Constructing a visual detection model for floc settling velocity using machine learning.

Area of Science:

• Water treatment
• Chemical engineering
• Machine learning

Background:

• Optimizing coagulant dosage is crucial but time-consuming.
• Real-time evaluation of floc settling velocity can predict coagulation effects.
• Accurate settling velocity assessment is key to efficient water treatment.

Purpose of the Study:

• To evaluate the accuracy of convolutional neural network (CNN) models in recognizing floc settling velocity from images.
• To develop a machine learning approach for real-time coagulation process monitoring and optimization.
• To compare the performance of different CNN architectures for floc settling velocity analysis.

Main Methods:

• Utilized Python and OpenCV for image segmentation and floc settling velocity detection.
• Constructed a dataset of floc images correlated with their settling velocities.
• Applied and compared Lenet5 and Resnet18 CNN models for image recognition.

Main Results:

• Floc settling velocity determination based solely on particle size yielded 88% accuracy.
• Lenet5 CNN model achieved 88% accuracy in recognizing settling velocity.
• Resnet18 CNN model surpassed 90% accuracy in recognizing floc settling velocity.
• CNN analysis demonstrated that floc structure complexity requires more than single parameters for accurate velocity prediction.

Conclusions:

• Machine learning, specifically CNNs, effectively evaluates floc settling velocity from images.
• CNNs offer a faster and more accurate alternative to traditional methods for assessing coagulation.
• This technique provides theoretical guidance for optimizing coagulant dosage and water treatment processes.