Sensors ageing in monitoring: a review of behaviours and a probabilistic framework for quantification of fault detection

Purpose This study aims to understand sensors’ behaviour under ageing and its consequences on fault detection: probabilities of false alarm and false silence, which have an impact on maintenance costs and reliability of structures. Design/methodology/approach A literature review is conducted. A drift model is derived from this review and its parameters are analysed. This model is then applied to a real case study. Measurement uncertainties and sensor recalibration strategies are then incorporated into the analysis. Findings A general drift model is derived from the literature review: drift=a+b.(1-e-c.t)+d.t. Its impact on the probabilities of false alarms and missed detections is evaluated: a positive drift will increase false alarms and reduce missed ones. The effect of multiple uncertainty models is also taken into account, and the combined effects of drift and measurement uncertainties are analysed. Building on these results, a probabilistic framework is developed to incorporate the random nature of drift. It is found that a gamma law can explain the evolution of false alarms and silence with this kind of ageing drift. The model was fitted to two available databases and proved to be versatile. It also highlights the wide range of parameter values after fitting, particularly “a” and “d”. The influence of sensor recalibrations on the probabilities of false alarms and missed detections is also assessed: they allow to recover the initial properties of the sensor, reducing the PFA and PFS values by 50%. Research limitations/implications This study presents several limitations. Firstly, the validation relies on aircraft and digital sensor datasets, as no suitable offshore data sets reporting the temporal evolution of PFA/PFS were identified. Secondly, the probabilistic framework assumes independent Gaussian distributions for drift coefficients, neglecting potential coupling effects or heavy-tailed noise behaviours observed in some harsh environments. Thirdly, recalibrations are modelled as perfect resets, ignoring hysteresis and calibration uncertainties. Lastly, the general drift model, while covering the majority of reviewed sensors, excludes rare polynomial or discontinuous behaviours. Future work will address these constraints by investigating non-Gaussian noise, parameter dependencies and site-specific offshore data. Practical implications This work offers a critical commercial advantage for the €80bn offshore wind market by directly reducing operations and maintenance costs, which account for 30% of lifetime expenses. The proposed framework transforms sensor ageing data into actionable reliability metrics, enabling operators to eliminate costly false alarms and prevent catastrophic undetected failures. Investment potential lies in three areas: SaaS integration: Licensing as a premium module for Digital Twin and maintenance platforms. Product certification: Empowering sensor manufacturers to validate “Lifetime Reliability”, commanding higher market prices. Risk modelling: Providing insurers with precise data to price offshore assets accurately. This scalable solution promises significant ROI through optimised maintenance schedules and extended asset longevity. Originality/value This work brings together aspects of sensor science that are typically studied in isolation: measurement uncertainty, drift and fault detection. It also establishes a connection between sensor ageing and the evolution of fault detection performance over time, which, to the authors’ knowledge, has not been explored in this way before. This relationship could be particularly valuable for industries that rely on long-term monitoring, such as civil engineering.