Concept Drift for Discrete-Event Simulation Modeling of Manufacturing Systems
Abstract
Many discrete-event dynamic systems face changing underlying conditions over time. Manufacturing processes are one example of systems that can be well modeled using discrete-event simulation. System changes include cyclical changes due to natural queueing behavior and changes due to random variability, either of which can involve significant structural changes to the system. For example, machines can degrade in performance, work shift schedules can change, or resources can be reallocated in a state dependent manner. The ability to detect such changes is an important part of concept drift, which attempts to determine when a model may be outdated due to changing conditions. This research aims to formally integrate discrete-event simulation with methods of concept drift using a military vehicle manufacturing case study. We use simulation modeling to build a comprehensive model of this notional system and collect time series output under a variety of conditions. These output time series of queueing data are used to fit a recurrent neural network that can be used to identify state changes of interest. This work has broader implications in developing methods for detecting unusual activities using concept drift.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2021
- Accession Number
- AD1164399
Entities
People
- Wei X. Ng
Organizations
- Naval Postgraduate School