To address the issues of data redundancy and information loss during multi-source data fusion in existing condition analysis methods for mechanical transmission systems, this paper proposes a condition analysis method based on probability box theory. Leveraging the strengths of probability boxes in handling uncertainty, this method allows for the flexible selection of different distribution functions according to specific scenarios, enhancing both the practical applicability and the interpretability of the model. Concurrently, this paper thoroughly investigates correlations in data fusion. Based on DS evidence theory for reasoning and using the Gaussian copula function as a framework, a probability box fusion method is constructed. This method enables more accurate assessment of correlations within the collected data, thereby effectively improving the efficacy of multi-source information fusion and further enhancing the accuracy of fault diagnosis. Finally, through a developed data acquisition experimental platform for mechanical transmission systems, operational data was collected, processed, and comparatively analyzed, validating the rationality and feasibility of the proposed method.