Pattern Recognition & Data Validation

Advanced Pattern Recognition for Critical Data Validation - Economic, Environmental and Maintenance Footprint Shrinkage Strategies
Early Discovery of Failing Equipment and Sensors in Power Plants using Advanced Pattern Recognition
Equipment Condition Monitoring Through Advance Pattern Recognition
Application of Advanced Pattern Recognition for Nuclear Plant Performance Optimization
Precision Data Validation Boosts Process Optimization Benefits
Applied Pattern Recognition for Plant Monitoring and Data Validation
Similarity Based Regression: Applied Advanced Pattern Recognition for Power Plant Analysis

Advanced Pattern Recognition for Critical Data Validation - Economic, Environmental and Maintenance Footprint Shrinkage Strategies
Download

Abstract

The rapid growth of historians and, the advent of cheap IT infrastructure in the last decade have resulted in an abundance of data collection at generating stations. There has also been a proliferation of sophisticated tools that use this data to various effectuations ranging from performance monitoring and vibration analysis to combustion optimization. One of the key reasons for the often unsatisfactory performance of these "data-using tools" has been that the information collected by the plants have been riddled with inaccuracies due to a number of reasons as varied as sensor drifts and failures to high dead-bands and exception reporting. Bad data being used as input to these advanced tools, no matter what the level of complexity of them, has resulted in bad decisions.

APR (Advanced Pattern Recognition) software has been commercially available for more than a decade and is used by many top-performing generation companies for advanced detection of impending failures (high impact low probability or HILP type events). Data validation focuses on ensuring that the data collected is accurate and, also provides replacement data where actual data is found to be inaccurate or unreasonable. This paper investigates the advantages gained by using APR software to validate data; showcasing it through real and documented examples. In effect, the premise of this paper is that data validation, often overlooked by companies, can be achieved relatively cheaply and easily using APR software, and this has many beneficial implications for the enterprise that include the reduction of their environmental and maintenance footprints, while significantly boosting productivity and economic performance.
________________________________________

Early Discovery of Failing Equipment and Sensors in Power Plants using Advanced Pattern Recognition
2006 Presented at the ISA POWID/EPRI Conference
Early Discovery of Failing Equipment and Sensors in Power Plants using Advanced Pattern Recognition
Ron Griebenow, President, Performance Consulting Services, Inc.
Marcus Caudill, Vice President, Performance Consulting Services, Inc.
Download

Abstract

This study describes the application of Advanced Pattern Recognition On-line Monitoring in a power plant fleet environment. Advanced Pattern Recognition has been able to successfully and routinely detect incipient equipment failures and sensor failures not typically detected by traditional control systems.

Typical failure identification and alarming is carried out applying high and low limits to the range of a single instrument measurement. This results in a rather wide fixed window of operation. Advanced Pattern Recognition technology has been employed to monitor multiple sensor systems using a moving window to identify normal and abnormal operation. Using a moving window takes into account the current operating conditions of the equipment and allows the early detection of many types of failures. The moving window is defined by a pattern recognition generated expected value and the threshold or normal range about that expected value. These abnormalities are then compared to typical failure signatures to suggest the failure origin. The expected value can also be used to validate the signal and provide a replacement signal value.

Practical applications of advanced pattern recognition to detect plant equipment and sensor failures will be described. The variety of incipient equipment failures, which have been identified in real plant environments, will be categorized. Several failures will be illustrated using case studies. Subtle and difficult to detect failures will be profiled. These will also be illustrated by case studies.

Advanced pattern recognition has been used successfully in plant environments to provide the early detection of equipment and sensor failures.
________________________________________

Abstract

Safety, reliability, availability and efficiency have always been essential in the operation of the electric power generating facilities. In an effort to employ continual improvement in these areas, many power plants have employed advanced computing technologies including neural networks, statistical process monitoring and expert systems. One such system uses advanced pattern recognition to identify any measurement, component or process within the plant that has deviated from "normal" operation. Interpretation of these deviations can provide early warning of equipment failures, saving the unit from catastrophic failures and costly downtime.

Through application of advanced pattern recognition, accurate estimated values for each instrument being monitored are provided and compared to the actual measured value. The difference between the measured and estimated values, referred to as a "residual", is then monitored and compared to statistically determined alarm levels. As the alarm levels are reached, the patterns of residuals are compared to various modes of equipment failure or component degradation. Given the precision of the estimated values, deviations of even a fraction of a percent are clearly identified, often providing days, weeks and even months of warning prior to a catastrophic event. In addition, the deviation of a single measurement without associated variations in other process variables will often result from instrument calibration drift or some other measurement processing failure. Identification of these deviations can be used to optimize instrument calibration programs and ensure accurate data is available to other on-line systems to assure quality results for effective decision-making.

This paper will outline the advanced pattern recognition process and present several case studies demonstrating to accuracy and value of this technology to equipment condition monitoring and fault detection.
________________________________________

Application of Advanced Pattern Recognition for Nuclear Plant Performance Optimization
2004 Presented at the 11th Nuclear Plant Performance Improvement Seminar Proceedings
Ron Griebenow, President, Performance Consulting Services, Inc.
Marcus Caudill, Vice President, Performance Consulting Services, Inc.
Tim Holtan, SmartSignal Corporation
Download

Abstract

In the increasingly competitive electric power generation market, it is critical that all generation resources be utilized in the most cost-effective manner. In particular, it is essential that the unit generating capacity, the efficiency of plant processes, and the effectiveness of maintenance on the equipment be optimized. In an effort to employ continual improvement in these areas, many nuclear plants have employed advanced computing technologies including neural networks, statistical process monitoring and expert systems. One such system uses advanced pattern recognition to identify any measurement, component or process within the plant that has deviated from "normal" operation. Interpretation of these deviations can provide early identification of equipment degradation and help plant analysts identify the root cause of the efficiency losses.

Application of advanced pattern recognition methods can help reduce operation and maintenance costs in a number of ways. First, advanced pattern recognition can differentiate between "normal" variations in equipment operation (e.g., changes caused by environmental variations or operating modes) and changes caused by performance degradation, providing early notification of efficiency losses and critical time to prepare for maintenance activities. Second, the technology can be applied to validation of data used by satiety, optimization, performance monitoring, and control systems to ensure that these systems provide accurate and reliable information. Third advanced patter recognition can accurately identify instruments requiring calibration, that calibration efforts can then be focused on only t hose instruments that need attention, reducing total hours required for instrument maintenance.

This paper will outline the advanced patter recognition process and present several case studies demonstrating the accuracy and value of this technology to equipment condition monitoring and data validation for plant performance improvement.
________________________________________

Precision Data Validation Boosts Process Optimization Benefits
1999 Article in November issue of POWER Engineering magazine
Ron Griebenow, President, Performance Consulting Services, Inc.
Marcus Caudill, Vice President, Performance Consulting Services, Inc.
Duane Hill, Dairyland Power Cooperative
Steve Hoffman, Hoffman Publications, Inc.
Download

Fuel costs are rising, emissions standards are tightening and skilled manpower is always in short supply. To remain competitive, many power producers are turning to plant process optimization - a field that is so "in vogue" today that prospering without it seems doubtful. But as large fossil plants around the world employ expensive consultants to install sophisticated process optimization tools, plant managers are discovering (if they weren't already aware of it ) that precise validation of the data going into these tools is the single most important prerequisite to success.

Precision data validation is the ability to recognize abnormalities in real-time power plant systems. Today, the usefulness of process optimization, as well as other automated systems and technologies for plant monitoring and control, is highly dependent on the reliability of plant instrumentation and the accuracy of the data it provides. But this fact really hits home when a power producer invites a data validation consultant into a plant. Invariably, the contractor identifies instrumentation problems with a variety of systems that supersede the results of various optimization analyses. These problems are reducing generation output, raising the heat rate, and increasing emissions of SO₂, NO_x, and CO₂ - each of which increases operating costs.

________________________________________

Applied Pattern Recognition for Plant Monitoring and Data Validation
1995 Presented at the ISA POWID Conference
Ron Griebenow, President, Performance Consulting Services, Inc.
Elmer Hansen, PhD, Performance Engineer, Performance Consulting Services, Inc.
A. L. Sudduth, P.E., Engineering Consultant, Duke Power Company
Download

Abstract

In an attempt to improve plant performance, reduce operating and maintenance costs, and meet the requirements of new regulation, many utility companies are implementing automated systems and new technologies for plant monitoring and control. However, the usefulness of these systems is dependent upon the reliability of the instrumentation. Erroneous input data can result in misleading performance assessments, inappropriate operator actions, inefficient plant operation, excessive plant emissions, and a host of other undesirable effects. Duke Power Company is applying advanced pattern recognition (APR) techniques to detect and eliminate faulty input data to on-line plant performance and continuous emission monitoring systems. This paper will present the basis for Duke Power's application, outline the pattern recognition methodology, and discuss its application to continuous emission monitoring systems at Duke's Lincoln Combustion Turbine facility.
________________________________________

Similarity Based Regression: Applied Advanced Pattern Recognition for Power Plant Analysis
1994 Presented at the EPRI Heat Rate Improvement Conference
Elmer Hansen, PhD, Performance Engineer, Performance Consulting Services, Inc.
Marcus Caudill, Vice President, Performance Consulting Services, Inc.
Download

Abstract

Substantial amounts of time, money, and manpower are expended in analyzing and controlling power plants. Many of these efforts are compromised due to faulty or missing data values that can result in erroneous performance assessments, inappropriate operator actions, inefficient plant operation, excessive plant emissions, and a host of other undesirable effects. To detect and eliminate bad data values associated with power plant equipment and instrumentation, a computerized analysis technique called Similarity-Based Regression has been developed which utilizes historical plant data to form the basis of a very accurate, fault-tolerant plant model. The method automatically provides substitutes for any faulty or missing data values and readily detects abnormalities in individual sensors as well as plant subsystems. Similarity-Based Regression is currently being adapted to a variety of power plant applications.