Loading...

Study and Prediction of the Onset of Biomass Fluidization by the Statistical and Machine Learning Approaches

Akhoondi, Amir Hossein | 2025

60 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58820 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Fotovat, Farzam
  7. Abstract:
  8. This thesis presents a comprehensive and rigorous investigation of modern methods for predicting the minimum fluidization velocity (Uₘf) and the voidage at incipient fluidization in binary fluidized beds, with a primary focus on the complexities inherent to biomass particles. The overarching aim is to improve predictive accuracy by synergistically integrating machine learning techniques with refined empirical correlations. The study follows a two stage approach: first, a systematic evaluation of diverse machine learning algorithms is conducted to assess their ability to predict key fluidization parameters with high fidelity; second, the established correlation of Ago et al. is carefully revisited and refined, leveraging the concept of particle segregation to more realistically capture complex nonlinear behaviors and intrinsic features of biomass particles. Among the models examined, tree based algorithms, especially decision trees, delivered the best performance for predicting voidage, achieving R² = 0.99 and RMSE = 0.001; incorporating these ML based predictions into the Ergun equation reduced the RMSE of minimum fluidization velocity estimates by 12.5% compared with the most accurate existing empirical correlation (Kumar and Gupta), and yielded 39% more predictions within the acceptable error range (>30%) relative to the Goossens correlation. Furthermore, the research introduces three new relations for the interaction parameter β, derived from a particle segregation model and the use of the gplearn library in bubbling beds; these relations are physically plausible and adapt dynamically to different segregation regimes. Employing this modified β reduced the mean absolute error of voidage prediction from 0.024 with R² = −0.42 in the original Ago et al. formulation to 0.010 with R² = 0.83 in the proposed model. Incorporating the revised β into the Ergun equation also lowered the mean absolute error in predicting the minimum fluidization velocity by 30%, decreasing MAE from 0.061 m/s in the study by Ago et al. to 0.044 m/s in the present work. In sum, by bridging data driven approaches and theoretical understanding, this research delivers a precise, robust, and practical tool for the design and operation of fluidized beds, demonstrating that combining machine learning with refined physical models can provide a reliable and scalable foundation for applications across chemical, energy, and pharmaceutical domains
  9. Keywords:
  10. Minimum Fluidization Velocity ; Machine Learning ; Empirical Correlation ; Particles Separation ; Segregation ; Porosity ; Biomass Fluidization

 Digital Object List

 Bookmark

...see more