A Machine Learning Approach to Evaluate Impedance Spectra for Glycine Crystallization Monitoring

In industrial crystallization, ionic impurities or additives like salts or acids can affect various phenomena that occur during crystallization. Process control that aims to achieve a specific crystal shape must therefore take these properties into account. A soft sensor based on electrical impedance spectroscopy is presented that holds promise to measure the concentrations of dissolved and crystalline target substances and the concentration of the additive using only a single probe. A feasibility study is presented here for the model system glycine–water–NaCl. The spectra are evaluated using artificial neural networks. 5140 spectra were recorded as training data during various cooling crystallization processes. The model predictions show promising quality when applied to data previously unseen by the trained model. For future process applications, the precision can be further improved by increasing the number of training data and/or restricting them to the actually relevant parameter range.