Improving the reliability of asphalt pavers through prediction of the residual life of main units and assemblies

Abstract

Maintenance strategies based on calendar intervals and fixed operating time correspond poorly to the actual degradation rate of asphalt paver units: the scatter of the service life of the hydraulic drive at equal operating time reaches values at which scheduled replacement turns out to be premature or belated. The aim of the study is to construct a computational scheme for predicting the residual life of the main paver units from diagnostic features and to estimate the attainable level of fleet availability. A two-parameter Weibull distribution, a gamma degradation process and a hybrid scheme with gradient boosting over the diagnostic feature vector are applied to an author's model of a fleet of twelve tracked pavers over the 2021-2026 seasons. The hydraulic drive accounts for 34,2% of failures and 36,8% of total downtime at a mean operating time between failures of 413 engine hours, and the decisive contribution to forecast accuracy comes from working fluid contamination and the volumetric efficiency of the travel pump, jointly 0,58 on the feature importance scale. The hybrid scheme reduces the mean absolute error of the forecast from 218 to 87 engine hours and raises the fleet technical availability factor from 0,782 to 0,871 at a horizon of 150 engine hours; extending the horizon to 300 engine hours degrades the result. The limit of the predictive strategy is therefore set by the rate of diagnostic information accumulation, and increasing the complexity of the computational scheme does not shift it.