Data-Driven and Physics-Based Modeling and Optimization for Smart Systems
Smart systems are systems embedded with various types of sensors and smart components (e.g., a robot) for superior systems performance and optimal utilization of resources. Due to the fast development of sensing and information technology, smart modern systems grow rapidly in all fields of engineeri...
Uloženo v:
| Hlavní autor: | |
|---|---|
| Médium: | Dissertation |
| Jazyk: | angličtina |
| Vydáno: |
ProQuest Dissertations & Theses
01.01.2024
|
| Témata: | |
| ISBN: | 9798383628638 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | Smart systems are systems embedded with various types of sensors and smart components (e.g., a robot) for superior systems performance and optimal utilization of resources. Due to the fast development of sensing and information technology, smart modern systems grow rapidly in all fields of engineering applications. Along with the rapid development, emergent challenges are posed to systems engineering. As the machines and workstations in modern engineering systems become smart and connected, advanced sensor techniques make it possible for information collection in a quick and accurate manner. Data collected from the smart systems are always high-dimensional and of various formats. Information selection or feature extraction is necessary to analyze the abundant data and avoid wasting computation efforts on non-beneficial information. In addition, the involvement of smart components, such as collaborative robots and automated tools, requires changes in the design of the entire system. While the utilization of smart components can improve both the ergonomics and the productivity of the engineering systems, the resources of available smart components are always limited and the redesign of the existing systems needs the support of optimal decision-making strategies. To address those issues listed above, three tasks are investigated in this report. Specific contributions are made to data-driven and physics-based modeling and optimization for smart engineering systems.A Deep Learning Model for Engine Event Logs Monitoring. Event logs contain abundant information on manufacturing machines, with time information. A Recurrent Neural Network model using time-to-event data from event logs was established not only to predict the time of the occurrence of a target event of interest but also to interpret, from the trained model, significant events leading to the target event. To improve the performance of the model, sampling techniques and methods dealing with censored data are utilized. The real-world case study shows that the model interpretation algorithm proposed in this work can reveal the underlying physical relationship among events.Structural Fault Diagnosis using Bayesian Optimization. A Bayesian Optimization method using a multi-output Gaussian process was introduced to solve the structural fault diagnosis problem. This method utilizes a physics-based high-fidelity finite element model (FE) of the structure and the impedance/admittance measurements from the structure to identify the location and severity of the damage. Thompson sampling approach was used to guide the search for the structural damage in the Bayesian optimization. The proposed method outperformed other benchmark methods on both simulated functions and a real-world structural damage identification problem.Task Allocation in Collaborative Production Systems with Cobots/Robots. Collaboration between humans and robots has great promise in manufacturing systems. Previous research on cobots/robots allocation focus on the decomposition of tasks for a single workstation into multiple work elements and split them between humans and robots, rather than studying multi-machine systems. To bridge the gap, we handled the allocation problem of limited cobots/robots to manufacturing systems with multiple workstations, considering the precedence relationship between different tasks and the ergonomics concerns. The problem was formulated into a constraint integer programming problem, and the optimal allocation of cobots/robots was obtained in a simulated production system with a reinforcement learning guided evolutionary algorithm.While the methodologies have been developed in the context of industrial systems, they can be generalized and implemented to similar problems in other fields. |
|---|---|
| Bibliografie: | SourceType-Dissertations & Theses-1 ObjectType-Dissertation/Thesis-1 content type line 12 |
| ISBN: | 9798383628638 |