Design research in the intelligent age is often inseparable from information and knowledge, and work on automated or semi-automated design emerges under this background. With this upsurge, we propose ConceptTM, supporting the conceptual design of physical architectures. Different from common knowledge organization forms such as semantic networks and knowledge graphs, ConceptTM is a probabilistic graphical model, which can capture the systematic correlation among design concepts rather than just the fragmented pairwise or triplet relationship. The architecture of ConceptTM is designed by imitating the thinking mode of human designers, taking relevant technical fields as a prior, and functional requirements as a likelihood, to carry out Bayesian inference to obtain a posterior of the physical architecture. It also refers to the topic model to obtain the clustering characteristics of design concepts. We built a technology-related training corpus with massive invention patents for ConceptTM and obtained a variety of instances under different hyperparameters. We evaluated these instances and selected the most appropriate one for the final case studies. The case studies show that ConceptTM can effectively support design automation and provide inspiration for human designers.