DIGITAL WATER. CLIMATE RESILIENCE.

Tackling Limitations of IOT Sensors in Digital Twins using Soft Sensors

Written by Ismail Weiliang and Evan Harwin, Data Scientist, Moata Data Science, UK

20 OCT 2022

3 MINS READ


ISMAIL WEILIANG

The Climatebender

EVAN HARWIN

Data Scientist

Moata Data Science

Views are entirely ours

and not connected to any company

Challenges from Physical Sensors

Physical sensors are commonly used to monitor water quality and are critical for water treatment plants’ operation. However, there are many inherent issues surrounding physical sensors that are challenging to tackle. Some issues inherent to physical sensors are:

  1. Experience sensor drifts over time. These are difficult to detect in a timely manner by operators especially if drift occurs slowly. Sensor drift affect operators’ decision making and even the process control system
  2. Require high maintenance with regular scheduled cleaning, monitoring, and event detection
  3. Prohibitively expensive with specific equipment and supplementary lab tests for data validation

_________________

Quick Take:

  • Physical sensors are critical for water treatment plants’ operation but has inherent issues such as sensor drifts which are challenging to tackle.

Soft Sensors for Digital Twins

Soft sensors are a digital version of a physical sensors where several measurements are processed together for a specific parameter. They are computational and statistical correlation-based methods. The challenges from physical sensors can be tackled through a multi strategy approach using soft sensors to augment an array of physical sensors, we can enable more traces and higher data quality in digital twins. We can use soft sensors to supplement, validate, and potentially replace physical sensors.

  1. Supplement: Access unavailable live data where sensor probe for specific parameter not available or too expensive, reduce latency on laboratory work which has a time lag in receiving results whereas soft sensors deliver results in milliseconds, scale up digital twin with more robust data.
  2. Validate: Identify sensor drift, filter out anomalies, increase trust in sensor data.
  3. Replace: With low deployment cost and maintenance, easy and quick testing and iteration - soft sensors can be used in place of physical sensors for secondary factors that might not be worth the maintenance burden of physical probes.

Soft sensors can potentially be deployed across domains and projects where physical sensors applications are valuable and where digital twins are being built.

_________________

Quick Take:

  • Soft sensors can supplement, validate and potentially replace physical sensors to better live data for digital twins in water treatment plants

Machine Learning for Digital Twins

Neural networks provide the most suitable architecture due to their ability to learn and identify relationships with little or no prior knowledge, handling multiple unstructured time series in parallel.

Machine learning using pattern-based matching methods to model sensor readings by processing the combined historical data of multiple adjacent and associated sensors in the water treatment process can be developed and assessed for performance by cross validation routine. 

_________________

Quick Take:

  • Machine learning by processing the combined historical data of multiple adjacent and associated sensors in the water treatment process can be developed to better live data for digital twins in water treatment plants.

Potential in Digital Twins

Mott MacDonald Moata data science and Smart Water recently built and delivered three live soft sensor deployments for a wastewater treatment plant in New Zealand. Mott MacDonald Singapore and Ada Mode from Spain is also currently working on PUB Global Innovation Challenge on this exact challenge: Innovation Challenge Sensor Data Integrity Monitoring (pub.gov.sg)

With the combination of soft sensors and ML for digital twins in water treatment plants, this opens countless opportunities where modularly designed digital twins can be potentially connected to other digital twins in the water sector.

Authors:

Ismail Weiliang is a climate resilience consultant with over half a decade of experience and specialises in flood risk advisory for Asia. His work involves advising governments and development banks on strategies to transform climate risks into resilience. He also founded a non-profit organisation “The Climatebender” that provides humanitarian relief to communities vulnerable to the climate crisis.


Evan Harwin is a data scientist from Moata Data Science team, Mott MacDonald UK. He is experienced in researching and building out mathematical and predictive models forming the computational core of digital twins spanning multiple sectors. Notable examples include predictive modelling for water levels in complex urban waterways, statistical modelling to outline possible configurations for a set of trains traversing a complex junction and building out digital twins for monitoring WWTP performance and health.

©2021 The Climatebender