Revolutionizing Concrete Production with AI - Andrew Fahim.pdf

1、Revolutionizing Concrete Production with AI:Bridging the Gap Between Innovation and ApplicationMay 16,2024Andrew FahimSenior Director,Research24.05.20242Our visionRevolutionizing the Concrete IndustrySmartRock Strength/Temperature SensorSmartRock International Footprint 85 Countries|15,000 Projects

2、5BackgroundMay 24,20245 We build models using the Largest dataset in the world with over 75M m3 of data across 1500 plants 300,000 unique mixtures 2000 raw materials 100+test methods 1000s of specifications and applications Predictive capabilities for fresh,hardened and in-transit properties Deep un

3、derstanding of raw materials0.000.501.001.502.002.503.006/14/173/10/2012/5/22r45uDate SampledRaw material dataHardened properties6Various data sourcesMay 24,202467SpecificationsMay 24,20247EN12350-5Flow table8Carbon footprint variabilityMay 24,20248What can we utilize this data for?10ModelsMay 24,20

4、2410ProportionsRaw Material DataDelivery timeTemperatureSupplierWater additionCuringHardened propertiesFresh propertiesQualitative metrics We develop a collection of models to predict and optimize Majority of models based on neural networks architecture Utilizes traditional concrete technology model

5、s to Enhance robustness Provide interpretability Models are utilized to Predict performance Project variability Optimize mixturesCustom layers for domain-informed models11Model development lifecycleMay 24,202411 Define a problem to solve Data collection Pre-processing,cleansing and exploration Featu

6、re engineering Model training and evaluation Model tuning Deployment Monitoring and evaluation12Data Challenges-specificationsMay 24,202412SlumpAlso slumpAlso strengthStrength13Also strengthMay 24,202413Data Challenges quality of measurement14Data Challenges lack of product and test identifiersMay 2