• We propose a text mining approach combining topic modeling, text clustering , and association rule identification algorithms for knowledge discovery. • We propose an enhanced latent Dirichlet allocation scheme to explore document-topic and topic-word distributions. • We introduce a refined DBSCAN method for text clustering. • The results demonstrate the capability and effectiveness of the proposed approach. Prompt responses to problems/faults arising in an assembly workshop are crucial in terms of production reliability and efficiency. However, human-dependent tasks are time-consuming and prone to error. In this paper, we propose a knowledge discovery approach. We extract the patterns of associations between texts in problem-solving records to generate appropriate solutions automatically. First, we use an enhanced latent Dirichlet allocation (EnLDA) technique to explore the document-topic and topic-word distributions of a text corpus recording assembly problems, causes, and solutions. To increase accuracy, we adjust the elements of the document-term matrix, and we assign term frequency-inverse document frequencies. Second, we use the Refining Density-based Spatial Clustering of Application with Noise (Rf-DBSCAN) algorithm for text clustering. This refines the distances among topic distribution vectors and incorporates noise objects into clustering. This clusters textual documents with similar semantic information, maximizing information retention. Third, we use the Apriori algorithm to identify pattern associations among document clusters that represent the problems, causes, and solutions. We perform a case study using field data from an automobile assembly workshop. The results show that the method retrieves hidden but valuable information from textual records. The decision support knowledge facilitates assembly problem-solving.