Thesis Topic: Automatic end-consumer device recognition


Currently, the shift towards an energy consumption awareness is on the rise, thanks to the increase of e-Mobility and also, especially in bigger citys, e-Scooters.

But the consumption of daily household utilities and devices are still unknown to many customers. Many reasons can be found (pre-installed devices, e.g. in the kitchen), devices that have been gifts, etc.

With the help of a measurement device and machine learning, this problem is about to be solved.

In cooperation with a StartUp the Diploma Thesis should tackle one of the following highly interesting research questions:

  1. Recognition of electrical components like motors, heaters (water/air), different power supplies, different lamp types, etc. from their power usage characteristics without prior training.
    1. Research questions:
      1. finding attributes of electricity consumption that make components of electrical appliances identifiable
      2. what are the minimum required attributes that allow high confidence recognition with the focus on broad end customer usage
      3. Attributes: sampling frequency, amplitude resolution, dimensions (W, VAR, CosPhi, Hz, V, …)
  2. Recognition of device types, brands and models from their power usage trace.
    1. Research questions:
      1. High quality prediction of pre-trained appliances in live sampling data – validation with customer feedback and annotations
      2. Expand the state of the art from 5-7 appliances that can be distinguished to a typical size of a household: 50+
  3. Recognition and device isolation with complex programs such as dish washers, washing machines or heat pumps – executed in the live data feed of the power usage recording.
    1. Research questions:
      1. Customer view: „Which program of my dish washer is the most efficient one“
      2. Expand software stack to allow annotations and analysis of different program modes per device
      3. Isolate and recognize program specific power usage of the device
  4. OCR and image recognition to allow the user to take a photo of a device and it’s type plate instead of typing model, make, type into the system manually.
    1. Research question:
      1. State of the Art Research
      2. Create UX design for close to zero effort device identification
      3. Identify and integrate device definition repositories for validation and enrichment

Technical environment for thesis work:

  • Installation of power meter at students home preferable
  • Access to anonymized data of Watt Analytics customers via REST API
  • Development Environment
    • Java 1.8, Spring Framework 2.1
    • Apache Storm for Analytics Pipeline
    • PostgreSQL, time series DB to be selected and implemented
    • JSON over MQTT for data transport
    • React for web and mobile apps
    • Cloud Deployment on Heroku
    • GIT Source Repository
    • Agile Tools: Jira, Bamboo, Confluence

Close cooperation for software development with the StartUp is encouraged and welcomed. Real world data can be provided by the StartUp.