Deep Learning Applications in Wastewater Treatment
Mentor 1
Rudi Strickler
Start Date
1-5-2020 12:00 AM
Description
Wastewater treatment currently requires careful testing at multiple points in the treatment process. These tests are often slow and expensive, requiring trained personnel and lab time. Here we report an application of neural networks to supplement these tests in real time. As wastewater flows across a gravity belt, water is lost and a polymer is added to the sludge, thickening it. In order for the next steps in the treatment process to go smoothly, this sludge must be appropriately dry, or thickened. If the sludge is dry enough, it is of interest to the company to reduce the amount of polymer used in order to cut costs. Our neural network aims to classify images of the polymer into “wet” and “dry” categories, in real time on a raspberry pi. To do this, we use open source technology like tensorflow and python. As a laser is shined into the sludge, it creates a pattern which the neural network uses to classify the images. Preliminary programs are successfully classifying sample data, and it is expected that further refinement of our models will yield faster and even more accurate results. By creating a more responsive process for testing the character of the sludge, we hope to set the stage for better water treatment systems.
Deep Learning Applications in Wastewater Treatment
Wastewater treatment currently requires careful testing at multiple points in the treatment process. These tests are often slow and expensive, requiring trained personnel and lab time. Here we report an application of neural networks to supplement these tests in real time. As wastewater flows across a gravity belt, water is lost and a polymer is added to the sludge, thickening it. In order for the next steps in the treatment process to go smoothly, this sludge must be appropriately dry, or thickened. If the sludge is dry enough, it is of interest to the company to reduce the amount of polymer used in order to cut costs. Our neural network aims to classify images of the polymer into “wet” and “dry” categories, in real time on a raspberry pi. To do this, we use open source technology like tensorflow and python. As a laser is shined into the sludge, it creates a pattern which the neural network uses to classify the images. Preliminary programs are successfully classifying sample data, and it is expected that further refinement of our models will yield faster and even more accurate results. By creating a more responsive process for testing the character of the sludge, we hope to set the stage for better water treatment systems.
