Unsupervised learning methods, and in particular clustering algorithms, have found many applications in manufacturing, ranging from customer segmentation to quality monitoring. It has, however, been demonstrated that no clustering algorithm can be suitable for all applications and data structures. Ensembles of clustering algorithms have emerged as a partial answer to this limitation, aiming at increasing the robustness of clustering algorithms by aggregating partitions discovered by many models. This robustness, however, comes at the cost of increased computational requirements to generate and aggregate partitions. The ability to quickly predict a cluster for new, out-of-sample data points without having to recompute the whole clustering algorithm from scratch is, therefore, a desirable property for many real-world applications. Such out-of-sample methods, however, are not straightforward in the context of clustering ensemble, and few models include one. As a step toward filling this gap, this article proposes a novel out-of-sample method for clustering ensemble algorithms following the median consensus framework. An application of this method is proposed for the detection and classification of defects in metal parts produced by additive manufacturing processes. The proposed method is compared with state-of-the-art algorithms on both artificial and experimental datasets, demonstrating its high performance and robustness.