The first delivery phase of this program of work established the Google Cloud environment for Senex and focused on some key, prioritised use cases.
The most promising use case focused on moving a torque spike model, which helps predict failures and faults, which has the potential for a significant impact on both the business and Senex’s bottom line.
A Torque spike is a strong predictor of pump failure. A torque spike means the pump is suddenly working harder, which could be because of solids contaminating the outputs or some other cause. The pumps are costly to run and replace, taking a forego of sales whilst the wells are offline.
Servian, working in conjunction with Senex resources and supported by Google Cloud, rapidly took an existing model and re-deployed it to Google Cloud, with some refactoring to help it leverage cloud capabilities.
The new Google Cloud data platform hosts a Torque Spike ML model that uses key SCADA measures (from sensors at the wells) to identify wells at risk of torque spikes. A data pipeline ingests live data from SCADA Historian application data into Google BigQuery tables – which then is the source for the Tensorflow Machine Learning model to predict torque events for each window of 15-minute data. A front end application (hosted in a VM on Google Cloud) visualises the raw data. It predicts torque events in several ways, allowing engineers to quickly identify torque events and diagnose underlying causes associated with a particular torque event.