Optimisation of parameters of a dual-axis soil remediation device based on response surface methodology and machine learning algorithm--颗粒学报PARTICUOLOGY

Institute of Process Equipment and Environmental Engineering, School of Mechanical Engineering and Automation, Northeastern University, Shenyang, 110819, China

10.1016/j.partic.2024.10.013

Volume 96, January 2025, Pages 26-43

Received 22 August 2024, Revised 3 October 2024, Accepted 10 October 2024, Available online 2 November 2024, Version of Record 14 November 2024.

E-mail: 448465250@qq.com; tongzhu@mail.neu.edu.cn; wangyz@me.neu.edu.cn

Highlights

• Machine learning and response surface methodology study of soil-agent mixing characteristics.

• The predictive performance of three machine learning models is compared.

• Improvement of mixing homogeneity in a new biaxial in-situ soil remediation plant.

• Provides reference values for the design of multi-axis soil remediation equipment.

Abstract

To accelerate the recycling of black soil, it is necessary to develop a new type of soil remediation equipment to improve its working efficiency. The one-way test was used to determine the mean level value of the steepest climb test, and the combined equilibrium method was used to determine the upper and lower interval levels of the response surface test for parameter optimisation. Based on the results of the response surface indices, machine learning was performed and the optimal model was determined. The results show that the predictive ability and stability of the decision tree model for the two indicators are better than that of random forest and support vector machine. The optimal parameter combinations determined using the decision tree model are: speed 73 rpm, homogenisation pitch 183 mm, homogenisation time 1 s, descent speed 0.06 m/s. The error between the optimal value of the machine learning prediction model and the actual simulation is 1.1% and 5.72%, respectively. The results of the study show that the effect of optimizing the parameters through machine learning achieves a satisfactory prediction accuracy.

Graphical abstract

Keywords

Black soil; Machine learning; Mixing homogeneity; Discrete element method; Response surface methodology

文章导读