Agriculture Vehicles Predictive Maintenance with Telemetry, Maintenance History and Geospatial Data

Abstract

Timely detection and prevention of agriculture vehicles malfunctions are key approaches to reducing maintenance costs, as well as updating and replacing equipment, and reducing the cost of growing agricultural crops. In this article an approach for Remaining Useful Life (RUL) prediction that utilizes a combination of telemetry, maintenance, and geospatial data (such as weather and terrain information) as input to a Long Short-Term Memory (LSTM) algorithm has been considered. The results have shown that the models trained on the dataset enriched with geospatial data outperformed the models that relied solely on telemetry and maintenance data, demonstrating the benefits of including location-specific factors. However, the model's RUL prediction applicability for electric and hydraulic systems needs further exploration due to the current dataset limitations.