WaterKnight, Green Lizard, Solugen, Fero Labs win at AkzoNobel’s Imagine Chemistry start-up event

Quality and yield improvement through use of explainable machine learning (Intelligent chemical plants)

Pamir Ozbay, Fero Labs, USA

Fero is an explainable machine learning (ML) software for the manufacturing sector which enables personnel to automatically build ML models to predict quality issues, identify production bottlenecks, and improve key process parameters. Fero models are explainable: each part is easy to understand and comes with confidence estimates. The software continuously trains, evaluates, and deploys models in the cloud or on site, delivering insights to a wide breadth of industries (e.g. chemicals, automotive, steel, oil) across many scenarios (e.g. planning, decision making, process optimization).

Examples of problems Fero software can address:

• Production quality at a plant is good overall but the color of the chemical is occasionally off, and the reason behind the color change is unknown.
• The yield of a particular plant has been maximized, but there are still fluctuations over time or between products with direct impact to profitability.
• A particular plant is seen as the gold standard for a given KPI. Other plants, even ones with very similar setup, fail to achieve this level.

These problems can be solved using the explainable nature of Fero Labs’ ML technology. They exhaustively analyze the entire process, qualify data, and constantly adapt to changing plant conditions, a feat not achievable using SixSigma, while offering root-cause analysis functionality and probabilistic predictions with confidence intervals, key features not available in standard ML algorithms.