Data Centric Faults Toward Urban Resilience and Intelligent Quality Management in the IEC Continuum

Abstract

The Internet of Things-Edge-Cloud (IEC) continuum is now the main infrastructure for urban and environmental services. It connects sensors, nearby computing devices, and cloud services that differ in capability and operating conditions. This arrangement allows data to be collected close to where it is produced and analysed in large volumes. It also means reliability varies across the system and delays, data loss, or errors can occur at any point. Applications such as air quality, water, traffic, and building management rely on IEC data for decisions and control. When data are incorrect or delayed, decisions degrade and suffer. This thesis examines data-centric faults, meaning problems that change the content or timing of the data, rather than faults in hardware or software. The fault landscape across the continuum is broad and diverse. Faults can arise in devices, networks, or processing, they may be transient or sustained, and they can propagate between layers. There is limited shared structure to relate causes, symptoms, and cross layer effects. Assessments take place at the edge and in the cloud, but their outputs are rarely recorded in a consistent structure, which limits an integrated view and clear provenance. Many detection approaches are built for uniform and complete observations, thus lose their reliability when sampling varies and is partial. To address these gaps, the thesis contributes three components. First, a taxonomy of data faults that structures patterns, sources, and temporal behaviour, aligning practice across deployments and layers. Second, a data quality framework that runs ongoing assessments at the edge and in the cloud and records results in a standard form for monitoring and provenance. Third, an unsupervised detection method suited to real deployment data that describes events using the taxonomy and supports diagnosis without prior labels. The approach is evaluated using data from live urban and environmental deployments. The taxonomy is applied across sites and layers to label data faults in a single scheme. The framework records data quality at the edge and in the cloud in a form aligned with that scheme. The detector identifies events under variable sampling and partial records and assigns taxonomy labels. Taken together, the results provide a practical route to managing data quality across in the IEC.