Fault localization analysis in multi-station assembly system

Modern large scale manufacturing systems require effective variation reduction to improve the final assembly dimensional quality. One critical issue is to diagnose the root causes of dimensional variation based on statistical pattern recognition, which can be very challenging due to the complexity of Multistage Assembly Processes (MAP) and the "design-induced noise (DiN)" that manifest in the measurement data.;The criteria and needs for data-driven fault localization to enhance diagnosability of the MAP systems under the impact of DiN are investigated based on the proposed fault transmission classification using bi-partite/n-partite graph, design matrix analysis and moderator analysis. This dissertation proposes a Contrast-Correlation method which uses contrast ratio threshold to identify the significant faulty Key Product Characteristics and uses correlation threshold to identify their relationships in order to produce Candidate Sensor Set (CSS). The proposed Model-driven fault localization approach extends the data-driven localization by using the CSS and MAP model to estimate the subset of candidate faults.;The proposed data-driven and model-driven fault localization approaches complement the state-of-the-art fault isolation method for 6-sigma fault diagnosis of large scale MAP systems. Given the faulty process data, fault-related CSS can be accurately estimated under the influence of DiN. Given the Steam-of-Variation Analysis model and CSS, the scope of fault can be reduced to a subset of candidate faults. The proposed approaches were tested and validated using industrial case studies in automotive body assembly process.