Abstract
Considering that compressive strength (CS) is an important mechanical property parameter in many design codes, in order to ensure structural safety, concrete CS needs to be tested before application. However, conducting CS tests with multiple influencing variables is costly and time-consuming. To address this issue, a machine learning-based modeling framework is put forward in this work to evaluate the concrete CS under complex conditions. The influential factors of this process are systematically categorized into five aspects: man, machine, material, method and environment (4M1E). A genetic algorithm (GA) was applied to identify the most important influential factors for CS modeling, after which, random forest (RF) was adopted as the modeling algorithm to predict the CS from the selected influential factors. The effectiveness of the proposed model was tested on a case study, and the high Pearson correlation coefficient (0.9821) and the low mean absolute percentage error and delta (0.0394 and 0.395, respectively) indicate that the proposed model can deliver accurate and reliable results.
Highlights
To establish a reliable concrete compressive strength (CS) prediction model and reduce the time-consuming and laborious laboratory tests, this study investigated the influencing factors of concrete
A genetic algorithm (GA) was applied to perform feature selection adaptively based on the predictive performance of the subsequent modeling algorithm and random forest (RF) was used to correlate the selected process features to the concrete CS
The proposed method was applied to the production of an actual ready-mix concrete enterprise in Southeast China, and the results proved its applicability and effectiveness
Summary
It has been proved in many areas that to comprehensively evaluate the quality of an engineering product, the influences from man, material, machine, method and environment (shortened to 4M1E), should be considered [27,28,29,30,31]. Each of these factors further represents the aggregation of various detailed influential factors. This paper proposes a machine learning-based predictive model that integrates a genetic algorithm (GA) and random forest (RF) to comprehensively evaluate the various influencing factors from different aspects, aiming to accurately predict concrete CS. Given the characteristics of the concrete production process, RF is considered a very suitable modeling algorithm, due to its versatile merits, such as its good tolerance for outliers and noises, its ability to avoid overfitting, and its ability to deal with multicollinearity [37,38]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have