The increased use of simulation to support training, testing and evaluation, and rehearsal of operations has resulted in the need for high-fidelity three-dimensional virtual environments. Such environments are often complex and can contain a large number of diverse objects. Manually producing such an extensive model library is an expensive and tedious task. However, reproducing identical objects repeatedly throughout a scene can decrease the underlying realism of the environment thereby reducing user immersion. As a result, various procedural/generative modeling techniques have been developed for automatically constructing variations of models. These techniques have provided methods for generating buildings/cities, vegetation, and terrain.  In this paper, we present a novel generative modeling technique centered on an inference-based construction algorithm for developing diverse models from a set of templates. Our approach takes as input a set of example models provided by the user. The algorithm samples and extracts surface patches from these models, and develops a Petri net structure used by an inference-based algorithm for properly fitting patches in a consistent fashion. Our approach uses this generated structure along with a defined parameterization (either user-defined through a simple sketch interface, or algorithmically defined through various methods) to automatically construct objects of varying sizes and configurations. These variations include arbitrary articulation and repetition of parts sampled from the input models. Our approach is capable of generating a rich collection of different solid model representations. Finally, we show an application of our approach for generating complex cluttered environments and show their use in simulation. This paper presents a complete overview of our developed methodology. We provide a survey of the related work, describe our algorithm in detail, and provide example results of varying data complexity. Finally, we describe the technical challenges we encountered, the solutions developed to address these difficulties, and affirm our motivation by providing future work and final conclusions.