Predictive Modeling for Filter Longevity in Recirculating Shower Systems
(2025)Department of Automatic Control
- Abstract
- Water reuse technologies such as Orbital Systems’ recirculating shower offer significant environmental benefits by reducing hot water consumption. However, these systems rely on depth filters whose performance degrades over time due to clogging. Current maintenance strategies use fixed time intervals or pressure-based thresholds, which do not account for individual usage patterns or actual filter condition.
This thesis explores data-driven methods for estimating filter health and forecasting remaining filter life using onboard sensor data. A flow-normalized pressure model, based on a second-degree polynomial, is developed to compensate for variation in user-selected flow rates. This model enables consistent pressure comparisons across... (More) - Water reuse technologies such as Orbital Systems’ recirculating shower offer significant environmental benefits by reducing hot water consumption. However, these systems rely on depth filters whose performance degrades over time due to clogging. Current maintenance strategies use fixed time intervals or pressure-based thresholds, which do not account for individual usage patterns or actual filter condition.
This thesis explores data-driven methods for estimating filter health and forecasting remaining filter life using onboard sensor data. A flow-normalized pressure model, based on a second-degree polynomial, is developed to compensate for variation in user-selected flow rates. This model enables consistent pressure comparisons across devices and sessions.
A lightweight, two-stage algorithm is proposed for real-time filter status estimation, designed for execution on embedded hardware. It uses normalized pressure and session-level smoothing to provide a continuous, interpretable health metric. In parallel, a machine learning model, based on a Random Forest regressor, is trained to predict remaining filter capacity in liters. The model emphasizes late-life accuracy through sample weighting and was evaluated using a real-world dataset of over 240,000 sessions, 700 shower devices and 850 filters.
Deployment strategies are discussed for both cloud-based and embedded implementations, along with ethical and privacy considerations. The results demonstrate that combining physics-informed normalization with machine learning can significantly improve maintenance timing, reduce waste, and enhance user experience in water reuse systems. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9208527
- author
- Meurling, Johan
- supervisor
- organization
- year
- 2025
- type
- H3 - Professional qualifications (4 Years - )
- subject
- report number
- TFRT-6287
- other publication id
- 0280-5316
- language
- English
- id
- 9208527
- date added to LUP
- 2025-08-08 15:07:47
- date last changed
- 2025-08-08 15:07:47
@misc{9208527,
abstract = {{Water reuse technologies such as Orbital Systems’ recirculating shower offer significant environmental benefits by reducing hot water consumption. However, these systems rely on depth filters whose performance degrades over time due to clogging. Current maintenance strategies use fixed time intervals or pressure-based thresholds, which do not account for individual usage patterns or actual filter condition.
This thesis explores data-driven methods for estimating filter health and forecasting remaining filter life using onboard sensor data. A flow-normalized pressure model, based on a second-degree polynomial, is developed to compensate for variation in user-selected flow rates. This model enables consistent pressure comparisons across devices and sessions.
A lightweight, two-stage algorithm is proposed for real-time filter status estimation, designed for execution on embedded hardware. It uses normalized pressure and session-level smoothing to provide a continuous, interpretable health metric. In parallel, a machine learning model, based on a Random Forest regressor, is trained to predict remaining filter capacity in liters. The model emphasizes late-life accuracy through sample weighting and was evaluated using a real-world dataset of over 240,000 sessions, 700 shower devices and 850 filters.
Deployment strategies are discussed for both cloud-based and embedded implementations, along with ethical and privacy considerations. The results demonstrate that combining physics-informed normalization with machine learning can significantly improve maintenance timing, reduce waste, and enhance user experience in water reuse systems.}},
author = {{Meurling, Johan}},
language = {{eng}},
note = {{Student Paper}},
title = {{Predictive Modeling for Filter Longevity in Recirculating Shower Systems}},
year = {{2025}},
}