AI based industrial automation: advanced sorting without human intervention

  • Swimlane: 2022-2023
  • Column: Draft
  • Position: 43
  • Assignee: Olivier Debeir
  • Creator: Olivier Debeir
  • Started:
  • Created: 03/05/2019 08:03
  • Modified: 01/06/2020 15:40
  • Moved: 01/06/2020 15:40
Description

Context
Industrial processes can be hazardous or dangerous and often menial. And where human intervention is both risky and/or costly, automation could be indispensable.
Sorting is an industry process which is already strongly automated with applications in e.g. waste processing, food sector, …

Typically vision is used to detect and identify objects. Current state-of-the-use algorithms typically employ little to no artificial intelligence, and the resulting performance is suboptimal: often human operators are required at the end of the line to sort what was not correctly classified.

Objective
The goal of this thesis is to explore the possibilities, relevance and impact of artificial intelligence in sorting applications.
You will build a simple lab hardware setup: camera(s) overlooking a stationary bin of objects which need to be sorted into other bins. Algorithms need to be selected/developed to identify the (result of the) sorting actions of the human operator: which items have been moved from the initial bin to another bin, which items were left in the initial bin.

The focus point of this thesis is to explore the possibilities of artificial intelligence: a model needs to be trained by observing the actions of human operator. Its ultimate goal is to reproduce the sorting logic of the operator. Based on camera images the model needs to detect and classify items like the operator would, so that the sorting could be automated.

The automation of the sorting action itself (grabbing and moving objects) is beyond the scope of this thesis.

Through this thesis, you will highlight the opportunities, challenges and pit-falls in the modeling and the training phases for sorting applications.

Contacts: Olivier Debeir (odebeir@ulb.ac.be), buvana.lefevre@sirris.be.

Sub-Tasks
Internal links
Comments
Olivier Debeir
Olivier Debeir Created at: 20/05/2019 10:02 Updated at: 20/05/2019 10:02