Presenter: Yasar Yanik

Description: The emergence of digital twin (DT) technology is reshaping industries, envisioning a future where every physical entity finds its digital counterpart. Despite garnering attention as a cutting-edge innovation, the full realization of DTs faces challenges due to the complicated nature of the process. Achieving this potential involves a complex and time-consuming endeavor. Researchers encounter the formidable task of modeling various components within objects or systems in this context, requiring collecting and integrating diverse data types. Consequently, implementing verification and validation (V&V) procedures, alongside model updating, becomes a crucial foundational service for DTs. This approach addresses the complexities of the technology and offers distinct advantages across various applications. The following research question is proposed in this context: Is implementing V&V, data imputation, and forecasting enhanced within DT technology? The following objectives are created to address the research question: determine the impact of implementing V&V in a rotating mechanical system, evaluate the impact of implementing forecasting in a PV system, and determine the impact of implementing data imputation in a PV system. The dissertation applied various services encompassing V&V and model updating within the DT framework to experimental setups involving rotating machinery and photovoltaic (PV) solar panels. Significantly, the validation metric criteria were met for the right bearing and disk, showcasing promising results aligned with validation Approach 2. The findings affirm the accurate estimation and reliable interpretation of numerical model outcomes, introducing confidence in DT applications for future fault diagnosis and prognosis in rotating machinery. Furthermore, the dissertation explores incorporating DT technology and applying innovative data imputation strategies to mitigate the effects of missing data, characterized as random, morning, midday, and afternoon, on data imputation methods. To evaluate the influence of these missing data patterns on power calculations, short-term PV power forecasting was executed using Long Short-Term Memory (LSTM) networks tailored to each specific pattern of missing data. For each scenario, the Wasserstein distance parameters were computed to assess the accuracy and efficacy of the forecasting in the presence of missing data. Additionally, model updating techniques, including support vector regression (SVR), were developed and integrated into DT systems to enhance the forecasting accuracy of power generation from PV panels. The process of updating data-driven models using DT through careful calibration of experimental data proves to be an effective method for forecasting solar power output. The updated DT model significantly improves forecasting accuracy in Eugene, Oregon, and Cocoa, Florida, outperforming traditional methods and demonstrating broad applicability for power production forecasting. The integration of V&V with DT technology advances the analysis of rotating machinery dynamics. Additionally, the use of data imputation and DT data- driven model updating to secure dataset integrity enhances our insight and supports more accurate forecasting for PV solar panels.

Hosted by: Applied Mathematics

Zoom link: https://ucsc.zoom.us/j/91802766063?pwd=K2w4cG8vc21XTE1DcTlpalI5cmRJdz09