AN OPEN ACCESS JOURNAL
 JJCE Submission Home


Submitted2020-09-29
Last Update2020-09-29
TitlePredictive Models for Evaluation of Compressive and Split Tensile Strengths of Recycled Aggregate Concrete Containing Lathe Waste Steel Fiber
Author(s)Author #1
Name: Stephen Adeyemi Alabi
Org: Department of Civil Engineering Technology, University of Johannesburg, DFC, Johannesburg, South Africa
Country:
Email: aalabi@uj.ac.za

Other Author(s)
Contact AuthorAuthor #1
Alt Email: aalabi@uj.ac.za
Telephone:
KeywordsRecycled aggregate concrete (RCA), Lathe waste steel fiber (LWSF), Compressive strength, Splitting tensile strength, Artificial neural network
AbstractThe increasing demand and growing pressure on natural aggregates necessitated recycling and reusing recycled concrete aggregate (RCA). The proper implementation of RCA for concrete production lags, possibly because there is currently no adequate stand-alone data to predict its quality. Consequently, many tons of RCA are accumulated as landfills and are a threat to the public. In an attempt to stop the recurrence of experimental studies and the wasting of scarce resources, the present study proposed statistical models for the evaluation of the compressive and split tensile strengths of the recycled aggregate concrete (RAC) comprising Lathe Waste Steel Fibre (LWSF), utilizing the Artificial Neural Network (ANN). Crushed granite (CG) was partially replaced with RCA from 0% to 100% in increments of 25% and LWSF as reinforcement at a constant amount of 1.5% by volume fraction. The fresh and hardened concrete's properties, such as workability, compressive and splitting tensile strengths, were studied. The results showed that 25% RCA with 1.5% LWSF (RACS1)
increased the compressive strength and workability, while the split tensile strength reduced substantially. The ANN model was developed based on six input variables; namely: ordinary Portland cement (OPC), river sand (RS), CG, RCA, water-cement ratio (WC) and concrete age (CA), whereas the compressive and split tensile strengths were the response variables. The input data was learned, verified and validated using the feed-forward back-proportion approach for ANN. The most probable model architecture, comprising a six-input layer, twelve-hidden layer and two-output layer neurons, was selected based on acceptable results in terms of mean square error, MSE, after several trials. As a result, the selected ANN model was found to be capable of
reproducing experimental results.
Topics• str
• str. dyn.
• con.mat..
• tra.-traf.
• surv.
• tra.-pav.
• wat. Res.
• env.
• geo.
• con.mgt.
Comments
Paper 5656.pdf (413KB)
 

http://www.just.edu.jo