Page 86 - IJAMD-1-2

International Journal of AI for
Materials and Design
AMTransformer for process dynamics

Figure 2. A schematic representation of additive manufacturing states and their transitions with melt pool area images as examples of melt pool’s state
properties

Figure 3. An overall architecture of the AMTransformer in an additive manufacturing process

data Z of current state φ into the embedding vector ε i (as per function f), and represents the AM state at time i
i
i
within the observable space, denoted as , as ( ) as described in Equation I.
shown in Equation IV. Our model uses observed data to ε = E (Ζ ) (IV)
model rate and state properties along with their dynamical i i
dependencies. As an input of the encoder, Z includes the By applying the embedding process to all paired sets of
i
observed state and rate property data of AM states. The state and rate property data, the AM state embedder equips
output of the encoder is an AM embedding, which captures the Koopman operator ( ) with a set of embedding vectors.
the data features representing the dynamical dependencies These vectors represent the observed data and capture the

Volume 1 Issue 2 (2024) 80 doi: 10.36922/ijamd.3919