Industrial-scale dynamic characterization and apparent heat-transfer calibration of a rotary drum dryer using passive operational data

Rotary dryers are critical in tobacco primary processing, where outlet moisture must be tightly controlled for product quality. We characterize a KLD-2 industrial rotary drum dryer using 1.5 years of 1 Hz passive plant data (5.53 million rows; 905 stable batches). Cross-correlation of the within-batch demeaned signals identifies a dominant delay of 171 s between the drum wall temperature and the outlet moisture, interpreted as a combined transport–thermal–mass-transfer lag rather than a pure transport delay. An inverse energy balance calibrates an apparent overall heat-transfer coefficient UA=2.83 kW/K (95% interval 2.74–2.91), corresponding to U≈41.7 W·m−2·K−1, within the range reported for indirectly heated rotary dryers, and the calibrated steady-state gain (−0.086 %/°C) is reproduced by the mechanistic energy balance. Under a strict chronological train/test split, a parsimonious ridge model is best or tied-best at all horizons from 30 to 480 s (RMSE 0.037–0.050%); deep sequence networks approach but do not exceed it. The results are explained by the low variance, high autocorrelation and noise-floor-bounded nature of closed-loop production data, and provide engineering-grade parameters for dryer control and soft sensing.