This post originally appeared on ARC Advisory Group's blog, Industrial IoT/Industrie 4.0 Viewpoints
Salt River Project (SRP), the oldest multipurpose federal reclamation project in the United States, has been serving central Arizona since 1903. Today, SRP is based in Phoenix and provides the area’s electric power generation and water delivery. This blog focuses on the power generation portion of the business. SRP monitors seven fossil and four hydro power generation sites containing a total of 27 units with more than 500 asset models covering generators, turbines, feed pumps and other assets with over 50,000 available tags.
As is typical in the power industry, SRP’s customers expect very reliable service. SRP’s Performance Monitoring Center helps achieve this goal while controlling costs by identifying issues before they lead to equipment failure with extended unplanned downtime.
The performance monitoring center uses GE Digital’s SmartSignal, part of the Asset Performance Management (APM) Reliability Management suite, to generate predictive analytics. SmartSignal uses machine learning techniques to automate the building of models, in this case, using SRP’s historical data collected from the plants. The algorithms in these models have been carefully trained by experienced engineers to replicate SRP’s operations. The software takes readings every 10 minutes, and if the equipment doesn’t perform as expected, the software raises an alert. From the assets and associated tags, thousands of individual data point streams go into the models continuously.
SmartSignal covers steady-state, baseload power generation. The solution also covers assets used for peak loads that often start up and shutdown and then compares the data at each point in the cycle to similar expected behavior from reference “golden” starts. Removing the expected variation provides indications of the subtle changes from cycle to cycle.
The analysts in the performance monitoring center have decades of experience at SRP. This provides them with a clear understanding of the equipment and how it functions, helping ensure buy-in from the plants. When alerts occur, the analysts review them to remove false positive alarms and further train the model. For the real alerts, meaningful information is added to make them actionable. These alerts are forwarded to the operating plants who take ownership for the associated activities.
Since implementing SmartSignal in 2012, SRP has identified more than 1,900 issues by 2016, of which 800 were “catches.” A catch is a problem that the plant was not previously aware of and, with new alerts, was able to take corrective action. With time and improved training of the algorithms, the rate at which the company identifies true issues and catches has improved.
Editor’s note: Andrew Johnson, Engineering Supervisor, Salt River Project, presented “Performance Monitoring Center” for asset monitoring and diagnostics during the ARC Forum session, “Moving to Predictive Maintenance with Industrial IoT (IIoT).” Readers can view a video of Johnson’s Forum presentation here.