IML DataSoft
IML DataSoft is the platform I built for my master's thesis. It compares data collected from a real industrial environment with simulator-generated data using deep learning algorithms for anomaly detection, and the results showed that the simulator data was close to the real data. The Next.js web app lets you enter real-world data, create simulator data, define anomaly types, train and run models, and compare algorithm results. The system has four parts: a Next.js web app, a Hasura backend, a Node.js API and a Python API.
- Development
- Year
- 2024
- Role
- Full-stack developer · Master's thesis
Technologies
- Next.js
- TypeScript
- Hasura
- GraphQL
- Node.js
- Python
- Flask
- RabbitMQ
- PostgreSQL
- MySQL
- Docker
- Arduino
Overview
Developed for my master's thesis. Real data is produced with an Arduino and compared with data generated by a simulator, then both are tested with deep learning algorithms. The result: the simulator data turned out to be close to the real data.
How it works
- Produce real-world data with an Arduino, enter it and build datasets.
- Create simulators and define the error types they should inject. The simulator generates synthetic sound, vibration and temperature data.
- Test the datasets with deep learning outlier detection algorithms, such as ABOD, Autoencoder and CBLOF.
- Compare the results of different algorithms, and real data against simulator data.
Architecture
Four parts: a Next.js web app, a Hasura backend, a Node.js REST API and a Python REST API. In the Python API, a Flask service reacts to new database records, and a RabbitMQ queue runs the algorithm requests one after another. Everything ships with Docker.