A NEW METHOD FOR PREDICTING THE REMAINING EQUIPMENT LIFE FOR HIGH-FREQUENCY DATA WITH NON-UNIFORM DUTY CYCLES

Abstract

Advances in mechanical engineering make it possible to create more advanced and efficient
equipment, but at the same time, its complexity and the requirements for managing its life cycle
and maintenance increase. Requirements for reliability and availability also create additional
challenges to life cycle management. There are various maintenance planning strategies. Among
them, the most promising is the predictive strategy based on forecasting the remaining useful life
of the equipment. Existing methods for predicting the remaining useful life of equipment focus on
the use of historical data aggregated by work cycles, while there are no widely used methods for
forecasting using continuous data, including high-frequency data, received from equipment and
containing work cycles of various durations and data recorded during downtime. To solve this
problem, a method for predicting the remaining useful life is proposed with the determination of
work cycles in the initial data and the aggregation of their values into one-dimensional vectors for
the purpose of further use for training the forecasting model. The results demonstrate the successful
applicability of the proposed method - in combination with the XGBoost forecast model, it is
possible to achieve accuracy on data obtained from a gas turbine engine with a root mean square
error of 14.02 and mean average error of 10.71.