Though several disaster warning and accident prevention mechanisms such as the tsunami warning system have been deployed in high-risk areas, the perfect system that would sense imminent danger even in the so-called normal situations has remained a scientific Holy Grail.

The research conducted by US-based Indian scientists S S Iyengar and Supratik Mukhopadhyay of Louisiana State University may well have the answer.

They have come up with a sophisticated and intelligent disaster and accident management system that can be applied in a wide range of disaster and security situations.

Dubbed Cognitive Information Management Shell (CIM Shell), it is said to intuitively understand the series of events that culminates in a disaster or in an accident.

“Each of those disasters, such as the earthquake or tsunami and accidents such as oil spills or explosions happen as a result of a set of events,” explains S S Iyengar, who is Professor and Chairman at the Department of Computer Science, LSU. “Today, we understand that the fundamental thing in Computer Science is not information but events,” Mukhopadhyay adds.

“Such events are very complex and happen for sometime before the disaster or accident. Though these are recorded, it would be highly impossible for human beings to study the events and the pace in which they take place,” Iyengar points out.

The CIM Shell, which has been developed in the last three years with wider support and funding for Iyengar and Mukhopadhyay from US defence companies and the Naval Research Laboratory, performs three basic tasks: recording the events that lead to accidents or disasters, comparing them with the database of similar events in the past to generate a common operational picture and acting on the information it has to prevent or predict the disaster.

Distributed databases

The network-based system is made of multiple sensors and distributed databases to continually detect and store events that can be cross-checked with the database of past events. The pre-defined rules allow dangerous trends and patterns to be detected in real-time and action is taken for prevention or prediction of the disaster.

“In order to achieve what we call the ‘goal’ (successful prediction or prevention), the AI or Artificial Intelligence autonomously and continuously adjusts parameters in system rules to bolster the accuracy of the result,” Iyengar says. Mukhopadhyay adds that unlike some current systems, the one developed at LSU is not perturbed by uncertainty; the software can adapt to the situation. Though the system has an effective AI technology, human operators are not entirely excluded from it. Operators are able to add or reconfigure the system according to the new critical information available to them.
Because such changes will have to reach thousands of sensors spread across the enterprise system, CIM Shell provides a single customised admin tool.

If human operators of the system come to know of any new information that would enhance the result, they can inject that into the system. “If you want to understand how effectively it works, you need to first get a clearer idea of the ‘goal’,” Iyengar said.

“The ‘goal’ is the desired outcome. The ‘goal’ could be virtually anything from reducing wasted electricity in a factory or preventing fraudulent transactions on a credit card account. These goals ask for specific actions such as switching off machines after a certain period of inactivity or automatically declining an unusual transaction. So the success of a goal depends on what action to take and when.”

This is where cross-referencing of current events with the history of past events plays a crucial role. Iyengar gives the example of how it is possible to use cross-referencing to identify an abnormal credit card transaction and abort it. All it takes, he said, is a simple rule:  “Decline the transaction if the amount is at least 50 per cent greater than the average for the past 50 transactions and the location of the transaction is at least 200 miles away from the location of the previous five transactions.”

“In this example, it is the deviation from the trend (the potential for fraud) that is of interest. In other scenarios, it may be the continuation (rather than the deviation) of a trend or pattern that is of interest. The point is that only when analysing multiple events over time is detection of a trend even possible. Furthermore, it is only upon detection of a trend that it becomes possible to predict an event and take preventative or opportunistic action,” he explains.

The system uses human-readable declarative language for defining rules and goals, which enables sophisticated user-interfaces to be developed as a layer on top of the declarative configuration language. Given the Indian connections of the researchers, they hope their invention will help their homeland.

“We have seen the human disaster that followed the December 2004 tsunami that hit India and other countries of the region,” Dr Iyengar, who hails from Karnataka, said.
“Also, the current BP oil spill disaster affecting the coastline of the US is spiralling out into a major ecological catastrophe for the world’s largest economy. We would like our invention to play a major role in stopping such tragedies from affecting humanity,” he adds.