When facing pipeline replacement decisions, asset managers face a dilemma. Factual information about the condition of the pipelines allows better replacement decisions and capital efficiency gains. On the other hand, gathering this information is costly. Performing a cost-benefit analysis is also challenging, as the benefits are difficult to project. This paper models the financial impact of pipeline condition assessment by considering the financial risk associated with decision making errors. An economic optimization equation using the model is presented. This equation yields the Economic Assessment Level: the amount of condition information needed to minimize the total combined spending on information gathering and incorrect decisions. Case studies and examples of the impact of different levels of information gathering are presented. The results of these programs are compared with the predictions of the model, illustrating how the calculations can be used to improve capital improvement program efficiency real world situations.
FINANCIAL RISK OF DECISION ERRORS
Financial analysis drives many decisions in an asset management program. In cases where decisions need to be made under uncertainty, the concept of quantified risk is crucial. Risk can be measured in dollars, as the product of the probability of an event and the consequences of the event.
Consider this variant of the ‘three shell game’. There are three upside down shells, one of which hides a pea. You must pick one shell, and if it doesn't have the pea underneath it then you lose $5. Your risk is $3.33 per play; it is found by multiplying the probability of a wrong pick (67%) by the consequences per wrong pick ($5).
Running a pipeline replacement program is similar, but on a larger scale. Among the many kilometers of mains you manage, some fractions are in poor condition, and others have substantial useful life remaining. You need to choose which mains to replace. If you choose a main that is still in good condition, you lose the remaining useful life of that main and the value which that represents.
For the purposes of evaluating a pipeline replacement program, mains can be placed into two groups: ‘good’ mains having a remaining useful life of ½ what a new main would have, and ‘bad’ mains which have less than this. The remaining value of a good main is then ½ the cost of replacing that main with a new one. The probability that a main is good will vary program by program, and requires expert judgement.
TOTAL COST CALCULATIONS
In the shell game example, you had no information about which shell hid the pea. Fortunately, this is not the case in a pipe replacement program. You will generally know a main's demographic information such as installation date, pipe diameter, and material. You may know its failure history, traffic loading, bedding and backfill, and perhaps other information as well. These can help you make a better decision, particularly when combined with a review of the literature or desktop modelling software to help understand how they typically impact a main's behaviour. This is a common approach for pipeline replacement programs, and will be our starting point.
More information can be acquired; however, it comes with a cost. Coming back to the shell game, what if, for a fee of $1, you were allowed to peek under one shell before choosing? This reduces your odds of picking the wrong shell to 33%, thereby reducing your risk to $1.67 per play. Adding in the $1 cost of this information, your total cost is now $2.67 per play, which is a better deal than the $3.33 per play with no information. So the $1 cost to peek before choosing is a good investment.
The Decision Error Risk changes as the Data Gathering Cost changes. With more data on pipe conditions, water companies are able to increase the portion of mains selected for replacement which are in bad condition. The Decision Error Percentage will never drop to zero, but will decline as more data becomes available.
The Data Gathering Cost can be broken down to two components: the Cost of Preparation and the Cost of Inspections. The Cost of Inspections covers what is directly spent on the inspection itself. The Cost of Preparation includes all ancillary costs, such as excavation, selection and installation of access points, or draining the pipeline.
This Total Cost includes only the ‘necessary evils’ in the pipeline replacement program. Funds spent correctly replacing bad mains need not be considered. The money spent is replaced with an asset of equal value, so the net cost is zero.
Calculating the lowest cost option
With the Total Cost clearly expressed, the goal now is to find the amount of data gathering which results in the lowest total cost. Generally speaking, more data gathering will result in increased costs for Preparation and Inspection, and a decrease in the Decision Error %. The Preparation and Inspection costs will continue to rise at a steady rate as more data is gathered. The improvements in the Decision Error % is expected to show a diminishing return as more data is gathered. The more data you already have, the fewer errors you will make. The fewer errors you are making, the less opportunity there is to improve your decisions based on new data.
When a handful of different approaches are available for data gathering, individual estimates can be made of the inspection and preparation cost, and the resulting total costs can be easily compared. See the hypothetical example below:
The Assessment Costs, Error Costs, and Total Costs are plotted in Figure 1 below. The lowest point on the Total Costs line shows the amount of testing which provides the lowest Total Cost. This can be described as the Economic Assessment Level. A utility operating at the Economic Assessment Level is spending as little as possible on data gathering and decision errors combined.
Condition assessment methods which use samples to assess the overall condition of a main offer a special case. By varying the number of samples per km (or the % of the main assessed, in the case of acoustic wall thickness), the situation changes from a handful of discrete options to a wide range of choices. Plotting these can give a smooth curve instead of the jagged chart shown in Figure 1. The acoustic wall thickness testing technology used for the case studies in this paper is such a method.
The case studies in this paper employ the Acoustic Propagation Velocity Method (APVM) for measuring pipeline wall thickness. APVM measures the average minimum wall thickness over a length of pipe. APVM uses two acoustic sensors to measure the velocity of an acoustic signal in a water pipe. Acoustic sensors are installed between 60–150 m apart bracketing a length of pipe. A noise is induced into the pipe, outside the tested length of main to create a known acoustic pressure wave. The speed of the acoustic pressure wave through the pipe is then measured with the acoustic sensors. A typical test set up is shown in Figure 2.
The wave speed in the pipe is a function of the wave speed in an infinite body of fluid and the pipe wall thickness. Changes in wave speed between lengths of pipe tested indicate changes in pipe wall thickness along the test segment. Other site variables such as temperature and fluid bulk modulus, that impact acoustic velocity, are measured on site and normalised in the collected data set. The APVM provides an indication of the general loss of structural strength over the test interval. This is a valuable screening tool to identify lengths pipeline with high levels of structural degradation (Beuken et al. 2014). Water utilities have used APVM to prioritise pipeline replacement efforts (Sinha 2014). Further, APVM has been used to target pipeline rehabilitation efforts by cleaning and lining structurally sound but hydraulically deficient pipelines, and using structural liners or pipe replacement for pipes that were both hydraulically and structurally deficient (Wolan et al. 2018).
While testing is usually done over entire pipelines, the method is also used as a sampling tool. When sampling, only a fraction of the main's length is actually tested, and results are extrapolated to evaluate the condition of sections of similar mains. Sampling targeted mains allows the amount of inspections to be tailored to pinpoint the Economic Assessment Level.
A calculator has been developed, based on the Economic Assessment Level model, to identify the optimal investment in pipe structural testing. The cost of preparations for pipeline inspection and cost of inspections are built in as a function of the percentage of the main to be tested. Estimates of the expected Decision Error % at different levels of coverage using APVM have been solicited from 10 experienced engineers in four organizations. An equation was fitted to these results to provide a reasonable estimate of the Decision Error % at any level of inspections. Levels of over 100% can be used in the equation, and represent testing portions of the main with a second complementary technology to supplement the results. An example of the calculator is shown in Figure 3 for a situation similar to the example used in Figure 1. In this example, the Economic Assessment Level falls at around 60% coverage with APVM.
CASE STUDY: ANGLIAN WATER
Anglian Water serves the second largest geographic area in the United Kingdom, supplying water to 4.3 million customers and water recycling services to 5.5 million. Water is supplied through a mains network of over 35,405 km.
The pipe replacement program is based on a risk model for likelihood of breaks taking into account a number of factors such as; historic break rates, customer minutes lost and risk to water quality. Over a five-year asset management period (AMP) Anglian Water are due to replace 402 km of water mains. In a drive to better target investment Anglian Water are investigating condition assessment methods and APVM was chosen as part of the pilot project.
Following some small scale pilot tests to explore the technology, AVPM was deployed on a live scheme which was being delivered through a third party as part of Anglian Water's @One Alliance. The scheme comprised of two parts;
Rehabilitation of 7.2 km of an existing line between a reservoir and water treatment works
Installation of an additional 7.2 km, 457 mm line between the two sites
The scheme was needed to support growth in the area; the population of Anglian Water's region is expected to grow by close to 20% by 2035 against a 2010 baseline. The existing system could not support the forecasted demand. Additionally, there was a higher than average number of breaks on the existing pipe due to corrosive soil conditions. This meant the main had to either be rehabilitated or replaced. Due to Anglian Water's commitment to reduce its carbon footprint, rehabilitation was the selected route.
During the course of the scheme Anglian Water identified a 198 m section which potentially could be removed from the program if a good mains condition could be proven. The section ran under an environmental protection area where it was found to suddenly sink and rise in level on either side. This added to the complexity of the work and the risk of negative environmental impact. Anglian Water needed to ensure that the right decision was being made and so deployed AVPM to ascertain the remaining average wall thickness and the remaining service life (RSL) of the pipe.
The test was carried out in less than a day, splitting the 198 m section into two parts. Existing fittings could be utilised for the assessment. The RSL value allowed Anglian Water to compare to their standard design life before making an informed decision on what work to carry out on the section.
Anglian Water work to a required design life of 50 years when installing new assets. The results showed the section to have a RSL of equal or close to that value and so a review of the scheme was held to decide on the most appropriate solution. Anglian Water uses a risk based scoring to evaluate investment decisions during the course of a capital delivery scheme. When taking into account the updated likelihood of failure due to the pipe condition assessment results, the revised RSL indicated that rehabilitation of the 198 m section of pipe could be deferred. Instead, valve arrangements were installed outside of the protected area to allow swift access if needed to repair breaks in future.
As a result of the AVPM assessment Anglian Water was able to make the most appropriate investment decision. This led to a cost saving of over S$150,000 along with the added benefits of reducing its environmental impact, carbon footprint and impact on local customers; reducing the estimated time on site from several months to just four weeks.
The calculator in Figure 4 shows that Anglian Water matched the economic assessment level by choosing to assess 100% of the length. While the expected net savings shown in the calculator are a small figure, this represents the probability-weighted average of two scenarios: either the assessment reveals that the main does not require rehabilitation (resulting in substantial savings), or it indicates the main does in fact require rehabilitation (resulting in a net cost increase). The actual results were the former, allowing Anglian Water to achieve a substantial savings.
CASE STUDY: BRABANT WATER
Brabant Water is a large drinking water company in the south of the Netherlands. It supplies 1.2 million customers from 30 water treatment plants via an 18,507 km network.
Data from interruptions and condition assessments is needed to support the annual renewal program of 299 km pipework (diameter between 38 and 203 mm). If opportunities for network renewal arise due to the planned operations of third parties such as councils or gas suppliers, APVM measurements can be used if these areas should be included in the Brabant Water renewal program. In this way Brabant Water can combine replacements with third party infrastructure activities to reduce disruption for customers and the surroundings.
The APVM measurements can be carried out in areas where third parties are working. The network in these areas is often characterised by the same material (asbestos cement or cast iron), the same year of construction, and similar ground conditions, pressure conditions and historical water quality. Brabant Water expects the same condition results in such networks, so less measurements than usual are needed.
In an area where potentially 2.9 km network must be replaced, two locations are selected where a 99.1 m APVM measurement is possible. This is 7% of the total length of the network. Care is taken during the selection to ensure that the pipework to be measured is the same material, has the same year of construction, and that there are no changes in material due to repairs. Costs can be reduced by using easily accessible infrastructure such as hydrants and valves for the measurements. This is an important factor in the selection process.
Since 2013, measurements have been carried out annually in 40 areas in third party projects. However, due to incomplete condition information about the whole network in an area, the assumption is that 20% of the areas could be selected incorrectly or not selected. In our experience this percentage is nearer to 14% by combining the condition measurements with data from leakages, statistical leakage prediction and the effects of leakages in the renewal decision.
The calculator in Figure 5 shows that Brabant Water has come relatively close to the Economic Assessment Level. At its current level of 7% testing, it spends $2.97/m on data gathering and decision errors combined, with a net savings of almost S$1,900,000/year. At its Economic Assessment Level of testing around 25% of mains, it would spend a combined $2.10/m, and save a further S$550,000/year.
There is an optimal amount of investment in decision support information. Too little information leads to increased error costs where good condition pipes are replaced early. Conversely, over-investing in decision support information can result in waste in cases where lower resolution decision support information is sufficient. This Economic Assessment Level can be calculated for water main replacement program using five utility-specific parameter values and a simple calculation.
The calculation model shows that the extreme ends of the testing curve (no testing, and high resolution inline testing) are often the least economic choices. The case studies demonstrate that survey-level testing of 25% to 100% of the mains is often the optimal choice. The principle driver of how much testing to perform was replacement cost, confirming the intuitive idea that more testing is justified on large mains