Chemical signatures to identify the origin of solid ashes for efficient recycling using machine learning. (25th September 2022)

• Record Type: Journal Article
• Authors: Qi, Chongchong; Wu, Mengting; Xu, Xinhang; Chen, Qiusong
• Is Part Of: Journal of cleaner production. Volume 368(2022)
• Publication Date: 2022-09-25
• Subjects: Solid ash -- Origin identification -- Machine learning -- Random forest model -- Classification; Factory and trade waste -- Management -- Periodicals; Manufactures -- Environmental aspects -- Periodicals; Déchets industriels -- Gestion -- Périodiques; Usines -- Aspect de l'environnement -- Périodiques; 628.5
• DOI: 10.1016/j.jclepro.2022.133020
• Languages: English
• ISSNs: 0959-6526

Abstract

Abstract: Solid ashes constitute the world's leading environmental problem and their efficient recycling requires specific origin information. Chemical signatures encode the physicochemical and structural characteristics may reveal their origins. In this study, we collected 310 solid ash samples from various continents, countries, and origins and combined particle swarm optimization, random forest (RF), and novel interpretation methods to build an accurate ash origin detection system. The RF model took major oxides as inputs, without further feature-engineering, and automatically classified the solid ash into four origins. Our model predicted solid ash origins with 98.5% accuracy on a training set and with 91.5% accuracy on an independent testing set, even without expert knowledge of the operating conditions. Our approach also demonstrated feature importance and interpreted the decision-making mechanisms underlying optimum RF models. These results demonstrated the feasibility of the proposed modeling framework as a blueprint for automated machine-learning origin detection of solid ashes. Graphical abstract: Image 1 Highlights: Chemical signatures were proposed for the origin identification of solid ashes. Origin detection can promote the recovery of solid ashes. A global solid ash dataset was collected by an extensive literature review. The RF-PSO modeling framework can quickly and accurately estimate ash origin. Interpretation techniques were used to reveal the decision-making of RF-PSO.