Statistical Machine Learning for Improving Failures Detection

本项目旨在通过整合统计机器学习和数据来创新油田部门 科学融入到日常运作中该项目包括将人工智能技术嵌入到 运行中的钻井和压裂设备的实时传感器。这些传感器将能够探测到 任何流体中的微观金属颗粒和污染物。因此，大量的测试数据将 包括：机械数据、冶金数据、地质数据、流体性质数据、环境数据 操作数据、CAN总线数据（涵盖发动机、电气、变速器的子系统）以及 电化学阻抗数据。我们的愿景是，这些传感器不仅可以检测趋势， 组件即将发生故障，但也将从数据中学习。 这个问题不会没有任何挑战。最大的斗争将是组织和选择最好的 这些大数据集中的功能可以充分预测机器故障。然而，为了克服这个问题， 我们计划实施用于特征选择的最先进的无监督学习算法。一旦特征 我们将联合收割机结合不同的监督学习算法， 和高效的学习者能够提高自己与传入的数据。

The present project seeks to innovate the oil field sector by integrating statistical machine learning and data science into its day to day operations. This project consists of embedding artificial intelligence technology in real-time sensors of running drilling and fracturing equipment. These sensors will be capable of detecting microscopic metallic particles and contaminants in any fluid. As such, the abundant amount of testing data will include: mechanical data, metallurgical data, geological data, fluid property data, environmental data, operational data, CAN Bus data (subsystems that span engine, electrical, transmission) as well as electrochemical impedance data. It is our vision that these sensors will not only detect trends indicating imminent failure of components, but will also learn from the data. This problem won't be without any challenges. The biggest struggle will be to organize and select the best features amongst this Big Data set to adequately predict machine failures. Nonetheless, to surmount this issue, we plan to implement the state of art unsupervised learning algorithms for feature selection. Once the features have been chosen optimally, we will combine different supervised learning algorithms to create an adaptable and efficient learner capable of improving itself with incoming data.