27/07/2019

Scientists use data to find bodies in Brumadinho.

Bombeiros buscam vítimas da tragédia em Brumadinho (MG) / Imagem: Divulgação

Reproduction from UOL

By Stefhanie Piovezan – Finalist in the Science Communication category of the 2019 IMPA-SBM Journalism Award

Researchers from various Brazilian universities and different specialties, such as data scientists, engineers, mathematicians, and physicists, decided to come together to find ways to help locate victims in Brumadinho (MG) and prevent further dam collapses. This is how the group "Science for Brumadinho" was born.

According to the latest survey by the Civil Defense of Minas Gerais, 182 people died in the tragedy at the Córrego do Feijão mine and 126 are considered missing.

"I can't say what motivates all the members of the network, but, in my case, it's outrage. It's not the first mistake, it's the second, and it followed one after another. It seems that the responsible bodies weren't concerned about preventing these tragedies from happening, and I understand that we can no longer wait for someone else to solve it. We have to act and help prevent this from happening," said Vânia de Oliveira Neves, coordinator of the initiative and researcher at UFJF (Federal University of Juiz de Fora).

Neves is a software engineering specialist and wants to use his expertise in this technology to help the Fire Department's work, but it's not easy.

Tired of waiting for the necessary information to develop support programs and applications, which never arrived, she began actively seeking direct contact with the rescue teams. However, she was told by the agency that the priority was locating the victims' bodies, so she decided to create a parallel network to chase after this data.

With the help of a volunteer working at the Ministry of Science, Technology, Innovation and Communications, she managed to contact researchers interested in helping the cause, especially computer scientists. Eighty scientists from universities such as USP, Unicamp, PUC-RS, UFAL, UFAM and UFF joined her, along with about 200 volunteers in the technology field – some already actively participating, others following the group awaiting their area of expertise to be requested.

There are several areas of work:

Studying fluid mechanics to understand how the mud dispersed and settled.
Using an algorithm to find the drag force and, based on it, indicate possible positions.
Using machine learning to simulate where the bodies are located.
Integration of models
Validation of missing persons

Analysis of the mud path

The first results that emerged in the researchers' WhatsApp group, which also includes representatives from the Public Prosecutor's Office and the Fire Department, came from Poli (Polytechnic School of the University of São Paulo).

Members of the Automation and Control Laboratory at the university used an algorithm to find the drag force in the tailings, determine the acceleration of the mud, and from there, estimate the distance traveled by the victims.

With this information, they were able to identify 30 possible locations for those who were in the cafeteria at the time of the dam collapse.

"This method can help firefighters because it indicates points that were not anticipated," says Rafael Fernandes Pinheiro, a doctoral student at Poli-USP.

He and his colleagues are now trying to refine the algorithm and, with the help of other researchers, indicate not only latitude and longitude, but also the depth of the bodies.

Despite the efforts, Neves is cautious about the impact of the initiative. She explains that there is difficulty in finding professionals who understand the dynamics of the mud and, since the work is voluntary, many can only help in their free time.

Furthermore, researchers have little information about the behavior of mudflows and say that it is not possible to guarantee the success of the undertaking. "It's an attempt. We are trying and doing our best to bring some comfort to the families," says the scientist.

Mapa enviado pelo Ministério Público de Minas Gerais mostra a localização das vítimas de Brumadinho / Imagem: Divulgação

Data analysis Another group collaborating with the Science for Brumadinho project, in this case using machine learning, is based at the Institute of Mathematical and Computer Sciences at USP, in São Carlos (SP).

When the request for help arrived in the inbox of computer engineer Luís Paulo Faina Garcia, a postdoctoral researcher at the institute, he contacted his colleagues at the Massive Data Analysis Laboratory and Professor André de Carvalho, a specialist in machine learning and data mining.

"The intention is to try to help the families and the firefighters in some way, even if we are not physically present there," says researcher Luís Paulo Faina Garcia.

To do this, they feed a program with the victims' locations before and after the accident and train this system to estimate the area where those who remain missing might be.

The data is provided by the Public Prosecutor's Office of Minas Gerais and includes files such as maps showing the topography before and after the dam rupture, information on the location of the bodies already found, the precise location of the buildings, and the dimensions of the flood area.

They are analyzed, applied, and begin to serve as the basis for the first results.

Pesquisadores do laboratório do Instituto de Ciências Matemáticas e de Computação da USP fazem parte de grupo de cientistas que estudam formas de localizar vítimas do desastre / Imagem: Reinaldo Mizutani/ICMC-USP

"Working with machine intelligence requires data, but the Brumadinho data took a long time to be released. So, we started by doing simulations with artificial intelligence," Garcia said.

Preventing a third tragedy

Another focus of the group is the prevention of new accidents involving dams. In this case, the proposal is to adjust and apply already known techniques.

"When I saw the first news report about Brumadinho, it caused me great anguish, because the techniques we are researching at universities could be used to prevent this type of tragedy from happening," Neves said.

"It would be possible, for example, to computerize the systems and connect the sensors to a central control room, so we could have systems that would warn of the need to evacuate areas and indicate, via cell phones, which escape routes to take," he exemplifies.

According to the researcher, alert systems linked to Civil Defense or the Public Prosecutor's Office could also be used.

"Mariana shouldn't have happened, nor should Brumadinho. We absolutely do not want a third tragedy to occur," he says.

According to her, Brazil has highly trained professionals who could be helping to solve these problems.

"Much has been said about spending on universities, but they don't just train professionals. There are the pillars of outreach and research, and the research being developed could have been applied to prevention. It wasn't, and look what happened," criticizes Neves.