Automatic threshold tracking of sensor data using Expectation Maximization algorithm

In this paper we present a novel method for automatic threshold handling and tracking of sensor data at drilling rigs. A hybrid system for automated drilling operation classification is extended by the Expectation Maximization algorithm in combination with the Bayes' theorem to find automatically threshold values required by a rule based system used in an automated drilling operations classification system. The streaming data from the rig site is gathered and analyzed, the main clusters in the sensor data are identified and monitored as in a real life case. The first part of the suggested method is based on the Expectation Maximization algorithm which is used to decompose Gaussian mixture models in the sensor data set. Bayes' theorem is used as a subsequent part to calculate optimal threshold values. The threshold values calculation concept is heavily depending on the likelihood probabilities of each data cluster. The work in this paper not only suggests a solution and analytical method for tracking this kind of thresholds in the sensor data but also verifies how to compute such reliable thresholds in real-time.