Klev

18-549 Team 17, Spring 2015

A trainable flexible device sensor for smart homes
Electricity

Description

We aimed to create a “node” device that can be attached using adhesive to common household devices and appliances and communicate with “hub”. The node uses a magnetometer, gyroscope, and accelerometer to periodically collect data about the device to which it is attached, and use that data to tell whether the device is on, off, or in an abnormal state. It periodically sends the results to the hub device using 802.15.4. The hub, a raspberry pi 2, runs a lightweight Linux distribution, is plugged into a wall socket, and can be connected to the Internet with WiFi.
The hub stores the information from the node locally; this information is displayed through a django web-app hosted by the hub. Users can view the web-app and “train” a device by turning the appliance off, and telling the web app that the device is off, then turning it on, and telling the the app when the device is on. Once the training is performed, the user can check the web-app to view current device states, state-history for a given device, and more general device information such as model number or location.

About

Team Klev is based in Pittsburgh, PA. We're one of many teams developing innovative embedded systems on behalf of Carnegie Mellon University's senior Electrical & Computer Engineering capstone course: 18-549: Embedded System Design taught by Dr. Anthony Rowe .

Abstract

Existing state-recognizing sensor systems transmit their raw sensor data to another more powerful system in order to process and classify the data. Klev is an alternate solution which bundles the sensor with a processor powerful enough to classify the data using machine learning techniques (SVM), but power efficient enough to last longer than the competing products which constantly transmit data.

Motivation

As the idea of a "Smart Home" becomes a reality, easy-to-use sensor arrays will allow homeowners to better customize their homes. Currently, sensor arrays require pre-defined rules to effectively monitor device state. Users will be able to train Klev to monitor device states the need for programmatic skills.