Based on the datasets recorded from LMD process, this project provides the basic sctructure to develop a convolutional neural network software for process parametrization and defects detection.

Example of power prediction comparing with width measure in a wall construction.

The data adquisition of big data required for the training is explained above.

Big Data Registering

The high bandwidth of the multimodal monitoring approach demands effective mechanisms to record and manage such a big data. Thus data are recorded in bag files (parameters in yaml files), and they are processed after that to provide a better way for analysis.

Cyplam main interface for recording and direct cloud connection.

Bag files are a good way to reproduce the system performance enabling the development of new modules and features as if it was the real facility. However, it is not a good solution to search and analyse data.

Exploring multiespectral recorded bag data with the rqt_bag tool.

To provide a better way for data analytics, bag files are parsed and stored in HDF5 files as Pandas dataframes. A dataframe is created for each rostopic namespace (e.g. /tachyon, /camera, /robot) using the message timestamp to correlate data.

Pandas dataframe used to manage data opened in a notebook.

Getting Started

You only must install ROS and clone OpenLMD projects on your computer, and download any bag file available on ZENODO to test the software.

$ roscd cyplam_data/src/data/
$ python analysis.py -f filename.bag

Contact

Adrián Pallas-Fernández (adrian.pallas@aimen.es)

Verónica Panadeiro-Castro (veronica.panadeiro@aimen.es)

Baltasar Lodeiro-Señarís (baltasar.lodeiro@aimen.es)

Jorge Rodríguez-Araújo (jorge.rodriguez@aimen.es)

---

Copyright 2016-2017 - AIMEN Technology Center