Data from a large solar-panel manufacturer show that wire breakage occurred over 25% from all failures and quality yield lower than 90% occurred 32% of the time. With this motivation, predictive maintenance models are developed to predict wire breakage and low quality yield. Specifically, easy-to-deploy multivariate limit models are developed, involving the following key steps: calculation of sensor statistics, data preparation to accurately group data, determination of most important sensors, development of univariate models of the effect of these key sensors, analysis of how well model predictions overlap, and building multivariate models using overlap analysis. The result shows that 57% of all wire breakage events and 71% of all low quality yield resulting in wafer loss are predicted.
展开▼