Data-driven process monitoring and control for smart manufacturing

Authors
Yerimah, Lucky , Eshemomoh
ORCID
https://orcid.org/0000-0002-9577-0512
Loading...
Thumbnail Image
Other Contributors
Przybycien, Todd
Hahn, Juergen
Paternain, Santiago
Bequette, B Wayne
Issue Date
2023-12
Keywords
Chemical engineering
Degree
PhD
Terms of Use
This electronic version is a licensed copy owned by Rensselaer Polytechnic Institute (RPI), Troy, NY. Copyright of original work retained by author.
Full Citation
Abstract
Smart manufacturing (SM) is a new paradigm that uses information from equipment and sensors to make decisions that improve productivity, flexibility, and cost of running manufacturing processes. Attaining SM requires advanced monitoring and control systems to analyze real-time data and make decisions that drive the manufacturing process's overall goal. This dissertation focused on developing advanced process monitoring and control algorithms for SM applications. Data-driven process monitoring methods include Statistical and Artificial Intelligence (AI) approaches. Statistical methods often assume a linear behavior, which is not valid in many manufacturing processes. AI-based methods have been proposed to address the limitations of statistical methods. Another challenge is the availability of actual manufacturing data. Most of the proposed methods in the literature use the simulated Tennessee Eastman Process (TEP). We developed a novel AI-based method called Probabilistic bidirectional recurrent network (PBRN) for process monitoring using real plant data. Our model uses recurrent neural networks to learn long-term dependencies in the process data. We demonstrate the superiority of our model by comparing its performance with state-of-the-art statistical and AI-based methods for two case studies: the simulated TEP process and process data from an actual industrial air separation unit (ASU). We also reduce the computational requirements of the PBRN by introducing a shared parameter network (SPN), which uses a shared parameter space to learn relevant features for process monitoring. This architecture achieved a 75 percent reduction in model size and a 48 percent reduction in training time while maintaining a similar performance with the PBRN. Numerous studies have affirmed the competence of AI-based techniques in managing process monitoring and detecting faults. However, there is a noticeable divide between the promising results achieved in experimental setups and the real-world implementation of these AI strategies. Research indicates that only 13 percent of data science initiatives make it past the experimental phase to actual deployment in academic and industrial settings. Addressing this shortfall, we employ a shell and tube heat exchanger setup in the laboratory to demonstrate the deployment of AI models in real-time via a cloud-based manufacturing platform. This work is a hands-on approach to how AI models can move from conceptual studies to practical applications, driving toward an actual realization of Smart Manufacturing. The successful use of deep Reinforcement Learning (RL) in controlling hard-to-control dynamic systems such as the Cart-Pole, inverted pendulum, and Robotic arms has provided an opportunity to improve current process control techniques in the smart industry. RL algorithms can learn the optimal policies for controlling a system through repeated interactions. Model-free Rl methods require many repeated interactions, limiting their usage in critical manufacturing processes. Model-based RL methods have shown promising potential for reducing the required number of interactions to learn an optimal policy. We developed model-free and model-based RL algorithms for feedback control of the three-tank, quadruple-tank, and Van de Vusse systems. Our results show that model-based RL methods can track the setpoints of the dynamic systems studied.
Description
December2023
Department
Publisher
Rensselaer Polytechnic Institute, Troy, NY
Relationships
Rensselaer Theses and Dissertations Online Collection
Access
Users may download and share copies with attribution in accordance with a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 license. No commercial use or derivatives are permitted without the explicit approval of the author.
Collections