Data Quality Field: A Physics‐Inspired Framework for Industrial Time Series Quality Assessment

ABSTRACT This article presents a physics‐inspired data quality assessment framework for industrial time series based on field theory. Guided by the ISO 8000 fit‐for‐purpose principle, the framework addresses deviations that impair downstream data use. Specifically, discrete time series data are first mapped into a unified space via time‐delay embedding and principal component analysis; deviant data points identified by a replaceable front‐end module then serve as negative field sources. Solving the Poisson equation yields a continuous quality potential field that captures spatiotemporal quality propagation, which existing assessment methods do not provide. The framework makes three contributions. First, the field formulation provides a unified description of spatiotemporal quality distribution, capturing how deviations influence neighbouring observations via the potential function and field strength. Second, solving the Poisson equation with domain‐informed boundary conditions yields physically meaningful propagation and interpretable quality maps: the potential reveals the local quality state, while elevated field strength delineates degradation boundaries. Third, hierarchical indices at the unit, region, and global levels support multi‐granularity monitoring. Experiments on real manufacturing time series from a cigarette factory show that the presented method achieves comprehensive quality assessment with an average processing latency of 1.5 ms and effectively reveals spatiotemporal degradation patterns that point‐wise scoring methods fail to capture.