Portfolio
Dženis Avdić
Summary for ETH Zurich
1 :: DATA MINING :: WEB SCRAPING
- Created a script that runs locally to read and store available weather data in CSV format
- Data obtained from Federal Hydrometeorological Institute for Sarajevo, Bosnia and Herzegovina
- Windrose plot is being updated with every new value and stored as PNG image
- CSV data is used for analysis of daytime winds in Sarajevo for specific periods
Sarajevo windrose based on dataset collected from september 2020 to september 2021
2 :: DATA MINING :: SENSOR LOGGER AND WEB API DATA
- Data logger is made with open electronics components to collect data inside observed architectural spaces
- 15-minute logs are being stored to microSD card in CSV format for later analyses
- Available meteorological and air quality data from web APIs is being stored simultaneously
- Long term monitoring and data collecting for scientific research in building physics and energy efficiency
3 :: DATA PREPROCESSING AND OBSERVATIONS
- Based on collected data specific analyses were conducted
- Missing values and sensor misreadings were recognised and filled in with neighbouring values average
- Physical phenomena were analysed for wind movement, natural ventilation and infiltration
Infiltration air change rate based on collected CO2 concentration level data - Python plot
4 :: DATA VERIFICATION AND VALIDATION USING CFD SIMULATIONS
- 2D and 3D CFD simulations were conducted for typologicaly specific building form
Building block with courtyard typology
Street corridor ventilation and apartments’ air change rate estimation
- Based on real data (weather stations data), natural ventilation potential was estimated for neighbourhood (urban area ventilation) and single apartment (interior ventilation air flow)
‘Marijin dvor’ building air change rate simulations
- Pedestrian and street level air movement phenomena was analysed for arguably misplaced high rise building in Sarajevo area
Air movement around building simulation for Sarajevo Tower
5 :: MACHINE LEARNING :: TIME SERIES FORECASTING
- Tensorflow Recurrent Neural Network (RNN) with Gated Recurrent Units (GRUs) for time series forecasting was deployed to Arduino Nano 33 BLE Sense using TinyML
TinyML time series forecasting deployed on Arduino
- Three different prediction algorithms were discussed and compared (ARIMA, facebook Prophet and Tensorflow RNN)
Accuracy of compared ML models for air moisture content
- Low-energy consumption AI assisted system is proposed for natural ventilation control and predictive maintenance (full paper)
6 :: LOW-POLY 3D MODELING :: OPTIMIZATION FOR UNITY AND UNREAL ENGINE
- Worflow includes various software for 3D modeling, retoplogy and texturing
- Optimization of highly detailed 3D models for web and UI/UX integration
Game 3D models optimized with low polygon count for mobile gaming implementation
Preview of low-poly 3D model, ready for implementation in any 3D game development engine
Animated preview