Effective pH conditioning is an essential step prior to corrosion testing, inhibitor testing, aqueous surface treatments, aqueous cleaning, and material-degradation studies. When dealing with buffered electrolytes, the final value of pH cannot account for any historical information regarding the conditioning steps due to repeated adjustments to reach the final conditions through repeated acid-base titrations. Such manipulations may cause a different concentration, ionic strength, buffer state, and transient reaction chemistry of the solution before reaching the intended condition. This paper explores robot-assisted, low-observation, pH conditioning for buffered electrolytes prepared using acetate, citrate, ammonium, and potassium dihydrogen phosphate. Eighteen binary systems of buffer solutions at ratio 1:1, 1:2, and 2:1, along with a 4-compound electrolyte (citrate, phosphate, ammonium, and acetate), have been considered. An addition of acid or base was expressed by the titrant volume with sign. As such, only one pH meter loop could adjust the next addition based on measured pH values. Several machine learning approaches, namely linear regression, random forest regression, artificial neural network regression, and Gaussian process regression, were compared in their capability to reach the target pH range with low observation numbers. The best low-observation performance was demonstrated by Gaussian process regression, needing $3.1\pm 0.6$ active observations, compared to $3.4\pm 0.3$ by random forest regression and $5.6\pm 1.0$ by artificial neural network regression. Adding a chemical descriptor did not have a significant effect on performance, needing $3.1\pm 0.6$ active iterations with the large descriptor set and $3.2\pm 0.6$ with small descriptors. Minimizing the measurement burden to only $5.8\pm 0.6$ pH measurements, two initial pH values were the most suitable. Transfer learning from relevant single buffer titration history further reduced the number of active observations to $2.2\pm 0.4$. Robotic experiments showed that the citrate/phosphate system needs two iterations for conditioning, the acetate/citrate systems need up to eight iterations, and a 4-compound electrolyte reaches pH 6.0 in $3.7\pm 0.4$ active observations.

buffered electrolyte,corrosion media,pH conditioning,robotic chemistry,Gaussian process regression,active learning,transfer learning,aqueous materials chemistry