Structural Condition Models for Sewer Pipeline
Fazal Chughtai1and Tarek Zayed2 1
Student Member ASCE, Graduate Student, Department of Building, Civil and Environmental Engineering, Concordia University, Montreal, PQ, Canada; Email: [email protected]
2
Assistant Professor, Department of Building, Civil and Environmental Engineering, Concordia University, Montreal, PQ, Canada; Email:[email protected]
Abstract
Proper management of sewer pipeline networks involves fulfillment of many technical requirements under economic constraints. Therefore, cost effective solutions are required to assist municipal engineers in prioritizing maintenance and rehabilitation needs. This demands a systematic approach to condition assessment of rapidly deteriorating sewers. Performance evaluation of sewers through random inspections is expensive. Therefore, there is an urgent need to develop a proactive sewer pipeline condition prediction methodology. This paper presents a method for assessing a sewer’s structural condition by utilizing general pipeline inventory data. Based on historic condition assessment data, condition prediction models for sewers are developed using multiple regression technique. The final outcome of these models produces most likely condition rating of pipes, which will assist municipal agencies in prioritizing pipe inspection and rehabilitation to critical sewers.
Introduction
The condition of underground sewer infrastructure across North America has been deteriorating day by day. The maintenance and rehabilitation of aging sewers have become an overburden in terms of budget allocation and investment planning for municipalities. Multiple objectives may exist for planning budget allocation that could be dependent upon certain constraints and available resources. This makes the task of planning, prioritizing and allocating funds a complex exercise (Ruwanpura et al. 2004).
Investment in sewer rehabilitation must be based on inspection and evaluation of sewer condition. However, random inspection of sewers is extremely expensive. Therefore, due to budget constraints, only 22% of Canadian municipalities have a complete condition assessment program (Rahman et al. 2004). Thus, it is important to prioritize inspections to those sewers which are more vulnerable to deterioration phenomena and have higher risk of collapse for proactive sewer rehabilitation planning.
There are two main avenues of improvement in sewer rehabilitation planning (Ariaratnam et al 2001):
1) Collection and storage of adequate inspection information regarding current condition of sewer system
2) Ability to predict sewer deficiency prior to failure in order to facilitate timely sewer inspection and repair
Therefore, predicting structural condition of sewers should be the first consideration of municipal managers in order to prioritize detailed inspections. In this context, current paper describes the development of an improved methodology for analyzing and interpreting historical sewer data. The methodology involves the use of this data in developing multiple regression models to predict the existing condition rating of sewers. The application of these multiple regression models would provide decision makers with a means to prioritize inspections of sewers, which have higher risk of failure.
Previous Decision Support Tools for Sewer Condition Assessment
There are numerous documented studies that focus on various aspects of drainage systems including different methodological approaches to predict the condition of drainage pipes (Ruwanpura et al. 2004). In terms of objectives, these studies can be categorized into two classes. The first approach is to help municipal engineers in preventing inadequate sewer inspections through developing automated condition assessment techniques. The second strategy is to develop proactive tools for prioritizing inspections to critical sewers.
There are many examples of studies for developing automated systems for the interpretation of sewer inspection data. Moselhi et al (2000) described image analysis and pattern recognition techniques of sewer inspection, based on neural network analysis of digitized video images. The neural network analysis technique was found helpful in identifying four categories of sewer defects: cracks, joint displacements, reduction of cross-sectional area, and spalling. Chae at al (2001) developed an automated sewer inspection data interpretation system. Artificial neural networks were used to recognize various types of defects in sewers through optical data obtained from inspection with Sewer Scanner and Evaluation Technology (SSET). Sinha et al (2006) presented an algorithm for the automated analysis of scanned underground pipe images. The algorithm consisted of image pre-processing followed by a sequence of morphological operations for the classification of different sewer defects: cracks, holes, joints, laterals, and collapsed surfaces.
The other approach is to predict a sewer’s existing condition prior to its detailed inspection for selective, cost effective sewer inspection. Hasegawa et al. (1999) developed a method for condition prediction of sewers on the knowledge of pipe material, length, diameter and other characteristics. However, it was concluded that the method could not evaluate sewer’s condition effectively. Ariaratnam et al (2001) developed logistic regression model for condition evaluation of sewers. The model was developed through historical data based upon factors; such as, pipe age, diameter, material, waste type and depth. Another approach for condition assessment of large
sewers was developed by assessing the impact of different factors; such as location, size, burial depth, functionality etc., on sewers (McDonald et al 2001). Similarly, Baur et al (2002) developed a methodology of forecasting condition of sewers by using transition curves. These transition curves were developed through the historical condition assessment data. Sewers characteristics; such as, material, period of construction, location were used to define the existing condition of sewers for scheduling detailed inspection. Yan et al (2003) proposed a fuzzy set theory based approach for a pipe’s condition assessment. Various linguistic factors: soil condition, surroundings, etc, were transformed through fuzzy theory into numerical format for assessing their impacts on pipes. Ruwanpura et al (2004) used rule based simulation methodology to predict condition rating of sewers. The model predicted the condition rating of pipe based on age, material and length of pipe. Najafi et al (2005) developed an artificial neural network model for predicting the condition of sewers based on historical data.
The above mentioned approaches tend to predict existing condition of sewers for prioritizing detailed inspections. This paper suggests a simple and easy to use approach towards condition prediction of sewers. The paper describes the development of multiple regression models in this regard. These models will provide adequate knowledge about condition of sewers in order for municipal agencies to optimize cost of sewer inspection.
Sewer Pipe Deterioration
Although pipelines are designed for a particular lifespan under standard operating condition, their deterioration never follows a set pattern (Najafi et al 2005). Pipe deterioration is a very complex process and related to various pipe characteristics such as pipe material, period of construction, location, diameter and gradient (Yan et al 2003). In general, these pipe characteristics or factors can be divided into three categories; physical, operational and environmental. Table 1 shows the subdivision of these factors into further categories. It also explains how these factors contribute in pipeline deterioration phenomena.
Physical factors comprise general pipe characteristics such as length, diameter etc. While operational factors deal with flow performance in which they adapt operational and maintenance strategies. The third category is related to certain environmental factors directly influencing a pipe’s criticality and deterioration. These factors include type of surrounding soil, traffic volume above pipe, etc.
Data Collection and Variables Selection
The effectiveness of condition rating regression model depends upon the quality of collected data and selection of predictors. Data are collected from two municipalities; Pierrefonds (Quebec, Canada) and Niagara Falls (Ontario, Canada). The collected data included general pipeline inventory record, AutoCAD drawings and CCTV inspection reports. One of the major problems in the preliminary analysis of data is
that both municipalities have adapted different sewer pipeline condition grading systems. Data from Niagara Falls consist of WRc (Water Research Centre UK) classification system, while the other data set is based on CERIU (Centre for Expertise and Research on Infrastructures in Urban Areas, Canada) classification system. As WRc classification system is known as the “Embryo Codes” for world wide sewer rehabilitation industry (Thornhill et. al 2005), all data from Pierrefonds are converted into WRc classification system for generalizing the model building approach.
Table 1: Factors that Contribute to Sewer Pipeline Deterioration
Factors Explanation
Pipe Length Pipe in longer length and having greater length to diameterratio are more likely to suffer from bending stresses Pipe Diameter Small diameter pipes are more susceptible to beam failure
Pipe Material Pipes manufactured with different materials show different
failure patrons.
Age More probability of collapse for aged pipes
Average Depth
If the depth is very low, the pipe is susceptible to surface live load. If depth is high, the pipe is susceptible to overburden. Moderate depths increase the life of sewers
P h y si c a l
Pipe Gradient Steeper slopes of pipe cause high flow velocity which
increases erosion phenomena
Infiltration/ Exfiltration
Infiltration and exfiltration wash soil particles which reduces the soil support along the pipe
O p e r a ti o n a l M & R Strategies
Good maintenance and repair strategies increase the service life of sewers
Type of waste Different types of waste react with different pipe materials in
a different manner causing pipe erosion
Type of Soil Different types of soils provide side supports to pipesaccording to their own physical and chemical properties Bedding
Conditions
The chance of pipe failure increases with improper bedding condition of pipes
Frost Factor The load on buried sewers increases due to additional frost
load in winter
Other Utilities Proximity of other underground installations increases thecriticality of a sewer
E n v ir o n m e n ta l
Traffic Volume The bending stresses in the pipe increase with the increase in
live load above pipe
The next step was data sorting for selection of input variables. These variables which have maximum historical information are selected for further model development.
The predictors chosen are pipe material, pipe material class, diameter, length, age, depth, bedding material class and street categories. The information collected regarding bedding material class and street category is described in a generalize manner in order to facilitate regression model application.
Table 2: Bedding Material Classes as per BRE and OPSD Standards and their transformation weights for model development (Adapted from Perkins, 1974 and Zhao et al 2001)
Bedding Factor Bf Bedding
Class Description BRE OPSD
Model Input Weights Reinforced Concrete Cradle or Arch 3.4
A
Plain Concrete Cradle or Arch 2.6 2.8 4
B Well Compacted Granular Material 1.9 1.9 3
C Well Compacted Backfill 1.5 1.5 2
D Flat Sub Grade 1.1 -- 1
Others Cement Stabilized Material 2.6 to 3.4 -- ----Five different types of bedding material have been specified by Building Research Establishment (BRE) UK, which are also acceptable in USA (Perkins, 1975). In Ontario, Canada, OPSD (Ontario Provincial Standard Drawings) defines four classes of bedding material (Zhao et al 2001). These classes have been defined on the bases of bedding factor Bf. In general, the bedding factor Bfis defined as (Zhao et al 2001):
eb S
W f
B = --- Equation 1
Where, W is calculated external load and ebS is 3 – edge bearing strength.
The bedding factors for specific classes in both classification systems have been shown in Table 2 where the input weights are also shown. The weights have been allotted according to material class.
Not only is the bedding factor redefined, but also is the average annual daily traffic (AADT) data. The AADT data depends upon location of streets and other factors. Therefore, instead of using the locally available AADT data (as shown in table 3), the local streets are categorized according to American Society of Civil Engineers (ASCE) Classification. These classifications along with their assigned input weights for model development are shown in Table 3. The AADT data is based on Niagara Fall’s classification.
Design and Diagnostics of Structural Condition Models
In linear regression models, a linear relationship is assumed between the response variable (Y) and the several independent variables (X1, X2, X3...). The major point in these models is that the combined effect of all variables on the dependent variable is investigated rather than individual relations between dependent and independent
variables (Dikmen et al 2005). However, linear regression model includes not only first – order models in predictor variables but more complex models as well. Consequently, model with transformed variables or with different interaction terms should be considered as linear regression models due to their respective linear parameters (Kutner et al 2005). Therefore, different consideration for different functional forms of predictors in regression model is the first key step in model development.
Table 3: Transformation of Traffic Data into ASCE 1990 Urban Street Classification System and Model Input Weights
ASCE Classification Description Approximate AADT (Niagara Falls) Input Weight for Model 1 Arterial 10,000 - 12,500 4 2 Collector 7,500 - 10,000 3 3 Sub-collector 5,000 - 7,500 2 4 Access < 5,000 1
Initially, in order to check possible interactions and multicolinearities among variables, matrix plots are developed for the input data. When two predictive variables in a regression model are highly correlated, they both convey essentially the same information. Therefore, their interactions and relationships among themselves have to be examined carefully. In the process, if matrix plot shows some possible interactions, the decision to include or exclude variables from the model is made through best subset analysis. Figure 1 shows an example of the best subset analysis for one trial.
In Figure 1, it is clear that the model with higher R2, R2-adjusted, lower S, and closer Cpto number of variables is the most appropriate model. Where, S is the standard deviation of residuals. In this case, the variable “pipe depth” should be excluded from the model (Cp = 6.0 & S = 0.83087). After diagnosing correlation, the developed models are checked for their statistical validity. The main diagnostics in this regard are R2, P(F), P(t), residual diagnostics, and LOF. The R2 (co-efficient of multiple determination) measures the proportional variation in structural condition explained by sewer’s attributes; age, diameter, material, length etc. The results shown in Table 4 illustrate that 72% to 82% of the total variability in structural condition can be explained through the developed regression equations. The R2-adjusted accounts for the number of predictors in the model. Both values indicate that the model fits the data well.
To determine P(F) for the whole model, a hypothesis test is carried out. The null hypothesis (H0) assumes that all regression coefficients, 0, 1… p-1 are zero i.e. 0= 1= p-1= 0. The alternate hypothesis (Ha) assumes that not all of them equal to zero. Based on the Minitab’s output the p-values for the test are 0.000 for all chosen models (Table 4). This means that null hypothesis is rejected. Similarly, to determine the validity of regression coefficient individually, “t-tests” are performed separately for the 0, 1… p-1. In case of 0, the null hypothesis (H0) of t-test assumes that 0=
0; while alternative hypothesis (Ha) assumes that 0 0. Similarly, the other null hypothesis assumes that 1= 0 and vise versa. The results of these tests, for all the three models, shown in table 4, indicate that the p-value for intercept is 0.000, 0.041 & 0.003, respectively. As a result, alternative hypothesis is accepted. Note that for performing F and t tests, the confidence interval is assumed to be 0.05; that means that null hypothesis can be accepted if the p-value is equal to or greater than 0.05.
Figure 1:An Example of Best Subset Analysis for a Trial Model
Similar procedure is performed to check the validity of other regression coefficients associated to each predictor in all regression models. The overall results of t- test are found satisfactory. Some of the t-test results shown in the Table 4 have p values greater than (0.05). This indicates that there could be a weak evidence of null hypothesis for that particular coefficient. However, due to large number of predictors in models and due to satisfactory results of other statistical diagnostics, these results are concluded as acceptable.
The next step is to check the residual diagnostics. Figure 2 shows an example of residual plots for a trial model. The normal probability plot shows that there could be a possibility of outliers in the data. After a logically based reexamination of data, it is concluded that the points are not outliers and these scenarios could exist. Therefore, the possibilities of outliers are rejected.
Figure 2 also shows the fitted value plot for the model under consideration. In ideal scenario, constant data would be distributed evenly across the plot. That would show the consistent variance across the fitted value range. However, figure 2 shows diagonal bands across the centre line. The reasons for these types of results could be due to:
Important variable(s) might be omitted from the model (Kutner et al 2005) Data variability issues: data composed of integer variables (Anderson et al 2005)
The careful examination of data in hand shows that both the abovementioned possibilities exist in this case. Some of the important variables which could have a strong effect on the existing pipe condition could be missing. For example, type of soil, maintenance and repair history, infiltration etc are important parameters, which affects the existing pipe conditions directly. As information regarding these kinds of parameters is not available; studying the effect of these parameters on the pipe condition is recommended for future research. Data variability could be explained in the case under consideration as the model has integer predictors; concrete class, bedding class factor and street categories. Therefore, the discrete values of these predictors could cause the problem of unequal variance. Consequently, the results are concluded as satisfactory.
Table 4:Summary of Statistical Test Results for Selected Models
The developed models are further investigated through statistical measures such as Durbin-Watson test for auto-correlation and lack of fit (LOF). The Durbin-Watson test considers the null hypothesis (H0) that there would not be any auto correlation among predictors. The alternative hypothesis (Ha) considers that there is a significance of auto correlation among the predictors. The results shown in Table 4 indicate that Ha should be rejected in case of PVC model, and the test is inconclusive in case of asbestos cement model. The results of concrete model indicate that there could be an evidence of Ha. However, this test is more conclusive for maximum of five predictors; so the results shown for concrete pipe model are not accurate, because in this case the predictors are 6.
The results of LOF test indicate that there are not much replications available to perform the routine pure error test for concrete and PVC models. Therefore, an approximate lack of fit data subsetting test, developed by Minitab® statistical software, is performed on these models. The null hypothesis H0is that model fits the data, and alternative hypothesis Ha is that model does not fit data. The decision
P (t) P (F) Lack of Fit Model R2 (%) R2 (Adj.) (%) P (F)
0
1
2
3
4
5
6
Durbin-Watson Statistics P u r e E r r o r D a ta S u b -se tt in g Concrete Pipes 72.7 70.5 0.000 0.000 0.064 0.000 0.010 0.000 0.014 0.000 0..95 D<dL - 0.0 49 Asbestos Cement Pipes 82.4 78.3 0.000 0.041 0.001 0.034 0.093 0.085 -- --1.43 dL D dU 0.8 74 0.1 PVC Pipes 81.8 78.6 0.000 0.003 0.000 0.008 0.021 0.000 0.000 --1.82 D > dU -- 0.0 56criterion in case of data subsetting test is that p-value should be equal or greater to 0.01 for an ideal fit model i.e. for the significance of null hypothesis. Table 4 shows a significance of Ha; however, this test is an approximation of pure error test and these models were giving satisfactory results for other statistical diagnostics. Therefore, the models are accepted on the basis of their overall performance of all necessary statistical diagnostics. In case of asbestos cement model, the lack of fit test results for pure error and data subsetting tests are found satisfactory.
Figure 2:Normal Probability of Residuals and Residual vs. Fitted Value Plots for a Trial Model
After all statistical diagnostics the three models selected for validation are as follows: Concrete Pipe Structural Condition Prediction Model
Factor Bedding Factor Bedding Depth Log Class Concrete Age Log Depth Log Category Street e Length Diameter Log Grade Condition Structural _ 1 75 . 5 _ 10 92 . 6 _ 10 6 . 1 10 22 . 3 _ 00681 . 0 10 592 . 0 94 . 3 _ _ 1 + + = --- Equation 2 Asbestos Cement Pipe Structural Condition Prediction Model
1 . 0 8 . 14 _ _ 742 . 0 207 . 0 10 542 9 . 20 2 ) _ _ ( Diameter Class Cement Asbestos Age Length Depth Log Grade Condition Structural = + + --- Equation 3 PVC Pipe Structural Condition Prediction Model
4 ) ( 3 . 0 ) ( 000013 . 0 _ 0405 . 0 _ 0302 . 0 01 . 0 89 . 1 00642 . 0 25 . 2 _ _ ) 1 . 0 ( Depth Diameter Category Street Factor Bedding Length Age Grade Condition Structural = --- Equation 4 The values of structural condition grades are according to WRc classification system. The structural condition grading varies from 1 to 5 as per WRc protocols; where 1 is
for a pipe in excellent condition and grade 5 means that collapse for the pipe is imminent.
Conclusions:
A methodology for predicting a sewer’s structural condition information through the use of historical data is proposed. To assess and predict structural condition of existing buried sewers, multiple regression technique is used. Three different regression models are designed for three different sewer pipe materials: concrete, asbestos cement, and PVC. Various forms of variables are experimented during the design procedure and the best possible scenario is selected for further validation. The selected models are validated through all possible measures to ensure their appropriateness.
It is observed that the selected predictor variables for condition rating model are not enough to completely explain the variation in structural condition of sewers. Therefore, it is recommended that future research should be performed in expanding the model for other pipe attributes, which contributes to sewer’s deterioration. The influence of factors such as infiltration, soil condition, and maintenance and repair history, on a sewer’s structural condition should be thoroughly investigated. It is further recommended that the model should also be expanded to include other sewer pipe materials; such as clay and bricks sewers etc., to facilitate municipal managers. It is concluded that the developed technique will assist decision makers in scheduling and prioritizing sewer inspection. Thus, this technique will be helpful in minimizing the cost of random sewer inspection.
References:
American Society of Civil Engineers (ASCE), (1990), “Residential Streets”, Second Edition, ASCE, National Association of Home Builders of the United states, and the Urban Land Institute (ULI) Publication
Anderson, M & Davenport, N, (2005), “A Rural Transit Asset Management System”, University Transportation Centre of Alabama Report 04401, USA
Ariaratnam, S, El-Assaly, A & Yang, Y, (2001), “Assessment of Infrastructure
Inspection Needs using Logistic Models”, ASCE Journal of Infrastructure Systems, Volume 7, No 4, December 2001
Barqawi, H, (2006), “Condition Rating Models for Underground Infrastructure:
Sustainable Water Mains”, Master’s Thesis, Concordia University, Montreal Baur, R & Herz, R, (2002), “Selective Inspection Planning with Ageing Forecast for
Sewer Types”, International Water Association (IWA) Journal of Water Science and Technology, Volume 46, No 6-7, page 389-396
Chae, M and Abraham, M, (2001), “Neuro-Fuzzy Approaches for Sanitary Sewer
Pipeline Condition Assessment”, ASCE Journal of Computing in Civil Engineering, Volume 15, Number 1, January, 2001
Dikmen, I, Birgonul, M & Kiziltas, S, (2005), “Prediction of Organizational
Effectiveness in Construction Companies”, ASCE Journal of Construction Engineering and Management, Volume 131, Number 2
Hasegawa, K, Wada, Y & Miura, H, (1999), “New Assessment System for
Premeditated Management and Maintenance of Sewer Pipe Networks”, Proceedings of 8thInternational Conference on Urban Storm Drainage, Page 586-593, Sydney, Australia
Kutner, M, Nachtsheim, C, Neter, J & Li, W, (2005), “Applied Linear Statistical
Models”, Fifth Edition, McGraw-Hill Inc, USA
McDonald, S & Zhao, J, (2001), “Condition Assessment and Rehabilitation of Large
Sewers”, Proceedings of International Conference on Underground Infrastructure Research, page 361-369, Waterloo, Canada
Moselhi, O and Shehab-Eldeen, T, (2000), “Classification of Defects in Sewer Pipes
using Neural Networks”, ASCE Journal of Infrastructures System, Volume 06, Number 03, September, 2000
Najafi M & Kulandaivel G, (2005), “Pipeline Condition Prediction Using Neural
Network Models”, Proceedings of the American Society of Civil Engineers (ASCE) International Pipeline Conference, USA
Rahman, S & Vanier D, (2004), “An Evaluation of Condition Assessment Protocols
for Sewer Management”, National Research Council of Canada Research Report Number B-5123.6
Ruwanpura J, Ariaratnam, S, and El-Assaly, A, (2004), “Prediction Models for Sewer
Infrastructure Utilizing Rule-Based Simulation”, Journal of Civil Engineering and Environmental Systems, volume 21, No 3, Page 169-185 Thornhill, R & Wildbore, P, (2005), “Sewer Defect Codes: Origin and Destination”,
U-Tech Underground Construction Paper, April, 2005
Perkins P, (1974), “Concrete Pipes and Pipelines for sewerage and Water Supply”, Cement and Concrete association, London, UK
Sinha, S and Fieguth, P, (2006), “Segmentation of Buried Concrete Pipe Images”,
Journal of Automation in Construction, Volume 15, Pages 47-57
Yan J & Vairavamoorthy, K, (2003), “Fuzzy Approach for Pipe Condition
Assessment”, Proceedings of the American Society of Civil Engineers (ASCE) International Pipeline Conference, USA
Zhao, J. Q & Daigle, L, (2001), “SIDD Pipe bedding and Ontario Provincial
Standards”, Proceedings of International Conference on Underground Infrastructure Research, Kitchener, Ontario, pp 143 – 152
