Application of Neural Network Technology to Highway Maintenance Management

Abstract

Modern maintenance management systems provide detailed information for managers. However, this information and the detailed descriptions of work and production do not reflect the influence that people have on each other. Team composition is critical to the attainment of productivity goals, but managers must depend upon intuitive skills to make team assignments. The objective of this work was to develop a system that managers could use to assist them in making team assignments. Neural network techniques were used to develop predictive models for team productivity. Historic data on team composition and productivity were used to train neural networks, and then the trained networks used to estimate production. Team productivity tables can be developed using the trained neural networks, and managers can use these tables to guide the assignment of workers to crews. Neural networks can be used to replace the intuitive skills that managers often lack. Because neural networks use historic data to train, the predictive models produce intuitively comfortable results. Moreover, because neural network programs take advantage of modern computer power and use brute force to generate results, the limits of other statistical techniques are avoided.