Method Provides Real-Time Anomaly Detection for Drilling-Fluid Properties

Abstract

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 212443, “Real-Time Anomaly-Detection Methodology for Drilling-Fluid Properties,” by Moacyr N.B. Filho, Thalles P. Mello, and Cláudia M. Scheid, Federal Rural University of Rio de Janeiro, et al. The paper has not been peer reviewed. _ Online drilling-fluid-measurement technologies have become an essential tool for drilling automation. While online density measurements are widespread, the availability of rheology measurements is increasing quickly, and additional properties are central to ongoing field trials. The complete paper presents the concept of supervisory and advisory systems dedicated to support the detection of abnormal events and to provide guidelines for fluid treatment. Introduction The Laboratory of Fluid Flow of the Federal Rural University of Rio de Janeiro and Petrobras developed a flow loop for the proposal and validation of online sensors for the measurement of drilling-fluid properties. In previous work, the authors described the development of a set of online sensors to measure fluid properties with control of operational conditions of temperature, pressure, and fluid flow rate. Following previous research, this study proposes a new method using a technique based purely on data to diagnose and detect deviations in drilling-fluid properties. The method proposed applies principal component analysis (PCA) for detecting anomalies in drilling fluids. Method PCA-Based Anomaly-Detection Technique. This method consists of the linear transformation of the space of monitored variables, which generates a new space of variables called principal components, in which each principal component is a linear combination of the process variable. This allows the new space of variables to be linearly independent—that is, the principal components are linearly independent, allowing the calculation of process monitoring statistics. The ability to detect anomalies in drilling-fluid properties is necessary to train the system in the normal state of operation. The training based on the normal state of operation was performed through the calculation of the principal-component matrix and the selection of the reduced principal-component matrix. Once the process model was trained, the monitoring of new samples was performed with the help of two statistical tools that described the health of the process operation: T² and Q statistics. The T² statistic measures variations in the main space. T² only detects variation in the subspace of the first principal components larger than those attributed to the dynamics of the process. The Q statistic denotes the change of events not explained by the principal-component model. It is a measure of the difference, or residual, between a sample and its projection on the model. The failure-detection procedure based on the use of PCA consisted of continuous assessments of whether process statistics were within or outside an acceptable range. A method previously proposed in the literature used control systems based on fuzzy logic to monitor properties of drilling fluids and propose corrections, which are performed automatically by an experimental unit containing actuators for additions of drilling agents as thickeners and viscosifiers. The application of fuzzy logic depends on inference rules, which must be provided to the system by the programmer. This is a limitation of corrective systems based on fuzzy logic.