IoT-SCADA Integrated Deep Learning Model for Leak Detection in Fluid Transportation Networks

Purpose: This study aims to develop an Internet of Things (IoT) driven fluid transportation monitoring and control system integrated with a deep learning-based leak detection model, addressing the persistent challenge of leakage caused by ageing pipelines, corrosion, and operational errors. Design/Methodology/Approach: A hydraulic transportation network incorporating transducers, a programmable logic controller (PLC) and a Supervisory Control and Data Acquisition (SCADA) system was modelled and simulated to emulate leak and no-leak scenarios under varying operating conditions. Each sample is a steady-state snapshot of 20 transducer readings (pressure, flow rate, temperature and vibration measured at five stations along the flow direction). The dataset of 866 samples (351 no-leak, 515 leak) was split into 766 training samples and a 100-sample hold-out test set with class proportions matching the full dataset. A compact 20–4–2 neural network (ReLU hidden layer, two-class softmax output, 94 trainable parameters) was trained with the Adagrad optimiser using cross-entropy loss, which for a two-class softmax with one-hot targets is mathematically equivalent to a single-sigmoid binary classifier. Performance was benchmarked against reference classifiers and reported with imbalance-robust metrics and 95% confidence intervals. Research Limitation: The results are simulation-bound: they were obtained on data from an emulated test rig using a sample-level hold-out split, and have not yet been validated on physical or operational network data. Scenario-aware cross-validation, comparison with hydraulic and classical machine-learning baselines, probability calibration, and broadening of variables such as temperature and total dissolved solids (TDS) are required before field deployment. Findings: On the hold-out set, the model achieved an accuracy of 91.0% (95% CI 83.8–95.2%), precision of 90.6%, recall of 95.1%, F1-measure of 92.8%, specificity of 84.6% and a Matthews correlation coefficient (MCC) of 0.81, compared with an F1-measure of 75.8% and an MCC of 0 for a majority-class reference classifier. The false-alarm rate of 15.4% identifies specificity as the main target for improvement. Practical Implication: The proposed IoT-SCADA-AI pipeline offers transportation network operators an accessible route to early leak detection, reducing resource losses and unplanned maintenance costs. Social Implication: Reliable, low-cost leak detection supports reduction of non-revenue water in urban fluid supply networks and contributes to sustainable resource management. Originality/Value: The paper offers a lightweight (94-parameter), computationally efficient alternative to the CNN/LSTM-based leak detection approaches common in the literature, achieved by coupling a customised multiple-model hydraulic representation and PLC/SCADA data acquisition with a shallow neural classifier suitable for edge deployment.

Authors

Publication Details

Journal
African Journal Of Applied Research
Published
2026-11-18
DOI
https://doi.org/10.26437/fhpx4k05
Primary Topic
Water Systems and Optimization
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

IoT-SCADA Integrated Deep Learning Model for Leak Detection in Fluid Transportation Networks

K. L. Bhoite, S. M. Bhosle, A. M. Deshmukh
1 links
African Journal Of Applied Research
Water Systems and Optimization
article

IoT-SCADA Integrated Deep Learning Model for Leak Detection in Fluid Transportation Networks

K. L. Bhoite, S. M. Bhosle, A. M. Deshmukh
article en

Abstract

Purpose: This study aims to develop an Internet of Things (IoT) driven fluid transportation monitoring and control system integrated with a deep learning-based leak detection model, addressing the persistent challenge of leakage caused by ageing pipelines, corrosion, and operational errors. Design/Methodology/Approach: A hydraulic transportation network incorporating transducers, a programmable logic controller (PLC) and a Supervisory Control and Data Acquisition (SCADA) system was modelled and simulated to emulate leak and no-leak scenarios under varying operating conditions. Each sample is a steady-state snapshot of 20 transducer readings (pressure, flow rate, temperature and vibration measured at five stations along the flow direction). The dataset of 866 samples (351 no-leak, 515 leak) was split into 766 training samples and a 100-sample hold-out test set with class proportions matching the full dataset. A compact 20–4–2 neural network (ReLU hidden layer, two-class softmax output, 94 trainable parameters) was trained with the Adagrad optimiser using cross-entropy loss, which for a two-class softmax with one-hot targets is mathematically equivalent to a single-sigmoid binary classifier. Performance was benchmarked against reference classifiers and reported with imbalance-robust metrics and 95% confidence intervals. Research Limitation: The results are simulation-bound: they were obtained on data from an emulated test rig using a sample-level hold-out split, and have not yet been validated on physical or operational network data. Scenario-aware cross-validation, comparison with hydraulic and classical machine-learning baselines, probability calibration, and broadening of variables such as temperature and total dissolved solids (TDS) are required before field deployment. Findings: On the hold-out set, the model achieved an accuracy of 91.0% (95% CI 83.8–95.2%), precision of 90.6%, recall of 95.1%, F1-measure of 92.8%, specificity of 84.6% and a Matthews correlation coefficient (MCC) of 0.81, compared with an F1-measure of 75.8% and an MCC of 0 for a majority-class reference classifier. The false-alarm rate of 15.4% identifies specificity as the main target for improvement. Practical Implication: The proposed IoT-SCADA-AI pipeline offers transportation network operators an accessible route to early leak detection, reducing resource losses and unplanned maintenance costs. Social Implication: Reliable, low-cost leak detection supports reduction of non-revenue water in urban fluid supply networks and contributes to sustainable resource management. Originality/Value: The paper offers a lightweight (94-parameter), computationally efficient alternative to the CNN/LSTM-based leak detection approaches common in the literature, achieved by coupling a customised multiple-model hydraulic representation and PLC/SCADA data acquisition with a shallow neural classifier suitable for edge deployment.

African Journal Of Applied ResearchVol. 12(8)
Industry, innovation and infrastructure
Water Systems and Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon