| Popis: |
Generating the structure of a simulation network is one of the most important steps in the process of modeling construction operations using discrete-event simulation (DES). It is, however, a complicated and time-consuming task that requires extensive expert knowledge and data pre-processing, which are needed to establish plausible assumptions to build the network. Often such assumptions fail to capture reality, producing a large discrepancy between the generated simulation model and the underlying actual operation, due to an absence of data or incomplete knowledge of the modeler or subject matter expert (SME). As an alternative solution, this paper proposes an approach to learning the simulation model structure from data. We introduce techniques for discovering workflow models from activity log data, which are time-ordered records of all the activities performed by various types of machines during a given construction operation. The latest advancements in data collection and processing techniques such as activity recognition algorithms have made it possible to harvest activity logs based on sensor-based time series data collected from construction equipment. Since activities are fully ordered and recorded sequentially, these activity logs can be used to construct a process specification which adequately models activity cycle diagram (ACD). We introduce a refined α-algorithm to extract a process model from such log data and represent it in terms of an ACDbased DES model. This paper demonstrates the proposed method in the context of earthmoving operations and shows that it can successfully mine the workflow process of the earthmoving operations represented by an ACD. |
