Researchers provide discussions on the problem of root-cause identification of faults in multistage manufacturing process when the output and inputs variables falls in the framework of linear regression model. We consider the similar problem here for the case that the outputs and inputs variables falls in the framework of generalized linear model. Root cause identification of faults in multistage manufacturing process will involve a large number of tools or equipments at each stage, this makes the problem falls in the framework of big data. We consider the case with the following three characteristics: (1) Sparsity will be assumed in generalized linear model. (2) The covariates of observable data are not independent with each other; (3) The number of the observation is less than the number of the covariates; We proposed a new approach to address the multiple testing problem. This method can achieve FWER and FDR control in finite sample settings with no assumptions on the design of covariate, or the number of variables in the model and have promising power.