Nonparametric life consumption modeling of high end drilling tools

Abstract

Equipment, process, and system prognostic techniques can be classified as belonging to one of three major classes of methods: (1) conventional reliability-based using failure times (Weibull), (2) population based with environmental considerations (e.g. proportional hazards modeling), and (3) individual based (e.g. general path model). This paper describes a new individual based, nonparametric life consumption model (LCM) that builds on the previously described path classification and estimation model (PACE). In the original implementation of the PACE, the general path model (GPM), which is the foundation of the majority of the modern individual based prognosis algorithms, is recast as a classification problem, where a current device's cumulative stress path is classified according to a series of exemplar paths and the results of the classification are used to estimate the remaining useful life (RUL) of the device. While the PACE was shown to produce accurate estimates of the RUL, its use in ?real world? applications was found to be limited since it was founded on the premise of estimating the RUL, given an observed cumulative stress at a given point in time. In other words, to estimate the RUL, the complete stress history up to that point in time must be constructible, which is rarely the case for large scale, worldwide device deployments. This short coming was overcome by redefining the prognostic question from, ?what is the RUL of the device?? To, ?how much life has been consumed in the latest deployment?? By making this simple modification, the problem can be approached from a different angle. Rather than attempting to classify the cumulative device stress at an instant in time, the ?character? of the stress accumulation over a period of time can be classified. To do this, regression analysis is used on a device's cumulative stress path for a deployment to generate a set of parameters that define the ?shape? of the stress accumulation over a particular deployment. The calculated parameters are then classified according to the regressed parameters of example failure paths. The results of the classification are then combined with the life consumptions of the exemplar paths to generate an estimate of the individual's life consumed for the deployment. These life consumption estimates can then be easily aggregated over deployments to accurately assess the individual device's remaining life. The newly proposed method will be applied to data collected from the hydraulic steering tool of a high end drilling system used for deep oil exploration, with the objective of estimating and accumulating the life consumption of example steering tools over different deployments. This paper builds on previous work in this area by developing a new life consumption estimation model that has been specifically designed to ensure that it can be viable in the ?real world?. The developed model was shown to be able to estimate the life consumed of an individual drilling tool to within 4-12% with uncertainties of ? 15-35%.