PhD: Smart-sensing for systems-level water quality monitoring

University of Glasgow

Glasgow, UK 🇬🇧

Project institution: University of Glasgow

Project supervisor(s): Dr Craig Wilkie (University of Glasgow), Dr Lawrence Bull (University of Glasgow), Prof Claire Miller (University of Glasgow) and Dr Stephen Thackeray (Lancaster University)

Overview and Background

Freshwater systems are vital for sustaining the environment, agriculture, and urban development, yet in the UK, only 33% of rivers and canals meet ‘good ecological status’ (JNCC, 2024). Water monitoring is essential to mitigate the damage caused by pollutants (from agriculture, urban settlements, or waste treatment) and while sensors are increasingly affordable, coverage remains a significant issue. New techniques for edge processing and remote power offer one solution, providing alternative sources of telemetry data. However, methods which combine such information into systems-level sensing for water are not as mature as other applications (e.g., built environment). In response, procedures for computation at the edge, decision-making, and data/model interoperability are considerations of this project.

Methodology and Objectives

Initially, the student will investigate the trade-off between edge computation, at the smart sensor and cloud computation. While cloud computation is powerful, it comes at a high cost of analytics and data storage. When computation is conducted on GPUs at the edge (especially preprocessing) this greatly reduces data loads (raw data are typically high-resolution) and enables analytics in near real-time. Some skills that will be developed:

  • Machine learning (ML): especially embedded/TinyML, from simple novelty detection to more complex models using embedded GPU/TPUs
  • Statistics: there will be a focus on statistical and interpretable ML with uncertainty quantification, to aid decision making
  • Software skills, programming (Python) 

At-sensor GPU computing is integral to this project, with demanding computations leading to impractical power requirements at the edge without the efficiency of GPU computations.

Teaser Project 1: Tiny ML for embedded sensing

The project will develop a smart monitoring system, designed to be embedded within sensing devices (such as NVIDIA Jetson, or Google’s Coral AI). Tools will include signal processing, monitoring algorithms, or more advanced machine-learning techniques. The required data collection and analytics will be scoped with project partners, and the student will develop models/software for edge implementation using GPUs. This would likely involve building models in Python (Tensorflow, Keras, or Jax) and then converting them into an edge-AI device format (e.g. LiteRT). 

Areas of focus:

  • Monitoring of water quality indicators
  • Monitoring of the sensing system itself (batteries, remote power generation)
  • Both current and future sensing technologies

Teaser Project 2: Systems-level analysis of aggregate models and data

The second stage of this project considers how one aggregates information from smart sensors, to inform a whole water-systems analysis. At this stage, the analytics consider interconnected sensors, and how distributed information can inform systems-level decision-making. For example, as smart sensors allow for active control, the study might consider how data collection activities, power schedules, and maintenance can be modified given the ‘bigger picture’. Some relevant topics include:

  • Adaptive experimental design
  • Model fusion and federation
  • Policy learning
  • Decision analysis

A Hierarchy of GPU Computation for Exascale Systems

This project will develop a proof-of-concept, federated GPU computation network, where smart sensors (or alternative computing devices) are integrated as low-power accelerators within the wider exascale system. These hierarchical architectures are referred to as multi-tiered exascale systems (Navaridas, et al, 2019), where a modular approach is designed to scale from the bottom up, for better flexibility. In our case, bottom-tier nodes are naturally suited to pre-processing, real-time analytics, and in-situ image processing. Within exascale systems, these sensing units must complement high-power CPUs and GPUs, where different tiers of computation are designed/conducted in view of the whole – from centralised computation to distributed GPU resources. This approach will demonstrate the potential to develop a federated network as the sensor network expands.

References and Further Reading

  1. JNCC (2024). UKBI – B7. Surface water status. Published here. Accessed 10th December 2024. Last updated 10th December 2024.
  2. Dutta, Lachit, and Swapna Bharali. “Tinyml meets IoT: A Comprehensive Survey.” Internet of Things 16, 2021 (click here)
  3. NVIDIA examples of embedded ML
  4. Navaridas, Javier, et al. “Design exploration of multi-tier interconnection networks for exascale systems.” Proceedings of the 48th International Conference on Parallel Processing. 2019

POSITION TYPE

ORGANIZATION TYPE

EXPERIENCE-LEVEL

DEGREE REQUIRED

You ad could be here!