Support Vector Regression for Adaptive Endpoint Detection in Automated Potentiometric Titration: A Methodological Integration of Machine Learning and Process Control

Potentiometric titration requires precise endpoint detection under measurement noise and sensor drift, yet classical derivative-based methods exhibit limited robustness in industrial environments. We investigate the application of support vector regression with feature engineering, such as physics-informed signal for real-time prediction of equivalence point proximity, integrated with a three-stage variable-rate control policy, for titrant delivery, and augmented with conformal prediction–based uncertainty quantification to ensure reliable decision thresholds under sensor drift. The method is validated through 240 titration runs across three acid-base systems (concentrations 0.12–1.3 mol/L) in a 30-day industrial deployment with embedded implementation by Support Vector Regression (SVR) titrant rate control. Compared to first-derivative baseline methods with optimized hyperparameters, the SVR-based controller achieves 97.5% endpoint detection accuracy (95% CI: [94.8%,99.2%]) versus 94.3% (95% CI: [90.9%,97.0%]) for the baseline ( p = 0.041, Bonferroni-corrected α = 0.01), corresponding to a 3.2 percentage point improvement. Root mean square error is 0.041 mL for SVR versus 0.067 mL for first-derivative methods. Under electromagnetic interference (50 V/m, 1 kHz), false-positive rates are 5.0% (SVR) versus 17.5% (first-derivative). The embedded implementation achieves 8.7 ms inference latency on industrial programmable logic controllers through fixed-point arithmetic and lookup table optimization. We provide stability analysis under bounded prediction errors and discuss applicability boundaries, noting that the system requires periodic sensor maintenance and demonstrates limited generalization to chemical systems outside the training distribution. This work contributes a systematic methodology for integrating supervised machine learning with classical process control in resource-constrained industrial settings.