The multi-process, cross-process, and heterogeneous and polymorphic characteristics of production data in smart workshops exacerbate the complexity of the process data association and fusion problem in the production process. In this study, a deep fusion modeling method that integrates time series clustering, association mining and complex network based on multi-dimensional and multi-scale workshop data in complex spatio-temporal domain is proposed. First, the Gaussian kernel function and one-dimensional convolution operation are used to describe the clustering characteristics of the workshop data, the Euclidean distance is used to calculate the similarity between the feature vectors of the workshop time series data, and the processed time series characteristics are introduced into the cluster analysis. Then, the inherent laws and correlations between the process parameters are extracted through the correlation rules of time series data, and the calculation of support and confidence are used to complete the in-depth mining of the association rules. On this basis, according to the characteristics of cross-process and multi-process collaborative operation in the workshop, a two-weight directed multi-layer network model of the production process with time windows with the process parameters of the multi-process as the node and the correlation as the edge is constructed, which provides a basis for the description of the complex relationship between the process indicators of the workshop across processes, multiple processes and heterogeneous polymorphism. Finally, a certain tobacco production line quality control is taken as an example to verify the validity and applicability of the proposed method.