Adapting FMECA for digital reliability assessment of soft sensors in drinking-water systems

The digitalisation of water utilities introduces new operational risks that conventional asset-management frameworks do not fully capture. As utilities increasingly study, test, and implement machine-learning-based soft sensors for real-time monitoring and process optimisation, the reliability of these digital assets becomes essential for maintaining safe drinking-water production. This study presents an adapted Failure Modes, Effects and Criticality Analysis (FMECA) for soft sensors, focusing on information-related failure modes such as data loss, model drift, and alarm malfunction. A modified Risk Priority Number (RPN) integrates detectability through a severity-reduction factor, enabling quantification of the mitigating effect of early fault detection. The framework is demonstrated using the CT (Concentration × Time) ozone-exposure soft sensor for disinfection control in Leiduin drinking-water treatment plant of Waternet, using design specifications and expert-informed assumptions. Ten failure modes were identified across data acquisition, preprocessing, model inference, and output interpretation. Critical modes included missing Supervisory Control and Data Acquisition (SCADA) input, model drift, and alarm failure due to their potential impact on disinfection reliability. Incorporating detectability reduced the RPN by up to 80%; this shows that improvements in information visibility reduce risk more effectively than marginal changes in likelihood or severity. The framework supports ISO 55000-aligned asset-management strategies within FIWARE-based decision-support platforms.