Pump Monitoring Reaches New Heights with Data Modeling in the Cloud

This pump installation is at a water-treatment facility in the Caribbean. Photo: Summit Pump Inc.

Data modeling in manufacturing is gaining steam as a strategy to optimize a process but also avoid capital expenditure (CapEx) budget hits. Companies are recognizing cheaper start-up costs with software as a service (SaaS) asset management solutions in operation expenditures (OPEX) and finding ways to provide quick, digital solution wins for management.

The problem with data modeling in years past was the lack of actionable, real-time data for plant operators/technicians. An ARC Brief from Paula Hollywood and David Clayton, titled, “Machine Learning Technologies Introduce a Step Change in Maintenance and Reliability,” discusses how Flowserve and a partnership with SparkCognition helped remedy data lag for key pump equipment.

Before SparkCognition’s involvement, Flowserve had been evaluating the potential of machine learning to improve its equipment monitoring capabilities for the past twenty years, but – until recently – these had proven too costly and difficult to commercialize, according to the ARC brief.

From the ARC Brief:

In the past, Flowserve custom engineered most of its failure analysis algorithms. The company would start with a failure mode and develop a failure analysis algorithm in-house. It could take as long as one to two years to develop the algorithm for each pump type. In addition, these customized algorithms typically could not adequately detect unknown operating states with variable process conditions.

The strategic change by Flowserve is due to more computing power and secure access to cloud analytics that can translate into real-time, actionable data that allows technicians to prescribe a scheduled downtime event.

The data modeling system also touches on “unstructured data sets” within the enterprise and how a system can continuously learn from past process batches or time periods. Unstructured data (or unstructured information) is information that either does not have a pre-defined data model or is not organized in a pre-defined manner. Unstructured information is typically text-heavy, but may contain data such as dates, numbers, and facts as well, according to Wikipedia.

As most legacy enterprises battle with mountains of field or plant data, plenty of unstructured (never analyzed) data exists. Now, historical data can be “understood” with advanced date modeling algorithms.

According to Flowserve, it can now address pump operating states, pump problems, real-time pump issues and past issues.

For the entire ARC brief, see “Machine Learning Technologies Introduce a Step Change in Maintenance and Reliability.“