Smart Monitoring and Diagnostics Systems at Major Power Plants

The electric power generation industry must use new technologies to shift how maintenance and diagnostic departments operate. More than 50 percent of the generation capacity within the United States is over 30 years old [NI1]. In many cases, these power plants depend on equipment that is operating close to the limits of its original, intended design lifetime. This increases the potential for equipment failure, which, depending on the type of equipment, can make the power being supplied more susceptible to outages and instabilities.

At one utility, a study claims that maintenance and diagnostic experts spend nearly 80 percent of their day traveling, sometimes across vast distances, to collect “health” information about equipment and only 20 percent actually analyzing this data for potential failure points. This utility estimates nearly 60,000 operating points must be manually collected each month by personnel [NI2].

The aging infrastructure and inefficient use of experts’ time combined with the fact that the number of industry experts is dwindling due to an aging workforce is quickly creating a critical resource bottleneck. This bottleneck may eventually cause more downtime on critical machines and lead to potential brownouts or blackouts across the grid.

Duke Energy, the Electric Power Research Institute (EPRI), NI, and a consortium of power generation producers are working on a solution to automate online equipment monitoring systems for decision support. The Smart Monitoring and Diagnostics Project (Smart M&D) [NI3] aims to continuously and remotely monitor plant equipment for changes in measured parameters, run prognostics and advanced pattern recognition routines, and enable more informed  real-time decisions to optimize plant equipment and prevent failure.

Gathering analog sensory information, the core of this project, poses some unique challenges. For instance, vibration information is a good leading indicator of equipment breakdown. However, to collect vibration information, it may be necessary to capture anywhere from 10,000 to 100,000 samples per second for several seconds in order to obtain a good “measure” of the machine. Imagine, as the real world scenario in Figure 2 highlights, if you had 30,000 vibration points each collecting five seconds worth of data at 100,000 samples per second, every hour; that adds up to nearly 60Gb of data per hour!  This collection of information, if not managed and architected properly, can quickly lead to a “big data” problem.

Another challenge is that analog sensory information on its own does not tell the operator whether a machine is “good” or “bad.” Correlating multiple sensing types, processing data using mathematical algorithms, and using advanced pattern recognition techniques provide a true picture of the machine’s health.

A core element of the Smart M&D Project is the CompactRIO platform from NI [NI4]. By connecting a Field Programmable Gate Array (FPGA) and an onboard real-time processor to the sensor, raw analog waveforms can be reduced to conditions indicating the “health” of the system at the node itself.

FPGAs help analyze and process the high-speed sensory information in a very efficient, parallel manner for real-time decision making. Because the “smarts” in the system are close to the sensor and intelligent algorithms can be implemented directly on the CompactRIO system, data can immediately be reduced to known events. This prevents the data overload condition in which subject matter experts are stuck looking for problems that are difficult to locate.

The distributed, open, and reconfigurable nature of the system also plays an important part. Because the systems are distributed, a wide range of “health” information can be collected by similar systems (similar boiler feed pumps, fans, motors, and more) directly by the network of machines, and intelligence can immediately be applied to the data at the source. As the systems are analyzing data constantly, this means operator rounds can be greatly reduced while dramatically increasing the frequency of collection. Data no longer needs to be collected every month, 6 months, or year—it can be collected several times PER DAY. Issues can be discovered and tracked on a more frequent, consistent basis.

Further, advanced diagnostic and prognostic algorithms, such as those contained within the EPRI Asset Fault Signature Database™ and EPRI Remaining Useful Life Database , can be used to predict equipment failures before they occur. For example, the EPRI Asset Fault Signature Database™ characterizes equipment failure mechanisms by a collection of typical attributes or symptoms, such as temperatures, vibrations, lubrication analyses, and other diagnostic results. Using this type of real-time data generated by a SmartM&D system, comparisons can be made to diagnostic models contained in the EPRI Asset Fault Signature Database™. When a set or subset of data coincides with the known attributes or symptoms of a known failure mechanism and/or location, these diagnostic tools can identify when a particular failure is impending..

Finally, the reconfigurable nature of CompactRIO means as standards change, new algorithms are developed, or additional sensing technologies become prevalent customers can update their intelligent nodes without having to physically go out into the plant to update them or having to reinvest new capital dollars in order to solve a new problem.

Asset Integration Architecture of Smart M&D

The figure below shows an overview of the asset integration architecture for the Smart M&D Project. The system can be broken into two main components:

• Data Acquisitions Systems—DAQs are a crucial part of the system and provide data from numerous different sensors hooked up to a variety of machinery components and types. The DAQs are distributed throughout a facility, region, or global geography. DAQ systems perform onboard processing and extraction of key sensory metrics for future trending, alarming, and analysis. They are intelligent devices that can perform in the absence of a network.
• Sensor Fusion and Analysis—The piece of the system responsible for providing actionable data to system operators, subject matter experts, management, and others. Through a variety of open communication protocols and file formats, data from various sensors is fused together to provide a complete picture of the asset health.

Lessons Learned

The following summarizes the key learnings from this project.

• The reconfigurable nature of the system provides interfaces so as new algorithms, industry protocols (61850, DNP3, and more), and sensor types are created, the infrastructure does not need to change. Simply download new information to the embedded systems and begin collecting new fault signatures. The system can expand from the classical method of diagnosing machine faults with measurements such as temperature, vibration, pressure, and more, to incorporating advanced measurements like thermal imaging, ultrasonics, smell, and EMI interference.
• Providing an open platform encourages other system vendors to adopt the Smart M&D standard of connectivity. It is naïve to believe there will only be one type of acquisition system, back end database, enterprise analytic software, or more for a given facility. Therefore, a system needs to be created that can be inclusive of many systems. No longer is the data hidden away in proprietary formats but rather open for users to run personalized algorithms, connect unique sensors in one package, and provide an ecosystem for expansion.
• The use of Internet of Things technologies will provide an open, integrated, and flexible framework for service providers, suppliers, and users; thus increasing the operational efficiencies of plants, reducing downtime, and increasing the availability of energy on the grid.
• Currently, at Duke Energy nearly 1,500 CompactRIO systems are deployed and managed by the Smart M&D architecture across 30 facilities.

We would like to thank Stuart Gillen and Jamie Smith from National Instruments for the contribution of this case study.