Big Data Analytics for Prognostic Foresight

摘要

Objectives/Scope: With the rise of big data analytics in petroleum engineering, a plethora of data collectors, sensors, transmission devices, and software tools is entering our lives at the personal and professional levels. Data is extracted, transferred, processed, and utilized-be it from wearable devices, refrigerators, or motorcycles. It also spans areas such as public facilities, transportation networks, and major industrial assets. In the sphere of petroleum asset management data analytics is certainly not a new concept. For decades, equipment condition data has been explored and exploited. Internal data networks, repositories, and historians, along with the condition monitoring, predictive diagnostics, and performance optimization applications show a long tradition of utilizing data to gain useful insight. Methods, Procedures, Process: Current asset management technologies can identify what is going on, where, why there are challenges, and how they can be resolved. Yet, new prognostic analytics emphasizes data-driven forecasts addressing "when" questions, such as "When will my compressor have a malfunction?". Presently, such forecasts have no real proxy in petroleum asset management. The questions "When will my bearing fail?", "When is my last chance for plant revision?", and "When will turbine replacement be cheaper than refurbishment?" are widely asked but rarely answered. Results, Observations, Conclusions: We will review herein which major challenges prognostics faces in the petroleum industry sphere, how they are currently being resolved, and what benefits can be achieved. We present, in addition, a detailed example of an application of prognostics in a petrochemical manufacturing plant. Condition and process data of a cracked gas compressor is utilized to generate remaining useful life distributions and future malfunction risk profiles. The main objective of the operator is to reduce both downtime and maintenance costs.