Big data oriented root cause identification approach based on PCA and SVM for product infant failure

Abstract

Due to the increasing complexity and huge number of uncontrolled operational factors in manufacturing, the produced product usually comes with an exceptional high infant failure rate, and the root cause identification of product infant failure has been a very challenging issue for manufacturers. Especially in the era of big data, the large number of data could be collected from the product life cycle easily, these high-dimensional big data always bear many un-correlation noise information, which has caused serious problem that not only the accuracy may not be remarkable, but also the model-training time may be redundant to most of the current small data-driven method. Furthermore, traditional small data oriented analytic techniques are not applicable to the new big data environment. In order to solve this dilemma, this paper proposed a new method to identify the root cause of infant failure from the big data using the principal component analysis (PCA) and support vector machine (SVM). Firstly, data collected from design, manufacturing, and usage related to product infant failure mechanism has been divided into training data and test data. Secondly, PCA is applied to eliminate redundancy and reducing data dimension of original process feature parameters from raw data in low-dimensional space so that the key variables as the potential root cause candidates can be extracted. Thirdly, an SVM-based optimal hyper-plane to separate these candidates' features data is presented, and one-versus-all SVM classifier is designed to identify the final list of the root cause for infant failure by radial basis kernel function. Finally, the feasibility and validity of the proposed methods are demonstrated through a case study of computer board failure analysis.