Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.

Non-WSU users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Access Type

WSU Access

Date of Award

January 2025

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Industrial and Manufacturing Engineering

First Advisor

Ratna Babu Chinnam

Abstract

In the rapidly evolving landscape of modern automotive manufacturing — characterized by increasingly complex assembly lines and extensive industrial IoT integration maintaining exceptional product quality while minimizing warranty claims has emerged as a formidable challenge. Addressing this critical need, our research introduces a multifaceted machine learning methodology that leverages high-dimensional sensor data collected from diverse assembly stations to preemptively identify potential warranty issues during the vehicle production process. Central to this approach is the deployment of advanced long-short-term memory (LSTM) neural networks, adept at capturing the complex temporal fault patterns that traditional quality control systems often overlook. By proactively detecting early indicators of system malfunctions, our method not only minimizes false alarms but also optimizes cost-based sensor tolerance limits across the assembly line, thereby achieving an essential balance between precise fault detection and the reduction of false alerts. In addition with this innovative strategy, we also present a novel transfer learning framework designed to improve quality control in scenarios marked by limited data availability, particularly when introducing new vehicle configurations. This framework employs robust domain adaptation techniques, notably the Correlation Alignment (CORAL) method, to transfer established knowledge from models trained in legacy vehicle configurations to new production scenarios characterized by variations in sensor setups, station tools, and vehicle dynamics. For instance, the transition from a four-door to a two-door configuration exemplifies the domain shift challenge, one that our methodology addresses by integrating sensor data and sophisticated machine learning algorithms. Empirical evaluations conducted on extensive real-world automotive datasets underscore the efficacy of our combined approach, yielding high fault detection accuracy and an area under the receiver operating characteristic curve, alongside significant improvements in precision and recall metrics. The convergence of proactive fault detection and adaptive transfer learning in our framework illustrates a paradigm shift from conventional, reactive quality control methods to a more predictive and resource-efficient strategy that enhances operational efficiency, reduces warranty related costs, and promotes overall product excellence. Furthermore, by harnessing cutting-edge machine learning paradigms and employing robust domain adaptation techniques, our framework not only mitigates immediate challenges associated with manufacturing variability but also anticipates future complexities arising from rapidly evolving industrial technologies. This forward-thinking approach fosters greater resilience and adaptability in production systems, thereby reinforcing the strategic significance of predictive analytics in maintaining a competitive edge within the automotive sector. In summary, our integrated methodology marks a significant step forward in achieving intelligent, efficient, and cost-effective quality control in high-tech automotive assembly environments, setting a new benchmark for scalable and adaptive quality assurance in dynamic production systems.

Off-campus Download

Share

COinS