Identifying Tomorrow's Problem Today: Predictive Analytics in Wastewater Collections

1. Introduction
The cost of pipeline failure in sanitary sewer systems serving cities and municipalities is increasingly high. A failure where a collapse occurs typically causes a sanitary sewer overflow (SSO) and requires bypass pumping, public notification, and costly construction under emergency conditions. Soil saturation from failing sewer pipelines can lead to collapses and result in imminent danger to surrounding infrastructure and the public. The liability associated with collapse increases in high density areas, high-density traffic regions, and nearby other critical above ground infrastructure. Additionally, the age of the sewer can have a vital impact on the longevity of the infrastructure asset. Most pipe structures often have lifespans of 80-100 years. As the length of use increases past this timeframe, the need for condition assessment becomes increasingly critical. This assessment provides information on the current conditions, but also informs capital planning decisions: pipes may have outrun their useful life due to increased population and demand on services. The pipes within a wastewater collection system have a finite service life due to a multitude of variables, including material, usage, operations and maintenance schedules, and external environmental conditions. In order to identify and better understand the conditions within these main lines, utility and system owners have recognized the importance of regular inspection and assessment of these structures. This inspection data can prevent sewer system overflows (SSO), service disruptions, and most importantly, pipe collapses. However, due to field conditions, it can be incredibly difficult to inspect all structures. Infrastructure assets that are inaccessible or in hard-to-reach locations can pose problems with inspection and often extend the amount of time and money required on a project. Further, the need for inspection does not go away and is only emphasized. It poses a unique conundrum: How does a utility owner assess and prioritize their entire system, if a portion of the system cannot be addressed? How does a utility owner budget their maintenance schedules if their financial budget cannot support a baseline of inspecting their entire system fast enough to prevent failures?
2. Predictive Analytics
Predictive analytics is the use of data mining, predictive modelling, and machine learning to analyze current and historical characteristics to predict future events. It has been used from advertisements, shopping, healthcare diagnoses, and virtual assistants to identify what someone or something is likely to do.
Predictive Analytics has also been utilized in the water industry. Drinking water infrastructure is mostly pressurized pipe (i.e. Force main), which is often difficult to assess without redundancy, costly bypassing, or taking the pipe out of service. This makes it even more difficult for system owners to determine current condition, remaining useful life, or even rehabilitation decisions. Through the use of artificial intelligence and machine learning, pipeline attributes are fed through proprietary analytical models to look at commonalities of pipes that have failures to determine patterns and trends. These patterns and trends lead to predictability.
These attribute datasets are enormous and include: - historical failures and when those failures occurred - types of failures - pipe size and material - pipe age and repair history - current conditions - maintenance schedules - geospatial data (GIS) - weather patterns - foliage environment - soil conditions
The output that is produced is a ranking list of pipes based on their likelihood of failure (LoF). LoF indicates the probability of failure based on the attributes above. The weight and severity of these attributes can vary from system to system, but some generalities can be assigned: A pipe that is larger in diameter or conveys more flow has a higher consequence of failure (CoF) than a smaller diameter line. Additionally, a pipe that is below a populated area or nearby critical piece of infrastructure such as a hospital or government building, could have devastating consequences if it were to fail, compared to those in more rural or less-populous areas.
3. Wastewater
Asset Management is not just for the drinking water side of our industry! There are plenty of utility owners responsible for the oversight, budgets, and maintenance of wastewater collections systems. Further, these systems are often gravity-fed, providing access to the pipe for condition assessment compared to pressurized lines. These conditions assessments typically involve CCTV to identify defects and deficiencies. Fortunately, NASSCO has set the standard for how these defects should be coded and classified, making it easier to compare datasets. The approach to maintenance and rehabilitation decisions are commonly based on 2 main facets of a pipe: whether the pipe had failed before and the age of the pipe. It is assumed that as the pipes get older or has an issue that requires maintenance, that pipe is likely to fail again or require continued maintenance in the future.
4. Approach and Case Study
With over one hundred (100) million linear feet of inspections completed to date, RedZone Robotics sought to see how this NASSCO-coded defect data could be used to better predict failures within a wastewater system. Four (4) million linear feet of CCTV data was provided to Voda.ai's machine learning algorithm. All data is PACP v6 compliant and is comprised of inspections ranging from 6-inch to 36-inch pipe of various materials.
This case study and presentation covers the methodology of cleansing the data, building the model, and the results for predicting structural and operations and maintenance failures within wastewater assets. In summary, the predictive model was able to identify 3x as many failures on pipes compared to prior failure history and 65x as many failures compared to the age of the pipe and. Additionally, the model could also identify the types of failures, providing utility owners with more knowledge about potential failures and how to address them.