The vision of sustainable mass customization calls for additive manufacturing (AM) processes that are resilient to process variations and interruptions. This work concerns a system-theoretical approach towards such a smart, reliable AM. Specifically, one focused example is laser-aided powder bed fusion for fabricating metallic and high-performance polymeric parts. By capitalizing on the fundamental precision heating and solidification, together with the layer-by-layer iterations of energy source, feedstock, and toolpath, we will discuss mathematical abstractions of process imperfections that understand the intricate thermomechanical interactions and are tractable under realtime computation budgets. A model-based decentralized sensing is then proposed to discard unnecessary information to make full use of data-intensive sensor sources like streaming videos. Finally, we discuss our results of controlling the proper energy deposition to ensure quality and reproducibility in a closed loop. Simulation and experimentation with industrial laser scanning and an in-house built PBF testbed validate the theoretic and algorithmic results.