Predictive Maintenance for Industrial Water Treatment Systems Using Real-Time Sensor Data and Machine Learning

Shevantikar, Pradnesh Amar (2025) Predictive Maintenance for Industrial Water Treatment Systems Using Real-Time Sensor Data and Machine Learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (765kB) | Preview

Abstract

The water-treatment plants are aimed at running continuously, but, nonetheless, maintenance is typically reactive in nature and, thus, it results in time loss due to costly down-time and unreliability of the system. This paper resolves this problem in industrial water-treatment systems based on a simulated dataset based on real operational conditions, where the researcher demonstrates how machine-learning-based predictive maintenance can be applied in predicting failure and improving operational functioning. The case under analysis was an experimental-simulated data of more than 40 000 records and 52 sensor variables (pressure, flow, vibration), environmental conditions (temperature, humidity, CO2) and treatment parameters (RO pressure, UV intensity, mineral dose). After removing and filtering variables, 22 important variables were obtained. The prediction of near-term failures was carried out with the help of the Logistic Regression, Random Forest and XGBoost supervised models and an Isolation Forest was employed to detect the emergence of anomalies. The model explanations that SHAP produced were able to show that the vibration, RO pressure, and UV intensity were the most important predictors. Live-like risk scores, anomaly indicators and SHAP based feature-attribution outputs are the outputs generated by the analysis that can be used to power a Power BI dashboard to support decisions by operators. The results indicate that an optimized XGBoost classifier with a tuned model reached a recollection of 1.00 and F1-score of around 0.73, indicating a high level of prediction ability of upcoming failures, although with a higher false alarm rate. The works provided are explaining predictive maintenance of industrial water systems specific to treatment that can be scaled and introduces a scalable model of data-driven reliability management. This implementation of streams of live sensors in the future will tend to facilitate precision and sustainability outcomes.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Muntean, Cristina Hava UNSPECIFIED
Uncontrolled Keywords:	Predictive maintenance; water treatment; Machine Learning; sensor data
Subjects:	G Geography. Anthropology. Recreation > GE Environmental Sciences T Technology > TD Environmental technology. Sanitary engineering Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	09 Sep 2026 09:01
Last Modified:	09 Sep 2026 09:01
URI:	https://norma.ncirl.ie/id/eprint/9910

Actions (login required)

View Item