Using machine learning for water filtering membranes maintenance

Anticipating when to clean or change water filtering membranes in desalination plants is complex. Veolia Water Technologies worked with Amazon Web Services (AWS) to develop a solution to optimize the timing of the maintenance. Historical time series data was fed into an algorithm to learn from previous patterns and predict the future evolution of fouling indicators such as differential pressure or conductivity of the water, to allow the operating team to efficiently monitor the state of the system and anticipate possible deviations.

Membrane filtration is the most advanced technical process for desalinating seawater, purifying fresh water for drinking water and for industrial processes. These membranes, when combined with design and operating expertise are reliable, efficient and durable. They are vital when it comes to removing salt, micropollutants or any other undesirable dissolved material.

But like any technology or component in contact with water, reverse osmosis (RO) membranes used in desalination plants age over time. To slow this aging as much as possible, membranes must be carefully monitored, rigorously maintained and appropriately cleaned. 

It is well known that RO systems can be very sensitive to changes in the operating environment, making it difficult for the operator to identify and anticipate the real state of fouling and aging of the membranes. In order to fully assess the state of membranes, it is important to rapidly and accurately normalize the raw data to eliminate the influence of external parameters and to diagnose and deal with problems before they become irreversible.

Digitalizing membranes in Oman

The Sur desalination plant is located in the Sultanate of Oman’s eastern region of Sharqiyah. The plant helps fight the depletion of the region’s limited groundwater resources by processing over 132,000 m3/day of seawater, supplying more than 600,000 inhabitants across the Sharqiyah region. 

The plant is operated by Veolia Bahwan, who sought a data-driven decision tool to help maintain the quality and continuity of water production as well as predict when the membranes needed to be replaced or cleaned before they failed or fouled. Doing so would also reduce the downtime and prevent excessive energy and chemical consumptions.

“An emergency shutdown can not always be avoided. But having the ability to better plan the short-term preventative maintenance as well as long-term curative maintenance provides the operator with the ability to optimize water storage capacity to limit the negative impact of unavoidable shutdowns.”
Aditya Akella
Operations manager at the Sur desalination plant

Veolia Water Technologies, who designed and built the Oman Sur plant, collaborated with other Veolia entities to see how data could be harnessed and developed into practical artificial intelligence. In March 2020, the teams started to work with AWS to create a system that would anticipate the timing of maintenance events and predict aging of the membranes. 

As a starting point, three years of historical data coming from the Oman Sur plant was pre-processed, cleaned and prepared for machine learning processing. The clean data set is then normalized by a machine learning algorithm developed together with Veolia’s membrane experts. Finally, AWS machine learning services, including Amazon Sagemaker and DeepAR algorithm, are applied to learn from previous patterns and predict future behavior of fouling indicators — such as differential pressure and conductivity of water — which allow operators to anticipate maintenance operations days or weeks in advance.

These advanced analytics and machine learning algorithms are integrated into the Smart Membranes module of Hubgrade Performance, which has been in operation at the Oman Sur plant since September 2020. Since then, Hubgrade has provided operators with a holistic visibility of the operations and processes, empowering evidence-based decision making when planning for membrane cleaning or replacement. 

Benefits have included predictive maintenance which has helped improve maintenance planning and decision making, and access to key normalized fouling indicators to monitor the effectiveness of CIP and production cycles. Hubgrade allows the Oman Sur team to save valuable time by preventing lengthy manual data extraction.

“Thanks to Hubgrade, it’s possible to identify any membrane issues sooner and be more proactive in planning the corresponding corrective action. Normalizing operational data can now be completed in two-clicks instead of 12 hours of data management and analysis.”
Grégoire Bourguignon
Maintenance Manager

Advanced analytics and artificial intelligence treatments on top of Veolia Water Technologies’ process expertise propelled the Oman Sur desalination plant into the future. Hubgrade has created a significant edge in the value that is delivered to customers, particularly in helping them better operate their water plants and reduce the risk of their business. 

“The applications of machine learning, and more largely of AI, are endless for us. We recently moved our data on a cloud-based data lake and automated all these steps, from data capture up to machine learning. I see us using it across the core of our operations and truly changing the way we look at our data, helping us to develop better services for our customers to be more sustainable and resilient!”
Aude Giard
Chief digital officer at Veolia Water Technologies