Robust system for monitoring the technical condition of construction and road vehicles with incomplete and noisy sensor data

Abstract

The article discusses the problem of ensuring fault tolerance of technical diagnostics systems for construction and road vehicles in conditions of incomplete and noisy data from on-board sensors. An architectural and algorithmic approach to building a robust analytical system based on adaptive models with dynamic delay and hybrid machine learning methods is proposed. A methodology for quantifying robustness has been developed that takes into account both the accuracy of predicting wear on components (hydraulic systems, chassis, engines) and the resistance to input data defects. The experimental test was carried out on real data collected during the operation of crawler excavators and pavers. The results showed that the proposed system maintains diagnostic accuracy at a level of at least 92%, even in the presence of up to 25% missing values and noise with a dispersion exceeding the nominal value by 3 times. The results obtained confirm the applicability of the developed approach to improve the reliability and safety of operation of construction and road equipment.