Part of my job is to keep track of water and sewer modeling research. I’m constantly reading journal papers or attending conference sessions to monitor the latest research. The journal papers always have attractive titles, promoting that they have solved real engineering problems, and the authors conclude that their work has greatly advanced the state of the world. However, when I peek under the hood at the underlying math, I’m almost always disappointed.
Real-world problems are ugly, and simplifications and assumptions need to be made to make the problem solvable. The researchers tend to gloss over these assumptions—if they mention them at all. I think many researchers may not even realize that they are making them. Most researchers are graduate students and their advisors, who have never worked on real-world problems. As a starting point for their work, they rely on prior research papers that contain assumptions that the prior authors did not mention. As a result, real-world complications are often lost in legacy papers.
Listed below are a collection of dangerous assumptions that are frequently made in papers I’ve read over the years. See how many you’ve encountered.
Demand forecasts are perfectly accurate
Pipes have a useful life—somewhere on the order of 100 years. It’s amazing how many pipes sizing optimization models ignore the uncertainty in demand forecasts and base their optimal designs on a forecast that will almost certainly be wrong. Think about how your city has changed over the last 100 years and what kind of demand forecast would have been made back in 1921. The real goal of water utilities is not necessarily minimum cost but rather a design that will provide reliability and capacity today while fitting into a long-term, uncertain system, at a reasonable cost.
The entire water system is constructed at once
Very rarely can a water system be constructed at time zero and work for the entire planning horizon. Utilities need a phased approach to the installation of improvements because they can’t afford to install the system for 2041 all at once. But many papers assume that phasing of construction isn’t needed, and a single planning horizon is sufficient. In fact, many of the problems being studied (e.g., aging infrastructure, optimization, smart metering) exist precisely because a perfect solution could not be foreseen, and water systems evolved over time, piece by piece. Planning based on a 2041 forecast essentially assumes that the world will end after 2041.
Costs are only a function of pipe diameter
The cost of installing a pipe depends on many factors and pipe size is only one of them. A 6-inch pipe can cost more than a 12-inch pipe, depending on laying conditions. However, almost all research papers use a single cost function because developing specific cost estimates is hard work and may not generalize to conditions somewhere else. When I started working in optimization, I thought that cost estimating was “trivial.” Now I realize it’s the most important component.
Hydraulic grade can be measured with incredible precision
Automated model calibration requires that the measured hydraulic grade (i.e. pressure and elevation of pressure gage) is measured, and it is assumed that it is measured at unrealistic accuracy and precision. In the real world, you’re lucky if you can measure hydraulic grade to within a few feet, yet researchers will assume they can measure it to an accuracy of 0.01 ft or some other unjustifiable value. Sure, some SCADA systems display pressure measurements like 62.487201 psi—but that doesn’t mean those values are correct. Getting useful field data is a challenge.
Assume you know why a model is out of calibration
There are many reasons for discrepancies between hydraulic models and data from the real system. I’ve compiled a list of over 30 reasons which can include connectivity errors, incorrect pressure zone boundaries, inaccurate pressure sensor, mistakenly closed valves, incorrect pump curve, etc. yet most researchers pick one (usually C-factor) or two (C-factor, demand) and assume that all the inaccuracy in the model is due to those parameters. Such a practice is what I call “calibration by compensating errors” where a modeler compensates for an error in one parameter by an offsetting error in another parameter. This can appear to work when you have one data set but falls apart when you have several.
Assume pump efficiency is constant
Researchers who want to determine optimal pump scheduling may not want to mess with determining pump efficiency, especially for variable speed pumps. The trick is to assume that pump efficiency is a constant. This makes the math a lot easier, but isn’t the idea behind pump energy optimization finding and avoiding those inefficient operating points?
Leak size doesn’t matter
There have been quite a few papers on using pressure anomaly data to identify leaks. In theory, this should work, but it takes a fairly large leak to cause a pressure anomaly that rises above background noise. Researchers will claim success with their algorithm when it can find a 200 GPM leak, when leaks of that size will almost always reach the surface and find you (without needing to look for them). How many people doing work in this area don’t have an idea of what 200 GPM looks like?
An 8-pipe system is real
I’ve seen a lot of researchers claim success with their algorithm when they can solve a model with 8 pipes. But don’t give them a medal yet. I haven’t seen many real systems with 8 pipes. An 8-pipe system is good for debugging your code, but the real test is whether the algorithm can scale up to a real system with thousands of pipes.
Valves don’t matter when it comes to system resilience
The impact of a pipe break on a system depends on the location of isolation valves to isolate the pipe break. Yet, many researchers assume that isolating a break involves removing a single pipe from the pipe network model, which is often not the case. This seems to arise from the necessary simplification of hydraulics models, where a closed pipe is the same as a closed valve. Isolation depends on the location of valves along the pipes, and solutions that ignore valve location can steer the solution towards providing misleading results.
I have more, but you get my point. A lot of research papers contain assumptions that render the solution questionable in many cases because the researcher doesn’t understand the assumptions being made and/or doesn’t disclose the assumptions. While most projects start from a real problem to be solved, it is important to get a deep understanding of the problem. Researchers should not solely rely on previous literature to define the solution, but communicate with engineers and operators who are faced with the problem on a day-to-day basis. If they must make assumptions, they need to state what they are and attempt to understand their impact.
Bentley OpenFlows tries to steer users away from many of these pitfalls, but engineers need to understand what is going on when they hit the “Compute” button.
Want to learn more from our resident water and wastewater expert? Join the Dr. Tom Walski Newsletter today!
