By Selmo C. Kuperman, M. Regina Moretti, Julio C. Pinfari, and Edvaldo F. Carneiro
CESP – Companhia Energética de Sà£o Paulo S.A., in Brazil, owns four hydro projects: 3,444-mw Ilha Solteira, 27.5-mw Jaguari, 1,551-mw Jupia (Eng. Souza Dias), and 85-mw Paraibuna. The dams associated with these projects are large concrete, earthfill, or rockfill structures, each more than 30 years old. CESP also owns Paraitinga Dam, which impounds water as part of the Paraibuna project. This earthfill dam also is more than 30 years old.
The 3,444-mw Ilha Solteira hydropower project is one of five locations where owner CESP – Companhia Energética de Sà£o Paulo, Brazil, established “limit values” on the readings of installed geotechnical instrumentation. These values allow CESP to identify alert and emergency situations that require action.
Because about 2,300 of the 3,000 monitoring instruments installed in these five structures still function, CESP has been able to compile an extensive history of structural behavior. CESP’s embankment dams contain inclinometers, settlement meters, Casagrande-type and vibrating wire piezometers, total pressure cells, and weirs. Concrete dam instrumentation includes standpipe and Carlson-type piezometers, rod extensometers, direct and inverted pendulums, Carlson-type and mechanical joint meters, thermometers, extensometers, crack detectors, reinforced and concrete stress meters, long baseline strain meters, and weirs. However, as often is the case with older dams, there are no “limit values” associated with the readings from these instruments. This meant CESP could not determine when readings exceeded acceptable limits and warranted further monitoring or remediation.
To remedy this situation, in 2000 CESP began a project to establish limit values for the readings of all these instruments. The goal was to:
- Determine how each of these instruments relate to structural safety by checking their behavior over the years and their response to variations in the upstream and downstream water levels and temperatures; and
- Establish “alert” conditions so that CESP can take further action when readings exceed calculated reference values or indicate a trend that differs from the pattern established from data gathered over the life of the structures.
CESP used a combination of statistical tools and engineering judgment to establish these values. The resulting methodology was easy to use and made the monitoring process more reliable. It also helped CESP change the frequency of readings from certain instruments, thus reducing monitoring costs by about US$12,000 a year.
Instrumentation already installed at CESP´s structures
In total, CESP’s five dams contain about 3,000 monitoring instruments. Some instruments, such as piezometers, were installed to gather information when doubts existed about the future behavior of the structure. Others, such as settlement meters and rod extensometers, were installed to develop national knowledge on their manufacturing and installation, or analysis of their data. Certain instruments — such as strain gages and pore pressure, creep, crack, and long baseline strain meters in concrete dams — were installed with research objectives. Finally, some instruments (thermometers in concrete dams) were installed to fulfill a temporary function, such as checking the efficiency of specific construction procedures or helping set certain parameters.
To date, about 700 of the instruments have stopped functioning. Some instruments that were installed for research or to advance civil instrumentation techniques stopped functioning during construction or just after the dam was completed. These include strain gages, thermocouples, and creep meters in concrete dams and vibrating wire piezometers and total pressure cells in embankment dams. Problems with humidity caused the strain gages and thermocouples to stop functioning during construction, while faulty construction of the instrument case for the creep meters or lightning strike of the piezometers were the cause of their malfunctions. The total pressure cells stopped functioning just after the dam was completed.
In addition to the instruments that have stopped functioning, some instruments are read less frequently. For example, readings from inclinometers and settlement devices were useful during construction, reservoir filling, and initial operation of the earth dams. But the behavior of the dams is almost stable, meaning this data is not as important today.
Deciding to set limit values
In 2000, CESP hired Themag Engenharia Ltda. of Sao Paulo, Brazil, to perform two vital tasks. The first was to use historical data gathered to assess the structural safety of CESP’s five dams. The second task was to set limit values for the monitoring instrumentation installed in these structures. At that time, only the foundation piezometers installed in concrete dams had limit values set by the design.
To accomplish the first task, Themag analyzed readings from 1986 to 2000 of all instruments installed at these five structures. By comparing these readings to what should be expected, Themag concluded that most of the readings showed coherent values. Only a small amount could be clearly classified as erroneous readings. These readings were discarded.
Once the readings were verified as accurate, Themag used them to determine the structural safety of the dams. Methods used to perform this work involved comparing loads and factors of safety used in the design of the dams with the behavior of all structures throughout the years, visual inspections, and instrumentation analyses. Reports on instrumentation and visual inspections were useful for this task because they covered all aspects of the dams since their construction. This analysis indicated that all dams were stable.
For the second task, Themag developed a methodology using statistical tools to determine reference values for the instrumentation readings. Establishing limit values is a complex task. If the values chosen are not sufficiently correlated to those observed over the years, operators would not be alerted to some alterations in the structural behavior revealed by certain instruments, such as piezometers. In turn, CESP might postpone the actions needed to prevent a bigger problem.
Performing a statistical analysis of instrumentation data
Because CESP´s dams had operated for about 30 years, the utility has an extensive historical series of instrumentation readings that depict the loads acting during those years. This helped establish one basic hypothesis for this work: If values measured by the same instruments continue to vary within a certain range, keeping all other conditions similar to those of the past, the structural behavior will stay within what can be called normal.
Two other hypotheses used were: The measured values follow a normal (or Gaussian) distribution, and the confidence interval must be 70 to 100 percent.
Other basic statistics concepts Themag used to establish limit values included: continuous distribution; variance; Student’s “t” distribution; hypothesis tests and significance F distribution; simple and multiple linear regressions; and coefficient correlation.
To perform the statistical analysis of the instrumentation data, Themag initially developed graphs of the entire historical series of readings for each of the 2,300 instruments. Themag personnel then inspected these graphs, looking for important features, trends, seasonal oscillations, event-related oscillations, and any specific characteristic that could serve as an orientation for the task.
Themag also carried out analyses to determine any correlations between the instrumentation readings and the acting loads. These loads included upstream water level, downstream water level, and the difference between upstream and downstream water levels.
Most of the instruments analyzed showed either stable readings (case 1) or variations that correlated with oscillations in the upstream water level (case 2). (See Figure 1.) By taking into account these two cases, there are two complementary lines of statistical treatment that must be used to define upper and lower reference values for instrumentation.
Figure 1: This curve illustrates the normal distribution of reference readings, with prediction intervals derived from a standard error. With a confidence level of 95 percent, X1 is the lower reference value and X2 is the upper reference value. Readings outside these limits would trigger an alert.
For the first treatment line, used with most of the instrumentation, the mean and standard deviation of the readings are calculated from a large amount of data collected for a long time. In this case, these parameters truly reflect the behavior of the instruments. Hence, they can be directly used to establish the reference values.
The second treatment line was used primarily with piezometers. When the readings show significant variations and high correlation with the upstream water level, it is necessary to link the values to the loads. Thus, the reference values for these instruments are established by following the statistical treatment of the historical data, in agreement with the water level.
It is important to note that, when installed in materials with low permeability, stand-pipe piezometers show a delay in the reading response when compared to the variation of the acting loads (upstream water level). This fact often results in a low coefficient of correlation. In such cases, the delays that implied better correlation coefficients were investigated. If considered significant, these limits were added.
Calculations that yielded values of less than 0.7 for the correlation coefficient “r” (less than a 70 percent confidence interval) were not considered. In addition, correlations whose significance “F” was below an adopted significance level of 5 percent were rejected.
For a confidence level of 95 percent, the prediction interval limits are:
X2 = m + (Pr x Sd)
- Sd is the standard deviation; and
- Pr is a value that depends on the confidence level and on the quantity of data used in the calculations. For more than 120 values, the Pr value according to Student’s “t” is 1.960. For the same confidence level but with 30 to 60 readings, Pr is 2.021.
Applying these statistical concepts, the prediction interval is the one that lays between the upper reference value and lower reference value lines.
As mentioned above, the instruments were analyzed based on one of two cases.
a)â€…Case 1 (reference values based on the mean and standard deviation of readings). Limit values were set by adding to or subtracting from the mean, which was obtained by multiplying the standard deviation by the Student´s coefficient “t” associated with the chosen probability. (See Figure 2.)
Figure 2: Readings from piezometer 10 at the Paraitinga earthfill dam were charted against the reservoir water level. The goal was to establish a correlation between the two values, as well as the margin of error from the correlation.
Using the following equation, the upper reference value is established by adding the standard deviation to the historical mean, while the lower reference value is established by subtracting the standard deviation from the historical mean.
- VR is the upper and lower reference values;
- VM is the historical mean value of the instrument; and
- Sd is the historical standard deviation.
In embankment dams and their foundations, this method was used to set limit values for those instruments that presented a small magnitude of variation and low correlation with the water levels.
b)â€…Case 2 (reference values established from the correlation between upstream water level, instrument reading, and margin of error from the correlation). In this case, the limits guarantee that 95 percent of the historical records fit into the established range, as shown in Figure 3.
Figure 3: Using the correlations between piezometric levels and reservoir levels at Paraitinga earthfill dam charted in Figure 2, Themag calculated upper and lower reference values.
Choosing alert situations
An alert situation occurs when readings lead CESP personnel to conclude that there is a situation of concern within the dam. The following conditions are considered alert situations:
- Any measurement that is outside the reference values interval evaluation for an instrument considered important to the safety of the structure;
- Measurements that are consistently outside the reference values interval, even after normal verification of reading consistency (such as measurement of the same instrument performed by another technician, integrity of the instrument checked, or integrity of the reading device checked);
- Average of at least five readings of an instrument taken during five consecutive weeks remaining near the measurement considered anomalous; and
- Change in the trend of the reading (either an increase or decrease of the measured value). This change generally can be visually noticed from graphs depicting the measured value against time.
An alert situation occurs if the trend has changed, even if the reference values were not exceeded.
Analysis of an alert situation implies instrumentation verification, increase in the reading frequency and its analysis, periodical detailed inspections of the dam, and overall checking of the behavior of all structures.
Because the statistical criteria considers the history of a certain time period, reference values should be reevaluated every five years if significant changes have taken place to the water levels or to other loads acting in the dams. If the loads do not change significantly throughout the years, this reevaluation should take place at least every ten years.
Whenever detected, problems should be solved. The alert condition will no longer be apparent when the problem that triggered the situation ceases and the instrumentation readings return to the reference values interval. This may result from special interventions deemed necessary to solve an eventual problem. Every alert situation should be analyzed immediately so that it will never be necessary to set in motion an emergency situation.
Results and lessons learned
This work provided CESP with cost savings of 42.5 percent, or 100 man-hours in the time spent to read instruments. For example, before this work, settlement meters and inclinometers in embankment dams were read once a month. Personnel now take readings from these instruments only once a year. For concrete dams, reading frequency for all thermometers, extensometers, and stress meters changed from once a month to once a year.
Through this project, CESP learned several valuable lessons. These lessons can help other dam owners seeking to implement a similar program.
- It is vital that instruments deemed important to the safety evaluation of a dam are read and interpreted according to established routines;
- Alert situations should be employed as a criteria for the analysis of results of the instrumentation readings and for the evaluation of dam safety. They are reached when the reading reference values are surpassed or when the measurements show the beginning of a trend that is outside the established pattern throughout time; and
- The main purpose for reading reference values is to guide the behavior analysis process of the instruments. They allow personnel to check whether the value is adequate while the instrument measurements are being taken. They have been useful for CESP during the preliminary analysis about the structural behavior of the dam.
Selmo Kuperman, PhD, consulting engineer with DESEK Ltda., was manager and instrumentation engineer for the program to establish “limit values” for geotechnical instruments. M. Regina Moretti, civil engineer with THEMAG Engenharia e Gerenciamento S/C Ltda., was geotechnical engineer for the work. Julio Pinfari is manager of the civil engineering division of civil maintenance and dam safety with CESP – Companhia Energética de Sà£o Paulo S.A. Edvaldo Carneiro is supervisor of the civil engineering division of civil maintenance and dam safety with CESP. Messrs. Pinfari and Carneiro provided the data needed to set the limit values and participated in all decisions regarding the project.
Dr. Kuperman may be reached at DESEK Ltda., Av. Nove de Julho 3229, cj.207, Sao Paulo, SP 01407-000 Brazil; (55) 3057-1514; E-mail: firstname.lastname@example.org. Ms. Moretti may be reached at Themag Rua Bela Cintra 986, 13th, Sao Paulo, SP 01415-906 Brazil; (55) 3100-1443; E-mail: email@example.com. Messrs. Pinfari and Carneiro may be reached at CESP, Avenida Nossa Senhora do Sabara, no 5312, Sao Paulo, SP 04447-011 Brazil; (55) 5613-3779 (Pinfari) or (55) 5613-3781 (Carneiro); E-mail: firstname.lastname@example.org or email@example.com.
The authors thank CESP and Themag for permission to publish this paper. Help from Sergio Cifu, Giacomo Re, Tarcisio Barreto Celestino, Kenia D. Kimura, Sergio Luiz Guimaraes Rossetto, and Ruitter Prada Reigada in developing the criteria is gratefully acknowledged.
This article has been evaluated and edited in accordance with reviews conducted by two or more professionals who have relevant expertise. These peer reviewers judge manuscripts for technical accuracy, usefulness, and overall importance within the hydroelectric industry.