pHbot: Self‐Driven Robot for pH Adjustment of Viscous Formulations via Physics‐informed‐ML**

Abstract

AbstractpH adjustment is crucial for many industrial products, yet this step is typically performed by manual trial‐and‐error. A particularly industrially relevant yet challenging titration is that of adjusting viscous liquid formulations using weak, polyprotic titrants (usually citric acid). Handling of viscous, non‐Newtonian formulations, with such polyprotic acids preferred for their chelation and buffering effects make a robotic solution challenging. We present a self‐driving pH robot integrated with physics‐informed learning; this hybrid physical‐ML model enables automated titration with weak‐strong acid/base pairs. To deal with the high viscosities of these formulations, we developed specific automated mixing and cleaning protocols. We hit the target pH within two to five iterations over 250 distinct formulations in lab‐scale small‐batch (~10 mL and 12 samples) titrations. In the interest of scaling up to match industrial processes, we also demonstrate that our hybrid algorithm works at ~25× scale‐up. The method is general, and we open‐source our algorithm and designs.

Accelerating Formulation Design via Machine Learning: Generating a High-throughput Shampoo Formulations Dataset

“…Next, we titrated the formulations to pH 5.8 ± 0.2 with citric acid or sodium hydroxide, as per an industrial requirement, using our fully automated pHbot 31 . Our target pH is typical for shampoo formulations 36 , 37 .…”

“…After this step, the formulations were mixed and characterised for their phase stability. We carried forward the stable samples to our pHbot 31 for pH adjustment, and once again, after a 36-hour wait (an agreed processing step), assessed stability of our products, before we measured turbidity and viscosity of the stable formulations. Rheology measurements were performed offline, as they are very difficult to automate, particularly for highly-viscous and non-Newtonian fluids 32 .…”

“…1 An overview of the liquid formulation workflow. (i) From an ML-guided design of experiments 29 ; (ii) preparation of the liquid formulations by automated viscous liquids handling 30 ; (iii) mixing on a multi-position stirrer plate; followed by (iv) pH adjustment on the pHbot 31 ; and finally, (v) characterisation including optical imaging for phase stability, proxy UV-vis absorbance for turbidity, and rotational rheometry for viscosity measurements. …”

“…Notebooks and scripts for the automated viscous liquid handling and gravimetric analysis to prepare and determine the composition of the liquid formulations can be found at: https://github.com/sustainable-processes/formulations-prep . The code to run the pHbot has been published separately: https://github.com/sustainable-processes/pHbot 31 . And all the software to develop the computer vision phase stability classifier can be found: https://github.com/sustainable-processes/stability-computer-vision .…”

Machine Learning-Guided Space-filling Designs for High Throughput Liquid Formulations Development

“…This is summarised in Figure 1 which shows that a set of experiments are executed according to the DoE. These formulations are then assessed for their phase stability, with the stable samples being titrated to an industrially specified pH target (Chitre et al, 2023).…”