Reverse Osmosis (RO) is a membrane-based filtration technique used to separate dissolved solids from a liquid solution. Today, RO offers the finest filtration currently available, with the system's semi-permeable membranes rejecting most dissolved and suspended solids, from minerals like salts, to viruses and larger particles.
Because of RO's impressive water purifying capability, it has seen widespread adoption in ultra-pure industrial applications (like the water required to manufacture smartphone components) and forward-thinking utilities like PUB’s NEWater in Singapore, and the 130 million gallons per day Groundwater Restoration System project in Orange County USA.
A 2016 study published by Global Water Intelligence (GWI) estimated a global online reverse osmosis capacity of a significant 74.7 million m3/day, encompassing both the municipal and industrial space. That's about 29,880 standard Olympic swimming pools worth of water. More recently, in 2018, the International Desalination Association (IDA) reported more than 20,000 desalination plants, with a combined production of over 104.7M m3 each day. Reverse osmosis has also seen exponential growth as both industrial and municipal end-users look to seawater and brackish water to meet their operational needs.
Want to dive🤿 deeper? Check out the full in-depth Reverse Osmosis Whitepaper
While the widespread adoption of reverse osmosis technologies shows little signs of slowing down at 10% CAGR since 2004, RO has not been without its share of challenges. In response to stricter water quality requirements, rising costs, and the need for resilient infrastructures (magnified by COVID), O&M professionals and plant managers continue to seek better ways to manage their membrane assets and service schedules, as well as optimize control set-points to maximize reliability and minimize their unit cost ($/m3) of product water.
The top three opportunities we see for machine learning to improve operational efficiencies at reverse osmosis facilities are:
1. MEMBRANE SERVICING
Reverse Osmosis performance (e.g., recovery, water quality) and economics ($/m3) are directly dependent on membrane health management.
“The average seawater reverse osmosis (SWRO) plant spends 21% of its annual operating budget on membrane servicing alone (cleanings and replacements), yet membrane servicing has seen little innovation in 30 years.” - International Desalination Association
The good news is RO facilities actively document cleaning activities and RO performance. Machine learning (ML) techniques, a branch of artificial intelligence (AI), allow models to be developed and trained on such datasets to provide accurate predictions for when a cleaning or membrane replacement is required on a pressure vessel in the plant environment.
2. SPECIFIC ENERGY CONSUMPTION
Reverse osmosis is an energy-intensive treatment process requiring water to be pressurized through a semi-permeable membrane. A typical SWRO facility spends 44% of its operating budget on energy use for the RO treatment alone. Determining the most cost-effective set-points to 1) ensure product water quality, while 2) minimizing energy consumption, requires detailed manual analysis due to degrading membrane conditions and changing influent conditions (e.g., TSS, flow rates). The complexity of such an optimization problem is further compounded in multistage systems with several trains, here operators must determine if running multiple trains at low recovery or a single train at high recovery will yield the lowest operating cost. These manual calculations are not always practical for operators and engineers whose primary task is ensuring safe and continuous plant operations. Machine learning (ML) techniques can be used to determine the optimal flow rates, pressures, and online train configuration to minimize energy use while operating within the constraints of mechanical assets (e.g., pumps, turbomachinery, seals) to meet target water quality.
3. PLANT WIDE RISK MITIGATION
Human error accounts for 24% of unplanned downtime at industrial plants. Traditional risk mitigation practices, like reactive and scheduled maintenance activities, are susceptible to system upsets. System upsets such as pump damage, undetected sensor drifts, and pressure vessel leaks often result in costly downtime for an organization since revenue water is on pause while troubleshooting and maintenance are performed. To prevent these kinds of upsets, plant managers can leverage machine learning’s predictive capabilities to forecast in advance when such issues (e.g., when a pressure vessel seal is leaking) may occur and notify O&M staff so they can plan and allocate resources before it's too late.
Want to learn more?
See our other posts where we expand on each of these three opportunities