Sustainability assessment (SA) has been one of the prime contributors to advanced manufacturing analysis, and it traditionally involves life cycle assessment (LCA) techniques for retrospective and prospective evaluations. One big challenge to reach a reliable sustainability assessment comes from the inadequate understandings of the underlying activities related to each of the product lifecycle stages based on expert knowledge. Data-driven modeling, on the other hand, is an emerging approach that takes advantage of machine-learning methods in building models that would complement or replace the knowledge-based models capturing physical behaviors. Incorporating suitable data analytics models to utilize real-time product and process data could significantly improve LCA techniques. To address the complexity and uncertainty involved in multilevel SA decision-making activities, this paper proposes a modular LCA framework to accommodate a hybrid modeling paradigm that includes knowledge-based and data-driven models. We identify and emphasize on two important challenges: (1) Generalizing knowledge-based and data-driven models into analytics models so that they can be uniformly deployed and interchanged, and (2) Modularizing the LCA decision logics and model structures so that the LCA decision process can be streamlined and easily maintained. The issues related to the decomposition, standardization, deployment and execution of analytics models are discussed in this paper. Three well-adopted standards - STEP (Standard for the Exchange of Product model data), DMN (Decision Model and Notation), and PMML (Predictive Model Markup Language) are employed to capture the product-related data/information, the decision logic decomposition of analytics models, and the structure decomposition of analytics models, respectively. The feasibility and benefits of the proposed modular, hybrid sustainability assessment methodology have been illustrated with an injection molding case study, incorporating an overall modular Scorecard-based LCA architecture with a Bayesian Network predictive model.