Overview

This software package offers real-time modeling and optimisation of a laser powder bed fusion process (AconityMINI). To do so, three scripts are executed simultenously: aconity.py, machine.py and cluster.py, the two former executed locally in the AconityMINI computer and the latter executed in a remote server for enhanced run-time performance.

• aconity.py: Makes use of the API provided by Aconity to automatically start, pause and resume a build, and to change individual part parameters in real-time.
• machine.py: Reads the raw sensory data outputted by the aconity machine, processes it into a low-dimensional state vector and uploads it a remote server for parameter optimisation.
• cluster.py: Computes optimal process parameters, at each layer, given feedback obtained from the machine sensors. Based on the deep reinforcement learning algorithm Probability Ensembles with Trajectory Sampling.

Program flow

• Layer is started by performLayer() in aconity.py
• Pyrometer data is read and processed in real-time by getStates() in machine.py
• When the layer is completed and all data has been read, the low-dimensional processed states are sent to the remote server by sendStates() in machine.py
• The states are received at the remote server by getStates() in cluster.py
• A new control action is computed (build parameters are optimised) according to the received feedback by computeAction in cluster.py
• The computed actions are saved to the remote server by sendAction() in cluster.py
• The computed actions are downloaded locally by getActions() in machine.py
• A new layer is built using the updated parameters by performLayer() in aconity.py

The Aconity API software package provided by Aconity3D must be installed in the computer connected to the Aconity machine according to Aconity’s guidelines. The two files containing the bulk of the functionality of the API are AconitySTUDIO_client.py and AconitySTUDIO_utils.py.