Multimodal machine learning for Cr(Ⅵ) removal and floc settling using image-based floc features and operating...

Area of Science:

• Environmental Engineering
• Water Treatment Technologies
• Artificial Intelligence in Environmental Science

Background:

• Electrocoagulation is effective for hexavalent chromium (Cr(Ⅵ)) removal but struggles with variable wastewater conditions.
• Factors like pH, electrolyte concentration, and stirring rate affect floc formation and settling, limiting process efficiency.

Purpose of the Study:

• To develop a multimodal machine learning framework for predicting Cr(Ⅵ) removal and floc settling during electrocoagulation.
• To enhance the adaptability and accuracy of electrocoagulation processes under diverse operational conditions.

Main Methods:

• Utilized deep learning (ResNet50) to extract image-based floc features.
• Integrated these features with operating parameters into classification-regression and direct regression models.
• Employed machine learning algorithms including Support Vector Machines, Bagging Classifier, and Extra Tree.

Main Results:

• A multimodal approach integrating image features and operating parameters significantly improved prediction accuracy.
• Direct regression models achieved high R² values: 0.971 for Cr(Ⅵ) removal and 0.986 for floc settling.
• The framework demonstrated superior performance compared to traditional methods in predicting electrocoagulation efficiency.

Conclusions:

• The developed multimodal machine learning framework offers a robust solution for optimizing electrocoagulation.
• This approach enhances prediction accuracy and process adaptability for effective wastewater treatment across various conditions.
• Pioneering the use of deep learning-based image analysis with operating parameters advances wastewater treatment optimization.