PREDICTION OF FAULTS IN TECHNICAL SYSTEMS BASED ON THE SIMILARITY MODEL OF THE REMAINING USEFUL LIFE
Abstract
This paper demonstrates how to construct a complete Remaining Useful Life (RUL) estimation workflow, including the steps of preprocessing, selecting trend features, constructing a health indicator by fusing sensors, training RUL similarity estimators, and verifying the prediction performance. The method was tested in a MATLAB demo program implementing this method for predicting the occurrence of faults in technical systems (https://www.mathworks.com/help/predmaint/ug/similarity-based-remaining-useful-life-estimation.html) based on data from the "PHM08 Challenge Data Set", NASA Ames Prognostics Data Repository (http://ti.arc.nasa.gov/project/prognostic-data-repository), NASA Ames Research Center, Moffett Field, CA. The method is focused on the use of reasonable technical characteristics of the equipment being estimated, which are sufficiently covered in the reference literature. Therefore, the method gives good results when assessing equipment whose operating conditions are close to the statistical average. This paper uses the Predictive Maintenance Toolbox™ in MATLAB, which includes several specialized models developed for calculating RUL from various types of measured system data. These models are useful when you have historical data and information, such as: ‒ failure histories of machines similar to the one to be diagnosed. The historical data for each member of the data ensemble is fitted to a model of identical structure; ‒ a known threshold value of some condition indicator indicating failure; ‒ data on how much time or how much use it took for similar machines to fail (service life). RUL estimation models provide methods for training a model using historical data and using it to make a remaining service life prediction. The term service life here refers to the useful life of a machine defined in terms of any quantity used to measure the service life of a system. Similarly, time evolution can mean the evolution of a value with usage, distance traveled, number of cycles, or another quantity that describes the service life. A general workflow for using RUL estimation models is: ‒ create and configure the corresponding model object; ‒ train the estimation model using the available historical data; ‒ using test data of the same type as the available historical data, estimate the RUL of the test component. It is also possible to use the test data recursively to update the model as new data becomes available, i.e. track the evolution of the RUL prediction as new data becomes available.
