With Cognite Data Fusion, data scientists at BHGE were able to begin modeling and training with reliable data faster, build and test more models in a shorter time span, and decrease the time to delivery for their client, Aker BP.
Time from signed contract to full data access
Solution space increased 10x over a typical project
Many energy companies are talking about the value and potential of AI, but few have managed to push real projects into operation. Baker Hughes, a GE company (BHGE) has the domain expertise to apply machine learning, artificial intelligence, and advanced analytics for their industrial clients. But they have continued to struggle with the biggest bottleneck for scalable ML/AI solutions: access to all necessary data in a format that is readable and clean.
Aker BP wants to predict gas turbine failure. With BHGE’s help, they prepared a test case, using historical data to train machine learning models to detect operational anomalies that preceded failure on two of their turbines (GT-100 and GT-300). BHGE’s failure prediction method began detecting anomalies in collected time-series data, then utilized these “outliers” to predict future asset failures.
Cognite Data Fusion enabled BHGE to access the relevant data with surprising speed. The day after the project agreement was signed, Aker BP was able to grant BHGE access to all the necessary data in CDF via a single, secure point of entry using APIs.
The pilot project revealed a way to predict failure approximately four weeks in advance, potentially saving huge amounts of time, money, and offshore manhours. BHGE and Aker BP are preparing to operationalize and scale the solution. As they hash out the details, the project has already proved another kind of impact.
Data scientists want to move fast from Search to Discovery. Cognite Data Fusion removes the grunt work of identifying, locating, accessing, and cleaning data. Freeing data experts to spend their time, expertise, and creativity on the more important and more potentially profitable question: What is the best model to build?
For this project, BHGE’s data science team was able to explore a much larger solution space. In a normal pilot project, BHGE’s data science team might have built between 5 and 10 models to test. This time, they built a few hundred. Increasing the possibility of finding the right model while simultaneously saving both time and money on data science resources.