A UK spin-out has launched and trialed an artificial intelligence (AI)-based tool that incorporates experimental and process data, as well as uncertainty. It helps companies achieve the best possible material or chemical optimization with about 80% fewer experiments. New materials and molecules have been developed and manufacturing processes optimized.
Details:
A UK start-up has launched a unique artificial intelligence (AI) toolset to train deep neural networks using sparse and noisy real-world experimental data. Most machine learning methods can only predict or optimize features for which they have a critical mass of fully populated training data. The new methods enable the following key functions:
- Enrich, validate and understand existing data to maximize return on investment;
- Guide experimental programs - e.g. by telling users what experiments to run next to maximize information with minimal effort;
- Suggesting new products and processes that optimize target characteristics;
- Testing product candidates or process changes, reducing the need for expensive experiments or simulations;
- Capturing and sharing knowledge by providing enterprise-standard models and tools.
The toolset also helps address the challenges many organizations face when trying to put machine learning into practice. Powerful machine learning techniques can be used by scientists, engineers and analysts through a simple Web browser interface, while data science teams can integrate the algorithms into existing tools and workflows.
Current and potential applications of the technology include:
- Developing and optimizing formulated products such as specialty chemicals, foods and beverages, inks, dyes, paints, and cosmetics
- Designing new materials and optimizing associated processes - e.g., metal alloys, ceramics, plastics, surface treatments
- Discovering drugs - e.g. Directing experimental programs to identify compounds of interest
- Additive manufacturing - exploring critical relationships between properties and processes and enabling data-driven decisions about AM processes
- Manufactured products - supporting informed decisions about product selection and process improvements, and enabling predictive maintenance
- Data science - methods can be applied to extract values from any numerical or categorical data set.
For those interested, a number of resources are available on the company's website, including white papers, academic papers, case studies, and webinars.
Wanted Partners/Notes:
The company is interested in commercial agreements involving technical support and technical collaboration on European projects. Partners typically have a need for a new material, chemical, or improvement to a process where low volume and inhomogeneity of data is an issue. https://www.technologiepartner.de/de/technologiedatenbank/details?ref=TOJP20210623001
For more detailed information on this topic, click here!