Prediction and compensation of material removal for abrasive flow machining of additively manufactured metal components

Abstract

Abrasive flow machining (AFM) is a surface finishing process for internal channels and freeform surfaces that is based on the extrusion of an abrasive-laden viscoelastic media. AFM can be applied to additively manufactured (AM) components with high initial surface roughness, but the final geometry after AFM may not meet the required dimensional accuracy due to highly inhomogeneous material removal (MR). In this work, we developed a solution to predict the distribution of MR for a component, and then compensate by adding materials to the component during design. The feasibility and advantage of this new method were demonstrated on a laser sintered test coupon resembling a nozzle guide vane (NGV) section. First, a baseline for MR distribution was established by applying AFM was applied to the NGV without compensation. Profile measurements of an NGV blade before and after AFM showed that dimensional error was almost 600 μm at some region of the blade profile. Then, based on the measured MR distribution, the NGV design was revised to compensate for MR and then fabricated. With this revised design, profile measurements before and after AFM showed that dimensional error was reduced to less than 200 μm. The proposed solution of MR compensation has thus been successfully demonstrated. Lastly, to demonstrate the potential to scale this solution, computational fluid dynamics (CFD) simulation of the AFM process and an MR model were also developed. MR distribution predicted by the simulation showed a good agreement with experimental. Finally, pressure, media velocity and geometry were identified as key factors affecting the MR distribution on the NGV blades. The present CFD and MR model can be a tool to predict and compensate for MR during the component design phase, negating the need to obtain experimental MR distribution.

Introduction

Additive manufacturing (AM) has enabled the design and fabrication of components with complex internal geometry such as closed impellers, nozzles and mould inserts. However, rough as-printed surfaces (Ra 5 μm–15 μm for laser sintering and Ra 30 μm–100 μm for electron beam melting) may cause problems such as carbon deposition, corrosion, and poor fatigue life, rendering the components unfeasible for actual applications (Frazier, 2014). Therefore, post-processing to improve surface roughness is needed. One particular challenge is post-processing of complex internal geometry, which is difficult to achieve with most conventional processes due to inaccessibility.

One method to polish these internal channels is abrasive flow machining (AFM). During AFM, a viscoelastic abrasive-laden media is extruded through the internal channel, typically by a piston driven by hydraulics. As the media flows in the channel, abrasive particles in the media abrade the surface to remove material and improve surface roughness. Media extrusion can either be two-directional (two-way flow) or one-directional (one-way flow). The key elements of AFM are illustrated in Fig. 1, where a one-way flow configuration (common for the complex component) is shown. The process principle of AFM was explained by Rhoades (1991) in detail, while a review of the many variants of AFM can be found in Petare and Jain (2018).

However, there are challenges when applying AFM on AM components. Due to high initial roughness of AM surfaces, the thickness of the material that needs to be removed is significant (typically 5–10 times the initial Ra). As a result, there may be a loss of dimensional accuracy after the AFM process. To compound the problem further, the material removal (MR) distribution depends on the media flow field during AFM and is typically non-uniform. This means that the final geometry after AFM will deviate from the target geometry in a manner that is difficult to predict and quantify.

One approach to address this is to model and simulate the process, so that the process outcome can be predicted upfront without extensive experiments. The general approach is to use computational fluid dynamics (CFD) simulation to model the media flow of the AFM process. Various attempts to model the AFM process have been reported in the literature. Simpler models include works by Wang et al. (2007) and Wan et al. (2014), with AFM media being modelled as non-Newtonian power-law fluids for numerical simulation of the AFM process. Specifically, AFM media was described as shear-thinning, consistent with general understanding of the AFM media (Davies and Fletcher, 1995). However, viscoelasticity was not modelled in these work. More recent attempts have fully embraced the viscoelastic nature of AFM media. For example, Uhlmann et al. (2013) characterized abrasive flow media as Generalized Maxwell viscoelastic fluids, Cheng et al. (2017) as Oldroyd-B type of viscoelastic fluids, while Singh et al. (2018) employed the Giesekus model. It is also known that AFM media has temperature-dependent characteristics (Jain and Jain, 2001), although these were not considered in many works in order to simplify the model.

Besides the differences in material model to describe AFM media, there were also differences in describing the interaction between AFM media and the channel surface. In their simulation setups, Singh et al. (2018) and Jain and Jain (2014) assumed no-slip boundary conditions. Under this assumption, there may be difficulty in modelling for MR, given that surface velocity of media is a key variable. On the other hand, Wan et al. (2014); Uhlmann et al. (2013) and Dong et al. (2015) explicitly included wall slip in the analysis. Inclusion of wall slip means that simulation is more accurate, but increases the complexity of the simulation because a model is required to describe the phenomenon. Outside of AFM, when a particulate suspensions such as AFM is modelled, wall slip is generally included in the model, examples being work by Buscall (2010) and Sergul and Ulku (2005).

Aside from CFD, there was also an attempt to use artificial neural network to develop a heuristics model for optimising process parameters and predicting surface roughness (Petri et al., 1998). Given the lack of continuation in the use of ANN, perhaps the advantage of its application is unclear. On the other hand, initial simulation work based on CFD has been expanded either by the original research group or other research groups. Uhlmann et al. (2015) has expanded their work to develop a better simulation for the AFM process, while Howard and Cheng (2013) have applied AFM simulation for their application on an integrated blade rotor process. The work on integrated blade rotor process was further expanded by Howard and Cheng (2014) and a two-phase model was considered in their most recent work (Cheng et al., 2017). Fu et al. (2016) also applied CFD simulation for the AFM process, but further considered the blade surface uniformity of blisk after the AFM process. In our previous work (Kum et al., 2019), we demonstrated the feasibility to use CFD as a tool to guide the AFM process to achieve uniform MR distribution for additive manufactured nozzle guide vanes.

In the present work, we build on existing body of work and propose a method to avoid loss of dimensional accuracy after the AFM process, especially for additive components. The method involves (a) predicting MR distribution accurately for the AFM process using CFD simulation and models, and (b) compensating for expected MR by adding materials to the component during design phase. This method is demonstrated on a laser sintered AM test coupon resembling a nozzle guide vane (NGV) section. Numerical works were carried out using an open source CFD software, OpenFOAM (Open Field Operation and Manipulation). In this work, OpenFOAM EXTENDED Version 3.2 was used. The exponential Phan–Thien–Tanner (ePTT) viscoelastic model, originally proposed by Thien and Tanner (1977) and computationally improved for stability (Chen et al., 2013), was selected to model the rheology of AFM media. MR model based on Preston equation (Preston, 1926) is used to obtain MR distribution on the NGV surfaces. The contours of pressure, media velocity, shear stress, and MR distribution on the blades on NGV are illustrated in detail. Finally, simulation results are compared with experimental MR distribution to provide a good understanding of current capabilities.