Artificial neural networks for quantitative online NMR spectroscopy

• DOI: 10.1007/s00216-020-02687-5
• Publication Date: 2020-05-09T05:02:35Z
• Authors: Simon Kern; Sascha Liehr; Lukas Wander; Martin Bornemann-...; Simon Müller; Michael Maiwald; Stefan Kowarik
• Keywords: Automation; Process industry; Artificial neural networks; 13. Climate action; Online NMR spectroscopy; Real-time process monitoring; 01 natural sciences; Research Paper; 0104 chemical sciences

Abstract Industry 4.0 is all about interconnectivity, sensor-enhanced process control, and data-driven systems. Process analytical technology (PAT) such as online nuclear magnetic resonance (NMR) spectroscopy gaining in importance, it increasingly contributes to automation digitalization production. In many cases up now, however, a classical evaluation of data their transformation into knowledge not possible or economical due the insufficiently large datasets available. When developing an automated method applicable sometimes only basic limited number batch tests from typical product development campaigns are However, these enough for training machine-supported procedures. this work, overcome limitation, new procedure was developed, which allows physically motivated multiplication available reference order obtain sufficiently dataset machine learning algorithms. The underlying example chemical synthesis measured analyzed with both application-relevant low-field NMR high-field method. Artificial neural networks (ANNs) have potential infer valuable information already relatively input data. predict concentration at complex conditions (many reactants wide ranges), larger ANNs and, therefore, required. We demonstrate that moderately problem four can be addressed using combination presented PAT (low-field NMR) proposed approach generate meaningful