A PROCESS-AWARE DATA PREPROCESSING AND TIME-DELAY IDENTIFICATION FRAMEWORK FOR CEMENT CLINKER ENERGY MONITORING

Reliable energy monitoring in cement clinker production requires distributed-control signals, laboratory measurements, and production records to be physically valid and temporally aligned. This study develops a process-aware preprocessing and time-delay identification framework for a 20-variable calcination dataset with a one-minute sampling interval. The framework integrates unit and range checks, three-sigma screening, Hampel verification, duration-dependent gap treatment, operating-regime segmentation, training-only standardization, first-difference cross-correlation, and mutual-information confirmation within physically admissible lag intervals. A controlled 9,000-point benchmark with five representative signals, 0.8% injected spikes of 5–9 standard deviations, and 2–5 min gaps was used for verification. Three-sigma screening achieved a precision of 1.000, recall of 0.997, and F1 score of 0.999. The Hampel detector achieved a recall of 1.000 and a precision of 0.676, additionally identifying 171 legitimate local changes. For four imposed delays, constrained cross-correlation recovered three exactly and one within 1 min, resulting in a mean absolute error of 0.25 min. Unconstrained mutual-information maxima yielded a mean error of 25.5 min under weak and noisy relationships, highlighting the value of applying physically admissible lag constraints. The results support the use of role-specific anomaly detection methods and physically bounded delay estimation prior to energy-model identification.

Hali tarjima qilinmagan