Physics Informed AI performs substantially where there is;
- a requirement for a higher level of predictive accuracy, limited data, and rarity of failures that make conventional ML and AI approaches inadequate,
- a need to overcome inaccuracies and strong limitations of a yes/no analysis framework,
- a need for greater diagnostic capacity,
- a requirement for accurate predictions of physical process causing failures, and
- a need to overcome limitations of narrower datasets, where spatial correlation is physically inaccurate.
Customer Quote from a major water utility
‘Over the last ten years we have partnered with a leading university, funding much of their research. Their models were found to be very accurate across a very small subsets of data. However, when this model was applied across data from our $43bn network its prediction accuracy fell to below 10%. Imagine reading a document where only one word in ten makes sense! Now overlay the costs associated with poor predictive accuracy.
In contrast, KartaSoft Physics-Informed AI quickly identified the parameters causing failures. This is the most important element of any risk mitigation strategy. Once this was demonstrated and properly understood risk mitigation strategies were supported with confidence’. Senior Manager, Australian Water Utility.
