Dynamic adaptability of water supply networks (WSNs) in terms of connectivity and hydraulic conditions is essential for their operation as there are increasing demands on serviceability (leakage, water quality, incident management and fire flow), resilience and cost efficiency. A common approach to achieve multiple control functions throughout networks is to employ automatic control valves (ACVs).

Advances in low-powered electronics and micro-actuators enable a wide range of novel control methods in WSNs, including the flow-based pressure control (or flow modulation control (FM)). The implementation of FM schemes has been steadily increasing as it has a major advantage of a closed-loop (feedback) control by utilising measurements to define the flow-pressure control profile. The performance of the FM scheme relies on continuous and accurate flow measurements. Hence, to achieve robust control in WSNs, high-level reliability of the control solution is required. Herein, two methods for the reliable operation of ACVs are investigated, namely (i) Flow estimation and (ii) Fault detection and diagnosis.

A novel flow estimation method for diaphragm-actuated globe valves has been developed and experimentally investigated. The method utilises three pressure measurements, namely the valve inlet pressure, the valve outlet pressure and the control chamber pressure (the 3P flow estimation method). The method relies upon the accurate computation of the valve stem position, the measured pressure differential across the valve and the flow coefficients of the valve (Cv, Kv). The developed valve stem position estimation model results in multiple solutions. Advances in signal processing are combined with a machine learning technique (support vector machine) to distinguish the correct solution. The proposed 3P method is compared with a method which uses sensor measurements of the valve stem position (the 2P&Pos method), and its performance validated against measurements from an electromagnetic flowmeter. The uncertainty bounds of the flow estimation methods are also derived.

For fault diagnosis, methods for early fault detection and diagnosis (FDD) are investigated.
Potential faults are categorised, and residuals and feature variables are defined to detect a fault and diagnose its likely cause. Experimental data have been generated and utilised from controlled laboratory conditions, from an operational network and also from a numerical simulation. The performance of the proposed schemes has been validated.