Ultrasonic and Resistivity Sensor Fusion for Discriminating Gas Injection Rates

Discriminating gas injection rates in drilling fluids is essential for monitoring downhole gas influx. However, each gas bubble traversing the sensing zone causes ultrasonic and resistivity readings to depart from the baselines recorded without gas injection, while a single acquisition cycle captures only a random snapshot of the bubble distribution. In this work, a sliding window framework is developed that combines ultrasonic and resistivity measurements to distinguish gas injection rates in a vertical static liquid column. Ultrasonic and resistivity indices are extracted from individual acquisition cycles, aggregated within a 2.5 s sliding window containing five consecutive acquisition cycles, and fused by normalizing classwise products without introducing trainable parameters. Experiments were conducted in freshwater and $2.0\mathrm{g}{\mathrm{cm}}^{-3}$ water-based mud at four discrete injection rates from 0 to $0.3\mathrm{L}{\min }^{-1}$. Evaluation used cross-validation with data split by experimental repetition. The fusion framework achieved a macro F1 of 84.7% ± 4.3% in water-based mud, a gain of 6.3 percentage points over the best result from either channel alone. The classwise product also gave the highest mean macro F1 among three evaluated nontrainable fusion rules in both media. The resulting fusion framework supports gas injection rate discrimination under static column conditions.