Electronic engineers in Singapore have developed and successfully tested a management system that increases the efficiency of wireless sensor networks for monitoring machine health1. The new system, known as an adaptive classification system (ACS), reduces the power consumption of individual sensors and increases their lifespan, while also decreasing network traffic and data storage requirements.
The ACS also achieves more robust results in terms of diagnosis of machine problems and prognosis of performance. “Other applications include monitoring patient health, disaster monitoring systems, such as fire alarms, and environmental monitoring for chemical plant accidents, air and water quality,” says Minh Nhut Nguyen of the A*STAR Institute for Infocomm Research, who led the research team.
Wireless sensors are now so inexpensive and flexible that their application in monitoring systems is widespread. Because of the environments in which they are deployed, sensors increasingly require their own portable power source, typically a battery, which means they have a limited lifespan (see image). Any way of reducing the amount of power the sensors draw would increase their lifespan, decrease the need to replace them and therefore reduce costs, Nguyen explains.
Reducing sensor sampling rates to a practical minimum is one way to lower power consumption; this can be achieved by halting monitoring when a machine is not operating. Typically, a machine functioning smoothly demands a lower and coarser sampling rate than one that needs attention. Nguyen and his co-workers therefore developed their ACS along these lines. Importantly, it incorporates an adaptive system of nested sensors. Some of the ACS sensors sample particular parameters at a low rate to provide data for a model whose purpose is simply to trigger more intensive sampling of other sensors when a potential problem is detected.
In addition, the system utilizes a set of models that is geared to sensors sampling at a particular rate. The ACS also integrates several different methods of classifying whether particular data patterns are of concern such that they require higher levels of sampling. Decisions are therefore made on the basis of multiple classifications. This not only increases the robustness of the system, but also means that it can be trained to detect problems using a minimal amount of data.
Nguyen and his team tested the ACS using a machinery fault simulator, a machine in which key components, such as bearings, could be replaced by faulty or worn ones. Encouragingly, on average the ACS outperformed current models in these tests.
The A*STAR-affiliated researchers contributing to this research are from the Institute for Infocomm Research