VOL. 18 November ISSUE YEAR 2017
in Vol. 18 - November Issue - Year 2017
Abrasive Flow Machining for Additive Manufacturing
Figure 1: Schematic diagram of the AFM process
Figure 2: Schematic diagram of two-way and one-way flow
Figure 3: Examples of AFM activities conducted in ARTC, (a) AFM of AM basic geometry and a sample component, (c) pressure measurement in flow circuit, (d) observation of media flow through transparent workpiece
Figure 4: Approach for CFD simulation of AFM media flow
Figure 5: CFD simulation of media flow velocity in a simple flow circuit
Abrasive flow machining (AFM) is an abrasive process that is applied for surface finishing of internal channels and selected external features. The initial embodiment of AFM was proposed more than 50 years ago by Extrude Hone founder Lawrence James Rhoades. AFM is based on the principles of extrusion – an abrasive-laden polymeric media is extruded through an internal channel, generating relative motion between abrasive particles and channel surface. At the same time, the highly viscous media provides support for the abrasive particles, allowing them to indent into the channel surface. The combination of relative motion and particle indentation causes material removal from the channel surface, and usually also result in surface roughness improvement. Figure 1 is a schematic diagram of the principles of AFM.
The AFM process may be applied in a bidirectional or unidirectional configuration. In a bidirectional configuration, also known as two-way flow, the component is clamped between two media cylinders, and media is extruded back-and-forth through the component. In a unidirectional configuration, also known as one-way flow, media is extruded through the component and exits directly to the atmosphere, passing through the component in a single direction only. Figure 2 shows the schematic diagram of both configurations. In both configurations, the AFM process involves the design of a flow circuit for the media extrusion. Flow circuit consists of the media cylinder, fixture and tooling, and the internal channel in the component itself, as seen in Figure 1. With appropriate fixture and tooling, it is also possible to create a flow circuit for external surfaces of a component. Application of AFM is therefore not limited to internal channels only.
Currently, design of flow circuit for AFM is an art that relies on the experience of AFM practitioners. Its difficulty varies, depending on the scope of the specific job. For a given simple component, design of flow circuit is straightforward, requiring only a basic fixture to guide the media from the media cylinder into the component. However, for a given complex component, design of flow circuit presents a greater challenge because there are now several considerations, such as pressure drop through channels, protection of selected surfaces, and dead zones for media flow. The availability of various fixture design options to direct and restrict media flow turns decision-making into a complex process. The challenge is greatest if scope of design also includes the internal geometry of the component. This gives great flexibility in designing the flow circuit, leading to an overwhelming number of permutations that can only be resolved through extensive approximations. Finally, the functionality of a flow circuit cannot be ascertained until it is validated through physical trials – which often translate into higher development cost.
Additive manufacturing – the game-changer
AFM is already applied in various industries, such as automotive, aerospace, energy and medical. Renewed interest in AFM is largely attributed to additive manufacturing (AM). With AM, traditional components can be drastically redesigned to reduce weight, lower costs, simplify or eliminate assembly process, and improve performance. Components such as nozzles, heat exchangers, impellers and manifolds have already been redesigned and prototyped to demonstrate potential advantages of AM, and other successful demonstrations are expected in the years to come. However, due to high surface roughness of as-built component (typically 5 – 40 µm Ra, depending on type of AM process and its conditions), in many applications, surface finishing must be applied to internal channels in order to meet engineering specifications. For AFM, this is an amazing opportunity to shine. However, unprecedented freedom in designing internal channels also means unprecedented challenges in designing flow circuit.
AFM in ARTC
In ARTC, AFM is an active area of research and development in the Data-driven Surface Enhancement (DSE) team. Work is done through in-house projects as well as direct projects with members of our industrial consortium. In 2016, ARTC commissioned a custom-made Extrude Hone EX4250 machine, capable of delivering pressure up to 20 MPa for media extrusion. One main research topic in ARTC is the application of AFM on AM components. Trials have been conducted on both basic geometries and sample components to understand the relationship between process parameters and process outcome, such as material removal and surface roughness. The key differentiator of ARTC is our emphasis on collection of meaningful data to better understand the AFM process. For example, observation of media flow is made using a transparent replica of the component, and another replica is made to be fitted with pressure transducers for direct measurement of pressure in the flow circuit. These enable deeper insights to be drawn and are especially valuable for complex components. Figure 3 provides illustrative examples of these activities in ARTC.
Through the work done in ARTC, several challenges of the AFM process on AM components have been identified. Firstly, a large amount of material removal is required in order to achieve a smooth surface on an AM component. A common guideline followed in ARTC is that the material removal should be approximately 10 times the value of Ra, which translates into anywhere between 50 to 400 µm of material to be removed. Secondly, surface finish in complex internal channel and external features is typically not uniform. Despite the ability of media to flow through internal channel, material removal is always highest in regions where flow is restricted. Currently, there is no simple solution for non-uniformity of surface finish. To some extent, non-uniformity can be mitigated through redesign of the flow circuit, or by over-processing the channels. Both methods are job-specific, and their success and effectiveness cannot be reliably appraised without physical trials. Of course, it is apparent that certain internal geometry will be beyond the capability of AFM. However, in many cases, whether or not a component is within the capability of AFM cannot be concluded without actual physical trials.
A common theme to the discussion above is the need for extensive iteration when it comes to complex tasks for AFM. Iteration is expensive and time-consuming, and the resources require scale with the complexity and value of the component. Therefore, the major direction for the AFM team in ARTC is to develop predictive AFM capabilities. The goal for predictive AFM is that the AFM process can be simulated to a reasonable degree of accuracy and produce reliable prediction of process outcome. Predictive AFM does not eliminate iterations; rather, it enables iterations to be done in a virtual environment, negating the need for extensive physical trials.
The concept of predictive AFM is not new. Several approaches have been proposed and demonstrated in recent years, but they have yet to mature sufficiently for wide adoption by AFM practitioners. Most commonly, it is realized through computational fluid dynamics (CFD) simulation, as the AFM process is fundamentally a fluid flow phenomenon.
The approach of the team in ARTC is to first simulate the flow characteristics (such as velocity and pressure distributions) using CFD, and then model the material removal and surface roughness evolution based on these characteristics, as shown in Figure 4. With this approach, there are two key challenges towards predictive AFM. The first challenge is in selecting and building a suitable material model for the abrasive-laden AFM media, which is crucial to correctly simulate its flow behaviour. AFM media is a complex material, with solid abrasive particles embedded within a viscoelastic polymer. The current approach in ARTC is to treat the AFM media as a continuum, neglecting individual particles in the media. This allows for simulation results to be obtained in a reasonably short time and negates the need for high-performance computers. Many reported simulation works apply basic models (such as power law) that do not sufficiently account for the viscoelasticity of the AFM media. Simulations done in ARTC use a viscoelastic model, which reflects the behaviour of AFM media more accurately, especially in complex internal channels. To establish the viscoelastic model, data captured from a rheometer was used as a baseline, and further adjustments were made after initial verification with experimental data.
The second challenge is to properly model material removal and surface roughness based on simulated flow characteristics, taking into account process parameters and media properties. Interactions between abrasive particles and channel surface are equally complex. The ARTC approach is to approximate these interactions semi-empirically, combining physical laws and experimental data. As of writing, models translating flow characteristics to material removal and surface roughness are still under development. Figure 5 illustrates a simulated flow velocity distribution for a simple flow circuit, from which the material removal and surface roughness will be modelled in the future.
From predictive to intelligent
Predictive AFM allows AFM practitioners and designers to freely tinker with the flow circuit on a computer and visualize the effects of geometrical changes. For instance, a designer can iterate the flow circuit design to compensate for material removed, ensuring that the final geometry after material removal fulfils dimensional requirements of the target geometry. This would have been prohibitively expensive to carry out physically.
Beyond that, the team in ARTC aims to further simplify or even eliminate iteration completely. This requires an intelligent model that is capable of suggesting the correct process input (flow circuit geometry, process parameters, type of media) based on a specified target final geometry. In an intelligent model, predictive capabilities essentially have to be applied in reverse, and it is a challenge because many different initial geometries can lead to the same final geometry. Lastly, there may also be opportunities to apply artificial intelligence in this next exciting phase of work.
Dr. Kum Chun Wai
Data-driven Surface Enhancement
Advanced Remanufacturing & Technology Centre