Modelling Production Rates for Construction of In-Situ Concrete Staircases Using Convolutional Neural Networks and Fuzzy Inference Systems

Abstract

The construction industry continues to contribute immensely to economies worldwide, with about 12% to Uganda’s GDP over the past 20 years. The increase in multi-storeyed building construction necessitates vertical circulation means for which staircases present the most common and energy efficient alternative. In-situ concrete is commonly used to construct staircases owing to its exceptional strength, durability, flexibility and versatility. However, the production rates for in-situ concrete staircases is affected by numerous factors, which this study aimed to identify, as well as collect data on the quantity of work done and durations for carrying out various staircase operations like laying formwork, arranging steel in place and casting concrete. Furthermore, the study intended to develop Convolutional Neural Network and Fuzzy Inference System models for predicting production rates for in-situ concrete staircases. The study was justified by the urgent need to collect reliable data for use in modelling realistic production rate estimates to facilitate proper project planning and control, thus reducing cost and time overruns on construction projects. The study was both quantitative and qualitative with use of both primary and secondary data collected from 24 construction sites and literature. The factors affecting production rates of the in-situ concrete staircases that this study concluded on were: the supervisor size, supervisor competence, crew size, crew competence, design complexity and location. Furthermore, descriptive statistics analysis on the data yielded the average durations and quantities of work done of 0.999, 2.366 and 1.338 hours, as well as, 2.277 m2, 67.895 kg and 3.387 m3 respectively for formwork, steel and concrete works. Furthermore, the average production rates for in-situ concrete staircases, which were 2.308 m2/hr, 27.959kg/hr and 2.654 m3/hr for formwork, steel and concrete respectively. Also, the optimal concrete CNN predictive model presented a high R-squared values of 82% in training and 81% in testing, with the best architecture having a total of 1 hidden layer and 12 neurons. Moreover, the optimal fuzzy inference system models had accuracies of 66%, 59% for formwork, 84%, 87% for steel and 90%, 88% for concrete in training and testing respectively. These relatively high accuracies showed that optimal models were realised in the study.