Report

Teaser, summary, work performed and final results

Periodic Reporting for period 2 - RUBSEE (Extending artificial intelligence revolution in the waste field beyond sorting)

Teaser

RUBSEE is a real-time monitoring system based on Artificial Intelligence and Deep Learning techniques. This system is able to determine the composition of material in Waste Treatment Plants in real-time. This monitoring platform allows the facilities to have real-time data...

Summary

RUBSEE is a real-time monitoring system based on Artificial Intelligence and Deep Learning techniques. This system is able to determine the composition of material in Waste Treatment Plants in real-time. This monitoring platform allows the facilities to have real-time data visualization, create alarms to prevent problems or unexpected downtimes, realize reports and save historic data to optimize the operation information of the plant. It allows reducing the amount of recyclable material that is currently being landfilled or incinerated.

Work performed

From the beginning to the end of the project, the following work and main results have been achieved:
• A dataset with more than 1.000.000 objects has been created, these objects are separated into 10 categories (due to customer requirements). Around 40.000 frames are labeled and more than 1.000.000 of annotations (polygons) have been done and classified. Recordings of three facilities are taking into account in this database.
• The system has been installed in three pilot facilities:
Light Packaging Material Recovery Facility in Hostalets de Pierola, Barcelona, Spain. Designed to process 7 tons per hour of light packaging waste.
Mechanical Biological Treatment Plant in Mancomunidad La Vega, Seville, Spain. Designed to process 40 tons per hour of municipal solid waste.
Light Packaging Material Recovery Facility in Baguim do Monte, Oporto, Portugal. Designed to process between 3 and 5 tons per hour of light packaging waste.
• The recordings have been done the three pilot plants involved in this project and are stored in Sadako’s servers.
• We have created a Neural Network able to detect different kinds of material like HDPE, Tetrapack, PET, PP, metal, so on (the initial Neural Network was able only to detect PET bottles) with the most advanced techniques of deep learning.
• We have developed different software modules needed to use the RUBSEE system as a single unit:
Acquisition module
Detection module
Communication module
• We have developed the RUBSEE monitoring platform that is already implemented in the three facilities.
https://hostalets.rubsee.com
https://lavega.rubsee.com
https://porto.rubsee.com
This platform allows:
Real-time data visualization
Create alarms to prevent problems
Create and download report to optimize the operation of the plant.
Save historic data
• A final design of the monitoring point has been done, as cheap and modular as possible.
• 26 monitoring points have been manufactured and assembled in Sadako’s premises in L’Hospitalet de Llobregat.
• Operational test has been done in Hostalets de Pierola to adjust the transformation between the information from individual images and the real weight of the waste seen by the camera.
• Training of the monitoring platform and maintenance of the equipment have been given to personnel of the three facilities.
• A business model of license has been agreed between Sadako Technologies and Bulk Handling Systems. With this kind of business model, Sadako Technologies can be focused on the development of Deep Learning technology and applications, its expertise, while letting BHS focus on all the rest of the required tasks to bring that technology to the market. With this
business model, BHS is responsible, therefore, for tasks like hardware components design and manufacturing, marketing, sales, customer service and so on.
• Sadako has attended to different fairs and events to promote the project to wider audience like Public Administration, Recyling Companies, Big Industrial Service Companies, Waste Treatment Plants, Technical Collaborators.
• New marketing material has been done to promote the RUBSEE project: promotional video, new website, articles in important magazines of different sectors (artificial intelligence, waste, robotic, so on).

Final results

Sadako’s technology is able to detect and measure a big diversity of objects existing in crowded or singulated waste streams, for mechanical biological waste treatment plants or light packaging material recovery facilities. This is a big challenge in terms of visual recognition.
Our disruptive low-cost technology let to increase the recovery rates, so decrease the quantity of material sent to landfills or incinerated. RUBSEE will be a key tool for waste treatment and recycling plants, promoting the circular economy.

Website & more info

More info: http://www.sadako.es/rubsee-project/.