Device fingerprinting

I came across a post on Engadget today [1]. It mentions Intel experimenting with the detection of specific household appliances being switched on/off through the use of a single sensor plugged into a power socket. I had read about a taxonomy of load signatures being proposed, based on device-specific transient voltage-current trajectories in [2]. Being able to single out individual appliances and switches from a single sensor sounds like the holy grail of domestic energy monitoring. Flukso data streams could be annotated with specific events (dishwasher turned on, hot water circulation pump running, …) by the use of just a single additional sensor. No need to start installing a plethora of wireless sensor nodes all around the house.

Some further Googling led to the excellent paper in [3]. It discusses similar research conducted at Georgia Tech. The researches detect the transient voltage pulses being generated on the power lines in the 100Hz-100kHz spectrum by a device being switched on/off. FFT transformation allows these pulses to be characterized by the system as vectors containing their spectral information. A Support Vector Machine technique is used for classification of newly detected pulses. The SVM does require initial training in order to successfully classify future pulses. The paper mentions a 90% success rate. Similar research from the Ubicomp Lab at the University of Washington is discussed in [4] and [5]. This time microphones or a single pressure sensor are attached to a home's water piping to detect the household's water usage patterns.

A very interesting read!

[1] Intel's experimental sensor analyzes appliance power consumption from single outlet
[2] A Taxonomy of Load Signatures for Single-Phase Electric Appliances
[3] At the Flick of a Switch: Detecting and Classifying Unique Electrical Events on the Residential Power Line
[4] Sensing from the Basement: A Feasibility Study of Unobtrusive and Low-Cost Home Activity Recognition
[5] HydroSense: Infrastructure-Mediated Single-Point Sensing of Whole-Home Water Activity

Update 18/04: Added a link to the Ubicomp Lab's project page.