How to predict a smart audio device's battery life using data analytics

DEWI is a European research project which ended in February 2017 and has provided key solutions for seamless wireless connectivity and interoperability in smart cities and infrastructure by looking at members of the public's everyday physical environments in buildings, cars, trains, and planes. In this way, DEWI has provided significant input for the emergence of smart homes and smart public spaces.

Within this framework, Sirris worked with Gibson Innovations and imec on a use case which aimed to reduce maintenance costs and make it easier for customers to use a Wireless Sensor Network (WSN). Gibson Innovations is a company that supplies music and audio products to consumer and professionals (headsets, audio, home cinema and video and so on). In this light, the activities of the Flemish consortium focused on smart home entertainment.

In the context of this project, Gibson Innovations' interest was in exploring and developing smart audio systems that interact with various Wireless Sensor Network devices (e.g. motion sensors, door sensors) in the home, so that for example when users arrive home, their music starts playing and then follows them around the house, while the lighting is adjusted based on the situation.

Against this backdrop, the collaboration between Sirris and Gibson Innovations more specifically focused on predicting the battery life of WSN devices and smart audio devices based on their use. Gibson Innovations' ability to predict how long a battery of a sensor module or a smart audio device will last has the potential to help to enhance customer satisfaction. For example, if a customer were to use the speakers when watching a film, the ability to predict how long a speaker would last, would mean the user could know how long he or she could watch the film for without having to charge the speakers.

Smart Home Entertainment Demonstrator

At Gibson Innovations' premises a permanent demo setup representing a small residential building was put in place, integrating a variety of heterogeneous wireless sensors (motion sensors, door sensors, a doorbell sensor and so on), actuators and wireless audio speakers.

Gibson Innovation Smart Home Entertainment demo setup

Predicting battery life

To collect data that can be used to predict battery life, Gibson Innovations fitted two battery-driven wireless surround speakers in this demo setup with wireless voltage sensors so that the battery's level of charge could be monitored. This setup allowed data on the battery level and playback time to be collected. Some of these collected data were used to train the developed prediction models, while the rest of the data were used at a later stage to check the models.

Three prediction models were developed (as set out in the figure below) based on three approaches: linear regression, Bayesian networks, and polynomial curve fitting.

For each of these approaches, the relevant requirements in terms of available data, model design, and deployment were studied, as well as the trade-off between the model's complexity and its accuracy. The results of this analysis are listed in the table below.

Finally, Sirris tested the prediction models developed for the data that had been collected at the start, and based on the results and the needs analysis, a decision was made to implement the prediction model based on the polynomial curve fitting in the Smart Home Entertainment demonstrator. This approach can also be used to predict the battery life of other WSN or smart audio devices.

Would you like to have further information on this approach and/or project? Contact us by e-mail!

Want to read more about this use case? If so, an article on this subject was also published in's Tech Magazine (in Dutch).

You can read this success story and many others in the Sirris Annual Report 2016.