Article: Wear prediction of agricultural machinery contact parts based on Bayesian optimisation Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.19 pp.76 - 90 Abstract: Accurate wear prediction of agricultural machinery contact parts is very important for maintenance decision-making. However, due to the scarcity of field data, data-driven methods are prone to overfitting and lack uncertainty quantification. To fuse probabilistic calibration and physical interpretability, a physics-informed Bayesian optimised Gaussian process regression guided by physical information was proposed. In this method, the classical wear law was integrated into the training as a priori, the dominant factors were automatically screened and the parameters were optimised, and the knowledge transfer between laboratory and field data was realised by using the dimensionless hardness index. Experiments show that the root mean square error of the model is reduced to 23.4 grams, and the negative log prediction density is 3.21, which is 22.3% and 25.5% higher than the optimal baseline, respectively. The framework provides a probabilistic reasoning basis with confidence boundary for the condition-based maintenance of intelligent agricultural machinery. Inderscience Publishers - linking academia, business and industry through research

Title: Wear prediction of agricultural machinery contact parts based on Bayesian optimisation

Authors: Feng Yan; Xiaofeng Zhang

Addresses: Agricultural and Rural Comprehensive Service Center, Linqu County Chengguan Subdistrict Office, Linqu, 262600, China ' Agricultural and Rural Comprehensive Service Center, Linqu County Chengguan Subdistrict Office, Linqu, 262600, China

Abstract: Accurate wear prediction of agricultural machinery contact parts is very important for maintenance decision-making. However, due to the scarcity of field data, data-driven methods are prone to overfitting and lack uncertainty quantification. To fuse probabilistic calibration and physical interpretability, a physics-informed Bayesian optimised Gaussian process regression guided by physical information was proposed. In this method, the classical wear law was integrated into the training as a priori, the dominant factors were automatically screened and the parameters were optimised, and the knowledge transfer between laboratory and field data was realised by using the dimensionless hardness index. Experiments show that the root mean square error of the model is reduced to 23.4 grams, and the negative log prediction density is 3.21, which is 22.3% and 25.5% higher than the optimal baseline, respectively. The framework provides a probabilistic reasoning basis with confidence boundary for the condition-based maintenance of intelligent agricultural machinery.

Keywords: wear prediction; Bayesian optimisation; Gaussian process regression; GPR; uncertainty quantification.

DOI: 10.1504/IJRIS.2026.155633

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.19, pp.76 - 90

Received: 28 Apr 2026
Accepted: 24 May 2026
Published online: 07 Aug 2026 *

Title: Wear prediction of agricultural machinery contact parts based on Bayesian optimisation

Keep up-to-date