and other sensors broadcasts in near real time on the Internet. This high-bandwidth wireless backbone and access data network connects Southern California counties. UCSD, the San Diego Supercomputer Center, and the Scripps Institution of Oceanography’s Institute of Geophysics and Planetary Physics team collaborate on HPWREN, which supports the data-heavy applications, such as real-time video, used in Internet-data applications in the research, education, and public safety realms.

Increasingly high-bandwidth research networks are connecting to each other. In the U.S., this includes research networks such as CENIC in California, which connects to other research networks around the world. These high-bandwidth networks could be used for data-heavy systems with-high bandwidth demands, such as video.

Fig 6: The Western Regional Network that connects research networks. Source: CENIC

Fig 6: The Western Regional Network that connects research networks. Source: CENIC

Fig 7: The Atlantic Pacific Research and Education Exchange (AP-REX) that connects research networks around the world. Source: CENIC

Fig 7: The Atlantic Pacific Research and Education Exchange (AP-REX) that connects research networks around the world. Source: CENIC

Larger areas can also be monitored by satellites, but lower power is still better. “In the outdoor cases, like forest fires and other wide area applications, our long range enables use cases that were not possible before. To have fire sensors in the forest, you would have needed to have a GSM network for the majority of the worst cases, and you would be stuck with a satellite-based network. With LoRaWAN’s long range, we can mount the LoRa gateways on a tower top in the range of 20 or 30 miles. You are able to reach a pretty large area of the forest. Then for the truly, truly remote areas where you couldn’t even face gateways at the high points, we have the option to use low Earth orbiting satellites above the ground 500 to 600 kilometers,” said Alper Yegin, vice president of Advanced Technology Development at Actility and vice chair of the board and technical committee chair at LoRa Alliance. “They would serve the LoRaWAN base as on the ground. Sensors, unlike the other satellite-based technologies, would still have a very low transmission power — as low as 25 milliwatts.”

Conclusion
The technology and systems for fire detection continue to improve. Industry, government, and researchers are learning from each other and connecting networks to the cloud for data analysis in easy-to-use dashboards and online monitoring systems that we are all starting to rely on. The software helps customers train and use AI and machine learning in their detection set ups. Increasing these systems, especially the publicly funded research-based systems, are connecting to each other to offer large scale insights accessible over the Internet.

But when it comes to reducing fuel over a large area, the prescribed burning is still the tool of choice. Modern technology just helps us identify where and when to apply that ancient method.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button