The Electric Power Research Institute has developed a pair of diagnostic tools that combine and integrate features from multiple sources of plant information. The Diagnostic Advisor and the Asset Fault Signature Database will improve diagnostics for and troubleshooting of equipment faults by providing a holistic view of the condition of plant equipment.

In competitive environments, electric power plants must operate under reduced operation and maintenance (O&M) budgets while maintaining high reliability and availability. Early detection of equipment faults and subsequent planning of maintenance actions can help cut costs while maximizing availability. However, the detection of equipment faults and subsequent troubleshooting often require information beyond what traditional process instrumentation provides. Improving the use of power plant information sources for equipment fault prediction and diagnosis can help electric utilities meet plant availability goals and reduce O&M costs.

In a new plant design, it is desirable to install sensors based not only on process control design but also on equipment fault detection needs, as identified through a structured failure modes effects analysis (FMEA) and/or fault tree analysis. At existing plants, however, information obtained manually through predictive maintenance or operator rounds can be used alongside process data for detecting and diagnosing equipment faults.

The Electric Power Research Institute (EPRI) has designed a new diagnostic analysis software application and database that assists electric power generation plant staff in the early identification of equipment faults. That early detection enables rapid incident response and prevents failures of critical power generation equipment. This article describes EPRI’s work in designing a pair of tools that combine features from multiple sources of plant information to assist with troubleshooting and diagnostics of plant equipment.

Recent Trends in Anomaly Detection

Many of today’s electric utilities store years of historical process data in servers known as data historians. This large quantity of data contains information about process and equipment history that, when extracted, can indicate trends that provide insight into equipment condition. However, without data-mining tools, it is impractical to monitor this large amount of data continuously. As a result, most utilities have used data historians mostly for looking back at trends after a failure has occurred, as part of a root-cause analysis.

New techniques in data mining, combined with today’s efficient computing capabilities, have enabled continuous online monitoring of plant process data. Advanced pattern recognition (APR) techniques recently have been applied to tens of thousands of data points that often reside in multiple data historians across a fleet of power plants. This technique allows a utility engineer or technician to compare historical data with current data for hundreds of systems and components across multiple operating units, thereby identifying anomalous trends.

Although plant process data can be monitored efficiently today with APR techniques, traditional condition indications from predictive maintenance and operator rounds, for example, remain key to troubleshooting equipment problems. Unlike plant process data, however, this type of information often is stored in stand-alone, non-networked databases and can be accessed only by the technician(s) who collected the information. Web-based enterprise asset management systems offer an interface that can provide summaries of predictive maintenance and operator rounds results to staff across an entire fleet of units. With this type of information, staff tasked with monitoring and diagnosing plant equipment can make a complete assessment of equipment health prior to maintenance planning.

Centralized Monitoring and Diagnostics Strategy

Although a large quantity of information is available for the purpose of monitoring and diagnosing equipment, some key hurdles for efficient troubleshooting remain:

  • Site technical staff are burdened fully with everyday duties and cannot dedicate additional time to trending data.
  • Equipment faults often progress to failure quickly, and staff cannot react in a timely fashion.
  • It is time-consuming to assemble all relevant data for a given piece of equipment when a fault condition is known.
  • Maintenance is performed on a periodic basis, and the latest results may be out of date.

Some of these problems have been addressed by centralized monitoring and diagnostic (M&D) facilities that are responsible for detecting equipment faults upon their earliest indication and working with plant staff to plan the appropriate maintenance in the most cost-effective manner. These centralized facilities require up-front investment including:

  • Multi-disciplinary staffing that includes experienced operators, maintenance technicians, and engineers.
  • Information integration, including the connection of plant data historians and enterprise asset management tools to the central facility.
  • Brick-and-mortar facilities in a location central to monitored units.
  • Executive support for establishing an implementation plan and for communicating the need and benefit across the fleet.

When implemented successfully, M&D staff provide solutions to at least two of the key hurdles: limited plant staff time for trending data and the need to identify equipment faults early enough that plant staff can respond in a timely manner. Figure 1 shows a typical M&D center.

1. The solutions squad. Centralized monitoring and diagnostic staff provide solutions to at least two key hurdles at power plants: limited plant staff time for trending data and the need to identify equipment faults early enough that plant staff can respond in a timely fashion. Courtesy: EPRI

Diagnostic Advisor Summary

The role of the Diagnostic Advisor (DA), which EPRI currently is developing, is to infer the presence of specific fault conditions by monitoring the fault effects in the observable plant operating data. Using inference, the DA matches current plant information with a possible fault indication that may lead to a known fault condition. It is important to identify the unique fault condition as soon as possible to enable timely corrective action. Sometimes, the observed data might not match a single fault condition uniquely. In this situation, the DA will identify and rank the most likely fault conditions. In addition, the DA will identify troubleshooting or corrective action options to improve the diagnosis or remedy the problem if that information is available in a fault signature.

It is desirable that the DA provide broad coverage for plant and enterprise assets. The DA will operate with a user-specific or industry-wide fault signature knowledge base.

EPRI’s DA software system will facilitate the conversion of observed plant data to specific diagnoses, not just identified anomalies. The companion database will provide an extensible reasoning framework that, over time, will be filled systematically with knowledge specific to fossil plant systems, equipment, and components (all referred to collectively as “assets” in this article). The software and database will be flexible enough to integrate and process data from several sources, including:

  • Observed signatures of actual asset problems (faults) contributed by users of fielded online monitoring (OLM) and fleetwide monitoring (FWM) systems.
  • Simulated asset fault signatures (AFS) from plant simulators and/or physics-based models.
  • Asset fault-related health status information, such as operator rounds results, predictive maintenance data, and engineering assessments.
  • Theoretical AFSs developed by subject matter experts from FMEAs.

The DA will perform anomaly interpretation, diagnosis, and condition assessment more effectively than has previously been possible. This will be accomplished by combining the data types listed above to assemble fault signatures for specific assets so that, over time, theoretical FMEA-based AFSs will be refined and enhanced using operational experience across the industry.

The DA will consist of two primary elements:

  • An advisory software tool that provides the user interface and reasoning capability for performing anomaly interpretation, diagnosis, and condition assessment.
  • An AFS Database for capturing and organizing observed, simulated, and theoretical AFS information used by the DA software and by other applications, such as vendor-provided OLM and FWM software.

Asset Fault Signature Database Overview

EPRI’s AFS Database archives experientially derived fault signatures and design-theoretical-derived fault signatures based on FMEA/failure mode effects and criticality analysis (FMECA) information. The AFS Database allows for streamlined entry of new fault signatures and for efficient retrieval of fault signatures for diagnostic purposes. This type of power industry knowledge is not documented effectively in any other database.

The fault signatures will be used in conjunction with EPRI’s DA software to accomplish the diagnostic process; however, the AFS Database is not exclusive to the DA and could be used in conjunction with other diagnostic software, such as future versions of vendor-provided online monitoring software.

One of the main goals of the AFS Database is to capture the fault signatures in such a way that they can be used and exchanged across the power industry. This role of the AFS Database has very high value, as it is unlikely that a single unit, plant, or utility will experience all faults possible during an asset’s lifetime. Furthermore, because many fossil-fueled power plants are of an age at which faults are expected to increase, sharing of technology and experience enabled by this pair of tools is necessary in order to maintain high levels of plant reliability and availability. The DA and AFS Database might be implemented within the framework of an enhanced enterprise asset management system, as shown in Figure 2.

2. Cyber sleuths. The Diagnostic Advisor and Asset Fault Signature Database might be implemented within the framework of an enhanced enterprise asset management system. These tools will be used to diagnose and troubleshoot potential equipment failures. Courtesy: EPRI

The signatures contained in the AFS Database are based on fault features that result from examining raw data. A feature is the result of evaluating the data to determine whether it is normal or not. For example, a feature might indicate that the parameter measured by the data is high, low, open, closed, noisy, dirty, or has some other quality relevant for indicating a fault. This approach recognizes that raw data values indicating a fault likely will vary for each individual item of similar equipment, and thus cannot be generalized. Conversely, fault features are very similar for like items of equipment experiencing the same problem. Emphasizing the fault features enables the AFS Database to be used broadly across the industry.

In addition, OLM, FWM, and thermal performance monitoring systems, data historians, and technology experts already provide site-specific information and the capability needed to examine and to convert raw data into meaningful fault features. It could be redundant, complicated, and possibly wasteful to duplicate such site-specific information and capability within the fault signatures themselves.

A Conceptual Example of Signature Matching

Because diagnostic success depends heavily on the selection of the best stored signature, an important issue for the DA’s development will be the design of an effective fault feature-matching system. Ineffective feature matching may lead a query down a path of secondary symptoms and factors. It is important to establish a feature-matching system that will weight features selectively to indicate or to counter-indicate the applicability of each stored signature most effectively. An indexing structure is being developed that enables only the most relevant signatures to be considered.

It is likely that observed fault pattern data will be incomplete or ambiguous, particularly in the early stages of progression of a fault. Case-based reasoning can use the knowledge in the AFS Database to determine and to present a diagnosis even with partial matching information. A simple illustration of signature matching is provided in Figure 3. For a new observed fault pattern, relevant signatures (those having at least one fault feature in common with the observed fault pattern) are retrieved from the AFS Database using indexing to prune the search space.

Relevant similarity metrics are used to rank the retrieved signatures on the basis of the signature feature sets that best match the observed fault pattern. The most similar signature(s) are used to determine and to present the diagnosis. The solution presented will include the most likely fault indicated, along with possible high-ranking alternatives, a listing of confirming indications, a listing of missing indications that might strengthen or alter the diagnosis, and information about the remedy or outcome recorded in similar signatures (see Figure 3).

3. The match game. A simple illustration of signature matching is shown. For a new observed fault pattern, relevant signatures (those having at least one fault feature in common with the observed fault pattern) are retrieved from the AFS Database using indexing to prune the search space. Courtesy: EPRI

Looking Down the Road

The next steps in commercialization are to develop software for both the DA and AFS Database that can be integrated with utility enterprise asset management systems and used within current business processes. The successful use of the diagnostic tools just described will depend on peer sharing of fault signatures, connection of the DA to web-based portals in enterprise asset management systems, and open connectivity to third-party software (such as rule-based expert systems and advanced pattern recognition systems).

This article is based on a paper presented by the authors at the 15th Annual POWID/EPRI Controls and Instrumentation Symposium.

Aaron Hussey (ahussey@epri.com) and Stephen Hesler (shesler@epri.com) work in the Fossil Operations and Maintenance Program at the Electric Power Research Institute in Charlotte, N.C. Randall Bickford (rbickford@expmicrosys.com) is president of Expert Microsystems Inc. in Orangevale, Calif.