ABSTRACT
Data mining methods can invoke substantial optimization potential, as demonstrated in the manufacturing industry. Looking at the construction industry and precisely the as-performed stage, though, this research area is in its infancy worldwide. By now, it is not clear how specific on-site activities can be monitored adequately and how diverse data sources can be combined to make transparent and invulnerable statements about particular on-site activities. The presented study investigates the application of modern data analysis methods to ongoing construction projects to reveal information about specific activities. Raw data from various construction sites has been gained using camera systems, Bluetooth Low Energy (BLE) sensors, and laser scanners to build a powerful foundation of data sources. A pipeline for integrating different data sources has been developed to handle a large amount and variety of data and its subsequent processing into higher-level information. Using a data mining approach, namely map-reduce, we scaled the significant amount of data down to particular problem-targeted databases. Object detection methods were applied to process images of the construction activities. It was possible to detect on-site construction workers' start and end times, breaks, and location. The introduced results have been verified by using fixed beacons and heterogeneous data types. In conclusion, the presented research provides fundamental methods for examining existing construction processes and collecting data for future analyses.
