Multi-sensor data fusion for Additive Manufacturing process control

 

ABSTRACT

Achieving cutting-edge mechanical properties of metal parts realized by Additive Manufacturing (AM) demands articulated process control strategies, due to the multitude of physical phenomena involved in this kind of manufacturing processes. Complexity is even higher for what concerns the Direct Energy Deposition (DED) technique, which offers much more potential flexibility and efficiency with respect to other metal AM technologies, at the cost of more difficult process control. The present work presents a multi-sensor approach able to combine on-line signals, collected while monitoring the deposition process, and data coming from off-line inspection devices, during the built part quality check phase. This data fusion approach constitutes the foundation for the process modelling phase and, consequently, for the implementation of an intelligent control strategy that would act on-line by adjusting the machine process parameters chasing part dimensional, mechanical and quality targets. The benefits of the proposed solution are assessed through a dedicated experimental campaign on a DED machine.

 

 

EXISTING SYSTEM :

Additive Manufacturing (AM) brought remarkable innovation in complex shape parts manufacturing. Unlike conventional techniques such as machining, which fabricates products by removing material from a larger stock, AM creates the final shape starting from a computerized 3D solid model and adding material, layer after layer. This makes it possible to produce extremely customized parts, even with high geometrical complexity.

Despite the multitude of appreciated benefits, improvements in process monitoring and control still have to be done for its stable adoption in industry. A metal based AM technologies, the focus of this work is on Direct Energy Deposition (DED), where a mixture of carrier gas and metal powder particles is blown out from a set of nozzles. The particles intercept a laser beam that provides the necessary energy to fuse them and to form a melt pool , i.e. a drop of molten metal. While the deposition head advances, moving the nozzle and the laser beam according to the desired product geometry, the melt pool cools down, evolving into a investigated and mature.

Both the research and industrial community are currently seeking for experimental studies that would enable a deeper understanding of the process, as well as process design guidelines and robust control methodologies that would enable right-the-first-time production.

his is mainly due to the high complexity of a process that involves several physical phenomena, ranging from fluid dynamics of the stream and heat dissipation to material growth and microstructure.

PROPOSED SYSTEM :

The present paper proposes a process control solution based on data-driven modelling. In particular, the process modeling phase relies upon the fusion of information extracted from data gathered off-line (3D shape of the deposited material) and online (melt pool images and machine tool trace). Successively, the control strategy acts on the machine control variables (laser power, tool velocity) to target the desired geometrical track dimensions (width, height), while optimizing the surface quality over time.

CONCLUSION :

This paper proposes a novel approach for integrating heterogeneous AM process data, with the aim of evolving AM machines to autonomous, intelligent systems. The benefits of the approach have been assessed with regards to a specific machine setup: by printing a deposition plate of 50 tracks (about 2 minutes 40 seconds of lead time) it is possible to generate process models that provide effective prediction and control, despite their low (but increasable) level of complexity. This is a promising result towards an efficient scalability to different machine setups. Under the chosen test conditions, the proposed control model enables the adaptation of machine parameters in 50 ms, thus allowing to target a number of part quality KPIs (e.g. superficial roughness and 3D geometry).