Predicting ash accumulation in industrial systems using machine learning: Enhancing maintenance and operational efficiency--颗粒学报PARTICUOLOGY

Godasiaei, S. H. (2025). Predicting ash accumulation in industrial systems using machine learning: Enhancing maintenance and operational efficiency. Particuology, 103, 41-54. https://doi.org/10.1016/j.partic.2025.05.008

Predicting ash accumulation in industrial systems using machine learning: Enhancing maintenance and operational efficiency

Seyed Hamed Godasiaei^*

School of Chemical Engineering and Technology, Xi'an Jiaotong University, Xi'an, China

10.1016/j.partic.2025.05.008

Volume 103, August 2025, Pages 41-54

Received 9 April 2025, Revised 3 May 2025, Accepted 7 May 2025, Available online 22 May 2025, Version of Record 28 May 2025.

E-mail: Hamedgoodasiay@gmail.com

Highlights

• Machine learning framework explores critical combustion parameters.

• High accuracy was achieved with Random Forest and XGBoost.

• Support Vector Regression shows superior speed in training.

• Correlation analysis shows strong ties between deposit thickness, time, and heat flux.

Abstract

Predicting ash accumulation in industrial environments is crucial for improving operational efficiency, enabling proactive maintenance, reducing downtime, and optimizing plant performance. Understanding of these processes requires the analysis of key parameters, including time, heat flux, particle size, velocity, excess air ratio, furnace temperature, heat load, and oxide concentrations, with a particular focus on deposition thickness. Traditional methods often fail to capture the complexity of these interactions, necessitating innovative approaches for accurate prediction and analysis. The experimental data, along with four algorithms, i.e. Support Vector Regression (SVR), Random Forest (RF), Deep Neural Network (DNN), and Extreme Gradient Boosting (XGBoost), were employed to analyze 20 features, providing a robust evaluation of their predictive capabilities. Furthermore, the use of SHAP (SHapley Additive Explanations) values introduces a novel dimension to the study, enabling interpretability and transparency in understanding the contribution of each feature to the model's predictions. The results demonstrate exceptional predictive accuracy for the RF and XGBoost models, achieving an R2 value of 0.99 and minimal mean absolute errors (MAE). A novel comparison of training times reveals that SVR outperforms the other algorithms in speed due to its simpler structure, making it highly efficient for real-time applications. Correlation analysis identifies strong relationships between deposition thickness and key parameters such as time, heat flux, and deposition probability at varying surface temperatures. Time directly influences deposition thickness, as particles accumulate and sinter over prolonged operation. Heat flux drives particle movement through thermophoresis, affecting surface adhesion and increasing deposition probability. Surface temperature modulates particle adhesion and slag viscosity, with optimal temperatures maximizing stickiness and deposition probability.

Graphical abstract

Keywords

Ash deposition; Boiler; SHAP; Predicting; Slag formation dynamics

文章导读