A Generalizable Dual-Engine Methodology for Predictive Process Intelligence and Analytical Quality Assurance
Year: 2026
Type: preprint
Abstract: Industrial and engineered systems are exposed to at least two distinct sources of uncertainty: deviations in the physical process and errors or interference in the measurements used to assess that process. A process may be abnormal while its measurements remain valid; conversely, a process may be nominal while the measurement system is unreliable. This paper formalizes the Universal Dual-Engine Framework (UDEF), a modular methodology that separates and then integrates (i) dynamic process-state estimation and anomaly detection and (ii) analytical measurement-validity assessment. A domain-specific state-space model generates predictive residuals and uncertainty estimates. A separate measurement-assurance module evaluates sensor health, matrix interference, calibration status, recovery, and method-suitability evidence. A decision layer combines these outputs into operational states without conflating process faults with measurement faults. The paper defines mathematical interfaces, a reference decision algorithm, implementation requirements, and a reproducible synthetic benchmark spanning four stylized domains: bioprocessing, battery manufacturing, chemical processing, and logistics. In the benchmark (5,000 synthetic episodes per domain), the dual-engine rule achieved a mean joint classification accuracy of 0.944, compared with 0.797 for a single residual-threshold baseline. Mean process-fault F1 increased from 0.776 to 0.882, while the dual engine also classified synthetic measurement faults (mean F1 0.989). These results demonstrate only behavior under the stated synthetic data-generating assumptions; they do not establish superiority on real industrial data. The contribution is therefore a formalized, testable integration methodology and validation protocol, not a claim that predictive monitoring, sensor validation, QbD, or analytical validation are individually new.
Source: Research Square
Author: Zhuofan Shen
Institution: Universitat Internacional de Catalunya
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225425698