Improved the efficiency of digital twin simulation, providing real-time predictive insights
Built an automatic workflow with GUI (PyQt) for dynamic reduced order model generation (prediction machine learning model for physical simulation) to improve the accuracy and efficiency of the model creation by 70%
Carried out internal A/B testing to compare the original half-automatic methods and automatic workflow for model creation to optimize the workflow and GUI.
Developed the prototype of co-simulation for a digital twin platform (Twin Builder) with a real-time GPU simulation-driven design tool (Discovery) to provide live data.
Created a large language model-based framework, which is able to automatically investigate the previous user design history and thereby generate the code to generate the simulation.
Engineered a prompt generator to seamlessly transform diverse raw data, comprising numerical values, text fields, and commands, into input prompts for language models.
Customized and compared various pre-trained language models with billions of parameters, encompassing encoder-based models like DeBERTa, and decoder-based models like GPT-2 with RedPaiama
Implemented and compared various fine-tuning methods for language models, including fully fine-tune, the most popular LoRA method, and the latest cutting-edge method, OLORA.
Improved the efficiency (remarkable 30%) of digital twin simulation, providing real-time predictive insights, resulting in annual cost savings exceeding $1 million for Ansys.
This case study regarding data-driven chiller optimization with domain knowledge deployed in an industrial plant aims to optimize chiller performances with real-time cooling load forecast and control.
Built AutoML pipeline by automatically comparing different regression models (Gradient Boosting, AdaBoost, and Random Forest) to forecast future demand for cooling load in the Bosch factory.
Minimized active chiller power by controlling chilled water supply temperature and flow rate by SQLSQP.
Designed GUI (PyQt) for field operators to upload data and auto-retrain prediction models.
Achieved energy savings for chilled water systems in the Bosch factory by up to 10%