Development of robust estimation and learning techniques for new chemical formulations

点击翻译按钮获取中文摘要

The proposed collaboration between Virox and the University of Guelph will yield the development and implementation of estimation and machine learning strategies on novel chemical formulations. Generally speaking, estimation strategies make use of system models and measurements to predict system states and outcomes. Machine learning methods, also known as artificial neural networks, are essentially input-output mappers that create 'black box' models based on large sets of data. These developed models and techniques will be used to predict new chemical formulations based on desired product design constraints (such as antimicrobial activity, toxicity, corrosion, product shelf life, cost, and others). Additionally, Virox's existing library of peroxide formulation records may be leveraged in this endeavor. The library includes tens of thousands of formulations and data points that have been created and recorded through manual testing. Furthermore, the proposed techniques could be fine-tuned to survey large libraries of chemical ingredients (such as those available with the US Environmental Protection Agency), and theoretically isolate further inert chemical additives that could synergistically increase the antimicrobial activity of hydrogen peroxide in blends. Both Virox and the University of Guelph will greatly benefit from the proposed research project and collaboration. The ability to develop novel formulations in a shorter period of time will reduce Virox's R&D costs for new product development, and will enable greater economic competitiveness. Furthermore, new estimation and machine learning techniques will be developed by the University of Guelph team, and highly qualified personnel (HQP) will be trained on a relevant industrial project.