One great challenge for asset management is to analyze large amounts of data to provide insightful information for decision making in a timely fashion. Analyzing all available data manually is infeasible and inefficient. It is essential to develop pattern recognition algorithms to recognize events-of-interest to achieve effective asset management. Conventional pattern recognition algorithms require a fairly large training set in which data points are carefully prepared and labeled (e.g., normal, gas locking, failing valve) by subject-matter experts. This process is very time consuming and is error prone when the training set is large. In fact, many other industries are facing the same challenge where the cost of acquiring labels is too expensive or time-consuming to be feasible and large amounts of unlabeled data are available. Studies show that, for most applications, a small amount of data points are already labeled by the experts through their routine activities. While these data points are usually not enough to form a training set for conventional pattern recognition methods, some of the newer methods can take advantage of them along with the hidden manifold structures manifested by the unlabeled data. If necessary, some methods can also carefully select a small subset of data and collect subject-matter experts’ feedback on them to improve model development. In this paper, we will review existing methods, such as semi-supervised learning and active learning, and report our observations about the effectiveness of these methods in some real-world asset management scenarios.
Number of Pages
Looking for more?
Some of the OnePetro partner societies have developed subject- specific wikis that may help.