Dynamic Data Driven Adaptive Simulation-based Optimization For Large Scale Systems

A decision support tool that enables real-time monitoring, planning, and control of large-scale systems at the operational level in an economical and effective way is important. One of the most widely used analysis tools to study large-scale, complex and dynamic systems is discrete-event simulation due to its capabilities in mimicking details of systems as well as incorporating randomness. However, the dynamic and complex nature of such systems makes utilizing traditional simulation models for coherent planning and control very arduous due to the following challenges. First, development of a detailed simulation model requires to handle massive amounts of measurements and accurate parameter estimation in order to address the embedded uncertainty within the system. Second, detailed simulation requires significant amounts of computation time which asks for intelligent information management in order to prevent unnecessary usage of computing and networking resources given the enormous amount of dynamically changing data. Third, there is no simulation software that allows direct dynamic change of model parameters based on the status of the system. To address these challenges, a Dynamic-Data-Driven Adaptive Simulation (DDDASO) framework is proposed. The core of this proposed framework is a detailed discrete event simulation model whose parameters are updated through state-of-the-art sensors and our proposed algorithms. The mentioned algorithms are designed to adaptively adjust the fidelity of a simulation model based on the available sensory data into the executing model through steering the measurement process for selective data update. The proposed conceptual DDDASO framework consists of three main elements. First, the real system contains the sensors (e.g., data gloves and Radio Frequency Identification (RFID)) and the objects (e.g., works in progress, raw materials, and workers) of interest. In this dissertation work, sensors are utilized for micro and macro level motion studies. As for the micro level, VMG 30 (Virtual Motion Glove 30) data gloves are utilized to wirelessly report the coordinates of the hand joints given the kinematics of the hand in order to estimate the task processing time and evaluate the performance of labor in real time. In addition, an ultra-high passive RFID system has been utilized to track the moving tagged object and estimate the material handling time. Second, the measuring unit estimates the parameters of interest, detects abnormalities, and determines the fidelity level of decision making for the proposed DDDASO framework. Although the measuring unit covers micro and macro level motion studies to evaluate the performance of workers and material handling system through data gloves and RFID system, the RFID based indoor localization and tracking are not the main focus of this dissertation. As for the micro level motion study, a sliding-windows filtering approach is applied to sample the incoming streams of data from data gloves. Then, the sampled data are fed to a dimension reduction algorithm (i.e., linear discriminant analysis) to reduce the number of features not only to avoid overfitting but also to reduce required data for classification model’s training. The reduced data are used to train a decision tree model to recognize the gestures in real time. The sequence of these recognized gestures defines the processing times of the tasks that generated those data streams. Finally, a personalized confidence interval is defined for each worker by utilizing Bayesian analysis, which helps to evaluate individuals’ performances and to identify any bottlenecks or anomalies in real time. Third, the planning unit utilizes the information reported by the measuring unit to update the developed simulation model and runs simulation-based optimization to find the solution for the problems of interest (i.e., labor management, layout design, and irrigation monitoring). To the best of our knowledge, the proposed DDDASO framework is one of the first efforts to present a coherent real-time decision-making framework for production systems involved with manual works such as assembly lines in manufacturing facilities. This framework is especially designed to address the real-world challenges encountered in greenhouse production systems such as grafting nurseries in terms of optimization of resources.Release after 07/11/202