Search

The objective of this work was to apply artificial intelligence techniques, such as computational vision, image processing, and machine learning, to develop a software for performing the analysis and monitoring of cuttings in vibrating screens to detect flow variations and possible instabilities during the drilling process. An experimental unit was built to emulate the flow of cuttings in the sieves, where three different vibration intensities were applied to the apparatus to allow for three distinct flow speeds. Flowing experiments using quartz, marble, granite, and actual gravel fragments were recorded in videos and used by the computational vision software developed on the Matlab©platform. Videos were processed by the proposed algorithm using different image processing and machine learning techniques. The 2D Discrete Fourier Transform associated with the convolution theorem was the segmentation technique used to detect the rocks in the frames. The estimation of the solids’ velocity was performed with the detectors of correspondence points in sequential frames of the video using the MATLAB’s functions “MSER”, “Harris”, and “SURF”. The MATLAB’s function "regionprops" was also used to estimate the area of the sieve filled with rocks as well as some geometric parameters of them, such as size, circularity, roundness, and eccentricity. The results showed that the proposed computational vision software could perform reliable estimates of the reported features for a range of rock types. Besides, in most cases, it was possible to estimate the area of the sieve filled with rocks and their velocities across the sieve with average errors of less than 10%. Results also suggested that it could be possible to implement a real-time monitoring system for vibrating sieves in drilling operations using the software developed in this work.

Although many advanced nonlinear process control techniques have been developed over the past decade, classic control based on feedback response still has its place. This is mostly so because feedback empirical control is robust and simple to implement and does not require fancy calculations or high-qualified operational manpower to operate it. This work has developed an application written in LabVIEW®environment capable of doing a fully automated single-input single-output control. Preliminary tests were performed in a drilling fluid production unit, controlling flow rate through manipulation of the pump power engine. In the future, tests in the same plant of pressure control by choke valves manipulation will be performed (as found in rig sites, where wellbore pressure is controlled by manipulation of such valves). The final goal is to implement such software in a real rig site, to help operators in drilling control areas such as flow rate and wellbore pressure. The produced software has embedded three self-developed features: automatic plant identification (API), auto-tuning (ABAP) and controllers auto-switch (CAS). The API determines automatically the linearity of the process determining the empirical parameters according to Sundaresan and Krishnaswamy technique. In sequence, it calculates the parameters for P, PI and PID controllers using Cohen–Coon and Ziegler–Nichols methods. The API method automates the sequence of tests necessary to implement the Sundaresan and Krishnaswamy empirical approach. The ABAP feature based on heuristic rules tunes in real time the controllers’ parameters to optimize its response. The CAS allows automatic switch between controllers and parameters to avoid instability, overshoots and creates a synergy with ABAP feature. The results have shown that the API feature is a good optimizer reducing the invested time to calculate all the parameters, from hours to a few minutes. The CAS results demonstrated an associative property with the ABAP feature to mitigate instabilities and overshoots. Therefore, the preliminary results suggested this software is a unique and important tool to improve performance, profitability and reliability during offshore and onshore drilling operations. Moreover, this application could be used in any industry with an approximate first-order dynamic system due to its robustness and a low human interaction need.