Interpretable Data Quality Assessment for Energy and Emissions Monitoring: A Semiconductor Manufacturing Case Study in Singapore

Abstract

Industrial energy and emissions analyses depend on transformations that connect activity records, conversion coefficients and aggregated indicators. An inconsistency in these transformations can resemble an operational anomaly while having a different physical and numerical meaning. This study develops and applies a rule-based framework that combines numerical consistency, semantic alignment and calculation-dependency analysis to explain anomalies in an industrial reporting dataset. A retrospective case study uses an eleven-sheet workbook supplied for a semiconductor facility in Singapore, containing 176.149 GWh of purchased electricity and a combined scope 1 and scope 2 inventory of 173,136.469 t CO₂e. Cross-view reconciliation identifies a 1,152.144 t CO₂e difference attributable to two source contributions excluded from one subtotal. A gas-factor mismatch produces a different response: replacing the displayed C₄F₈ coefficient changes a reverse-calculated quantity but leaves the emissions total unchanged. A fixed-mass calculation would instead suggest a hypothetical difference of 172.105 t CO₂e, illustrating the importance of dependency direction. Monthly auxiliary gas records exhibit a sixteen-fold change in an implied conversion multiplier, whereas the electricity factor cannot uniquely resolve competing reporting-year labels. The results distinguish aggregation, semantic, conversion and temporal inconsistencies within one dataset. The methodological contribution is an explanation procedure connecting each detected inconsistency to its computational mechanism and downstream effect. The case supports the use of dependency-aware data quality assessment before energy-performance modelling, while further datasets are needed to evaluate detection performance and transferability.