Pre-analysis of superlarge industrial data sets