Manohar, KrithikaKrishnan, Anand2025-10-022025-10-022025-10-022025Krishnan_washington_0250E_28875.pdfhttps://hdl.handle.net/1773/54059Thesis (Ph.D.)--University of Washington, 2025Despite the success of data-driven techniques in achieving superhuman performance across various domains, significant challenges remain in applying these methods to manufacturing processes. Unlike fields with large-scale datasets like ImageNet, manufacturing lacks comparable datasets to train machine learning models for tasks such as directly estimating ergonomic risk. Additional challenges, including sensor failures, limited data availability, and the time-intensive nature of data collection, constrain the performance and practical usability of machine learning models in this domain. These limitations highlight the need for tailored approaches that leverage domain-specific data and methodologies to address challenges unique to manufacturing. Composite layup for small, complex parts is still done manually, presenting a compelling case for data-driven approaches due to the intricate motions and ergonomic risks involved. This work develops a data-driven ergonomic risk assessment system with a special focus on hand and finger activity to better identify and address ergonomic issues related to hand-intensive manufacturing processes. The system comprises a multi-modal sensor testbed designed to collect and synchronize data on operator upper body pose, hand pose, and applied forces. Ergonomic risk is assessed using the novel Biometric Assessment of Complete Hand (BACH) formulation to measure high-fidelity hand and finger risks, alongside industry-standard risk scores for upper body posture (RULA) and hand activity (HAL). Using the collected dataset, machine learning models are developed to automate RULA and HAL scoring, achieving over 95% classification accuracy and generalizing well to unseen participants. The assessment system provides ergonomic interpretability of manufacturing processes and can be used to mitigate risks through workplace optimization and posture corrections. Building on these human factors insights, this work investigates the transition from human-centered ergonomic assessment to large-scale manufacturing automation through data-driven methods. A hybrid correction framework is developed that combines Finite Element Model (FEM) simulations with experimental data, resulting in a computationally efficient approach for improved prediction accuracy. The approach achieves an average 91.9% reduction in RMSE compared to FEM predictions across eight actuator locations, with prediction uncertainties quantified to ensure model reliability. The corrected model is then applied to optimize actuator placement using QR decomposition of the displacement matrices, identifying optimal locations that achieve superior shape control performance compared to conventional placement strategies. Looking ahead, the automated ergonomic risk assessment system and adaptive tooling techniques developed in this work establish a comprehensive framework for data-driven manufacturing optimization that addresses both human factors and system performance. The research demonstrates pathways for technology translation from laboratory demonstration to industrial deployment, representing a holistic approach to modern manufacturing challenges that prioritizes worker safety while enabling advanced automation capabilities.application/pdfen-USCC BYData-Driven TechniquesErgonomicsMachine LearningSensor FusionMechanical engineeringMechanical engineeringThesis