PLM software applications should enable engineers to develop and manage requirements throughout the product’s lifecycle. However, PLM activities of the beginning-of-life and end-of-life of a product mainly deal with a fastidious document-based approach. Indeed, requirements are scattered in many different prescriptive documents (reports, specifications, standards, regulations, etc.) that make the feeding of a requirements management tool laborious. Our contribution is two-fold. First, we propose a natural language processing (NLP) pipeline to extract requirements from prescriptive documents. Second, we show how machine learning techniques can be used to develop a text classifier that will automatically classify requirements into disciplines. Both contributions support companies willing to feed a requirements management tool from prescriptive documents. The NLP experiment shows an average precision of 0.86 and an average recall of 0.95, whereas the SVM requirements classifier outperforms that of naive Bayes with a 76% accuracy rate.