A Model Based Material Removal Simulation for Vacuum Suction Blasting of Composites

Automated scar�ng of carbon reinforced plastic (CFRP) layers is on its way to support commercial aircraft repair. In the industry, still, the manual scar�ng operation is the quali�ed method. However, automated techniques as milling, laser removal and water jet cutting are in development and showed already good results. Another promising method is vacuum suction blasting (VSB) that was until now in particular used for the roughening of surfaces before adhesive bonding. To �nd the right adjustment for the parameters many experiments would be necessary accordingly to different CFRP parts with changing layer thicknesses. Simulation is a way to avoid this and to predict the removal result and the settings for the machining parameters. The VSB model uses a pixel method dividing the simulated part surface into smaller volume elements. Experimental data of VSB static blasting spots are the basic for the source matrix. The simulation feeds the matrix with the blasted depth and removal volume for different blasting times. A shifting of columns of the source matrix in the blasting movement direction simulates the movement of the blasting nozzle on the work piece surface. With this, the model can also predict the nozzle feed to remove exactly one complete layer for each scar�ng step. In addition, it visualizes the seamless overlapping distance between two blasted tracks. With further adjustments, the model will predict the dynamic removal for varying input parameters such as negative pressure and nozzle distance or blasting agent.


INTRODUCTION
Damages on CFRP surfaces especially on outer fuselages of airplanes require often repair steps that attach a bonded patch in the end [1].To prepare the substrates surface for the bonding a removal of a wider area around the damage point is necessary to increase the adhesion zone [1].The damaged zone is removed and the layers around are prepared for bonding by scar ng [2].Scar ng is the stepwise or conical removal of the layers with certain layer widths or angles around the damaged area depending on the damage size [2] (s. Figure 1a).The quali ed scar ng operation in the industry is the manual removal of layers by grinding [3].A hand-held grinder tool is guided over the CFRP surface and eye-vision controlled by the operator [3].The precise removal relies highly on the experience of the worker and the dusty working environment requires protection clothing [4,5].Automated scar ng can replace this exhausting manual removal process.These solutions need to be reliable and precise.Vacuum suction blasting (VSB) is an approach to ful l these requirements.Parameters of extensive experimental studies for VSB to remove layer by layer would be necessary for scar ng.Hence, employed simulations can perform these tests reducing the experimental and measuring effort [6].The simulation tool shall imitate the existing removal method [6].For changing layers and CFRP materials, simulation models can therefore help to estimate and predict the process parameters and removal results [6].As many parameters in uence the blasting process when scar ng with VSB a lot of experiments have to be carried out to nd the right set for different CFRP substrate layer thicknesses and bre-matrix types.To avoid this amount of experiments, the here presented developed simulation model uses a pixel method to predict the removal geometries and according blasting machine parameters.This model visualizes removal rates, layer depths and widths and de nes the feed speed for the stepwise layer removal by VSB on CFRP.This also includes the prediction of the seamless overlap distance between two blasting tracks when blasting for larger areas of the same depth.

AUTOMATED SCARFING AND MODEL APPROACHES
Automated removal methods are in approach to avoid the exhausting work for future repair operations [8, 9,10].Scar ng with automated milling tools is an advanced approach developed for using it in the production line at Airbus [11].Milling can remove layers very precisely and fast [11,12].Reduction of tool wear and bre breakouts are the main challenges as well as an activated surfaces afterwards [12,13].
Laser scar ng is also a scar ng approach using the photochemical effect of UV Laser to degrade the surface by heat [14,15].Choosing the right wavelength scar ng can produce good results with short process times [16,17].For the matrix material and the bre different intensities are necessary what requires a precise process control [18,19].Automated water jet cutting also shows good results as a scar ng approach regarding removal time and less tool wear [20].Here, the reduction of water emission and occurring delamination is a progress towards an industrial use [21].VSB is an abrasive surface treatment technology using negative pressure to accelerate the blasting agent to the substrate surface [22] (s. Figure 1b).The removal happens in a closed chamber around the blasting nozzle to get rid of all the removal particles by off suction [22] (s. Figure 1c).The industry uses VSB mainly for the roughening of topcoats on CFRP parts before adhesive bonding [22].The process uses corundum-blasting agents with corn sizes up to 100 µm as well as a working distance of the blasting nozzle of 10-20 mm to not enter and harm the bre layers, low negative pressure of 10 kPa and short blasting cycles of 1-5 s [22,7].
When ignoring these adjustments, the removal rate rapidly increases and the bre area is touched [7].Scar ng with VSB takes advantage of this potential effect for a larger scale removal of CFRP parts before a bonded repair [7].
The usage of silicate blasting agents, a speci c closer blasting nozzle working distance of 7 mm and a higher under pressure of 30 kPa leads to a smooth removal of single CFRP layers while moving the blasting head over the substrates surface [7].In comparison to conventional methods, removals with VSB show the advanced progress for the scar ng approach especially regarding smooth surfaces, low emissions and low loads regarding heat or process forces [7].Limitations for the removal with VSB are the inner diameter of the outer nozzle of the blasting head regarding the removal width.Either static blasting or linear blasting show a decreasing removal width for increasing removal depth forming a cavity (s. Figure 2a).This is due to a particle ow effect when the blasting nozzle working distance increases.For an e cient scar ng with VSB, the removal has to be layer by layer because of this cavity effect (s. Figure 2b-c) [7].For larger areas, the tracks need overlap in the area of their edges to generate a seamless removal [7].
Similar to VSB is the removal with contact-less micro abrasive blasting techniques.Prediction models for the removal rate on glass materials are done using erosion models with force and velocity calculations of the ow [23,24].A simulation model for the removal of CFRP with milling operations includes a geometrical 3D milling simulation algorithm [25].In this model, milling tools are simpli ed as circular layers while CFRP laminates are simpli ed into layers of dexel lines.It simulates the cutting angle and cutting length with high e ciency.Also it does a milling force prediction and delamination prediction providing a new method for online or long-time-interval simulation of CFRP machining.Another Numerical simulation for different cutting-edge radia predicts generated machining forces due to cutting edge rounding in a milling process [26].It implemented adaptive convergence control and progressive remeshing of the tool-chip interface.FE models predict successfully the effects of machining parameters with an unworn cutting tool [26].A heterogeneous simulation model considered carbon bre and resin phase about laser ablation of CFRP.The ablation process simulation of CFRP by short pulse laser showed the mechanism of the laser scanning direction and angle on the width of the heat-affected zone (HAZ) [27].A three-dimensional macroscopic nite element model simulates the temperature distribution during the laser process.Transient-thermal analyses perform the material removal process.Simulations run for a unidirectional composite structure and different cutting speeds [28].A three-dimensional numerical simulation about laser processing of CFRP using the nite volume method shows how the HAZ is generated considering time development about removal of each material with difference of removal speed for the cut surface of CFRP [29].A nite element model precisely simulates the multi-particle impact in the radial mode abrasive waterjet turning (AWJT).An explicit dynamic analysis predicts the crater pro le resulting from the impact of the abrasive particles along a limited segment of the jet pass over the work piece surface.It shows the effect of both momentum transfer loss and abrasive load ratio while calculating the impact velocity of the abrasive particles [30].A cutting edge preparation with the aid of robot-assisted compressed air wet-jet cutting used a geometric and kinematic simulation model.The concept bases on a computationally intense element method showing the dynamic removal rate of the waterjet cutting on metal surfaces simulated for different removal speeds [31,32].The concept of this model takes removal groove measurements as the input to the simulation model while the jet source and CFRP surface are divided into small dexel elements.Further, matrix column operations perform the movement of the developed source on the work piece [31,32].

THE VSB SIMULATION MODEL
To generate a numerical simulation for a geometric-kinematic model of the dynamic removal volume of the VSB process a dependency between the blasting source and the removed volume on the work plate is the basis.For an accurate and fast computational removal prediction, a pixel method divides the blasting source and work plate surface into a numerical mesh or grid with smaller elements of cells and pixels (s. Figure 3a-b).The intensity distribution of static blasting in the blasting source -that is the intensity of the blasting particles divided in pixels/cells -generates the removal volume by time on the work plate surface (s. Figure 3c).Eq. 1 shows this general dependency, where I is the intensity distributed in the source, V pix the removal volume and T b the blasting time. 3 The used pixels are numerical elements which can store x, y, z values and determine the position of the pixel in the three-dimensional grid mesh with x, y as the plane width and length in the blasting source or removal area and z as the depth coordinates for the intensity at its position or the removal depth.The grid eld has a pixel resolution of 0.0127 mm in x and y due to the measurement point distance on the line of the 2D-sensor.Now, the model can simulate the volume for each pixel (s.Equ. 2).As the area of one grid element in x and y is the same the z value is the focussed value for the calculations.Therefore, Eq. 3 describes the intensity of a pixel area with Z pix in dependence on the untreated reference surface and the maximum depth value Z max measured on it.The nal equation describes the whole removal volume Vw for all pixels with two intensity integrals in x, and y direction what is equal to the sum of the intensities in x and y for the complete blasting source (s.Equ. 4).
The basic concept of the dynamic blasting simulation model is the assumption that the removal volume of the static blasting for a certain blasting time is the same for a dynamic blasting by distributing the volume over a certain length of a removal track line dependent on the nozzle feed.The model shall nd these velocities.The input values for the model intensities are the experimentally measured removal depth values of the static blasting.The measured values are saved in a CSV-le and transferred into matrices for the calculation (s. Figure 4).With these inputs, the software python is used for the implementation to calculate the removal.The time step ∆t is the feed factor simulating the movement of the nozzle and therefore the intensity of the source over the work piece plate (s. Figure 5).Each ∆x component of a pixel presents a path step for the line blasting.Dependent on the removal volume and geometry of the static blasting source higher intensities achieve a higher depth.With the simulation of the linear blasting, the time and depth steps of each pixel reduce the original depth values (s. Figure 5).Before moving the source, each pixel of the surface has a depth of 0 mm which is further on increased by the simulated source pixels.Each pixel of the source has an intensity factor saved as measured after static blasting.The edge pixels have a lower impact and the centre pixels an increasing factor dependent on the removal rate of each pixel.By moving the source, each pixel position in the blasting source covers the following pixels on the working plate in their row.The intensities overlap and are added to the removal of the leading pixels for the removal.

RESULTS AND DISCUSSION
A 2D-Line scanner measures the pro les in reference to the untreated surface of the CFRP plate every 0.5 mm in the spot area to generate the removal volume geometry in 3D (s. Figure 7).Each measurement line of the sensor consists of 3000 points forming the pro le of one cross section of the 3D geometry.It shows the rst step of the simulation and visualizes the measured static removal for a blasting time of 40 s and average depth of 1.9 mm, so nearly completely through the 2 mm thick plate.The blasting pro les show the cavity with decreasing removal width.Measurements of the static grooves show that there is a lower amount of removal at the central bottom of the cavity about 0.1 mm higher than then on the edges.The maximum material removal happens around this centred section.This happens due to a shifted particle ow towards the edges when being sucked off the surface.This effect does not appear during linear blasting due to the adding distribution of the concentric intensities along the blasted track.
To generate the calculation matrix of the static blasting the scanned depth values ll a measurement le and a matrix operation inserts the depth values for each point at the right position of the simulated static blasting matrix A. The 3D simulated cross sectional pro les of the linear removals with approximated homogenous removal over the track length for 1 mm/s and 3 mm/s are visualized (s. Figure 9, a-b).It shows the higher removal at lower speed.
The removal volumes for robot feed speeds from 1 mm/s to 7 mm/s are predicted with the focus of single layer removal.The simulation can approximate the robot feed speed for the removal of only one layer with a depth of 0.125 mm.Sample experiments of dynamic blasting tracks validate the simulated and visualized removed geometries and volumes.The simulated and afterwards experimentally measured removal depths and cross sectional areas for feed speeds of 1 mm/s, 3 mm/s, 5 mm/s and 7 mm/s are in good agreement with the validated measured values of the measured experimental blasting (s. Figure 9).The accuracies vary between 90% and 98%.For the depth, the differences are not higher than 0.01 mm what lies in the tolerance of the 2D-sensor and the accuracy of the robot movement.Especially in the area of single layer removal it becomes visible that the intensities are not dependend on the static blastic times.
Figure 10 shows the simulated linear blasted track pro les with homogenous removal over the track length.It also shows the lower removal for 3 mm/s when compared with 1 mm/s.The dynamic blasted tracks with the different nozzle feeds are then simulated showing good agreements to the compared experimental tracks with only small deviations.The graph shows the removal feed for the removal of the 0.08 mm thick top coat is at 7 mm/s and for the removal of the 0.125 mm thick one layer between 4 mm/s and 5 mm/s (s. Figure 10).
The experimentally blasted tracks are then carried out for also 4.5 mm/s showing the best result for this speed for the removal of exactly one layer (s. Figure 11).
For the overlapping distance of two parallel tracks, the simulated paths next to each other t until the pixels subtract each other to an even level.The closest results to a seamless overlap show for simulated linear overlap bastings a distance of 11 mm between the track centres (s. Figure 12).
When compared to the experimentally overlapping distance the value is around 11,2 mm.Overlapping tracks are carried out with the feed of 4.5 mm/s at which exactly one layer is removed (s. Figure 13).The measured values of the experimentally blasted tracks of the overlap of two tracks with one-layer removal show good agreement when compared with the simulated results.

CONCLUSION AND OUTLOOK
The here presented numerical simulation model predicts dynamically removed depths and volumes on vacuum suction blasted M21E CFRP surfaces for linear blasted tracks.The method divides the blasting source and removal surface in smaller pixel elements taking the static blasting depth measurements as input values.The dynamic removal with a moving blasting nozzle is then simulated by intensity calculations with matrix operations.The simulation can show visualizations of the removal 2D cross sections and the removed volume in 3D.It simulates removal depths, areas for different nozzle feeds.
Compared with experimentally blasted removals the prediction is more accurate when the input values of static removals have higher blasting times and removal volumes.The accuracy reaches 98% of the real measured removal depths.The simulation results predict important scar ng parameters as the feed speed for the removal of exactly one layer as well as the overlapping distance between two blasting tracks for a seamless removed surface.Measurements of the input values for static removals show a slightly lower removal in the centre while this phenomenon does not happen during the dynamic line track blasting.With this model the VSB process can rely on simulated values for scar ng on CFRP even when the layer thicknesses change and not every single experiment has to be done.Other blasting parameters as different negative pressures and mass ow can be predicted with similar accuracies in the future.The model will be advanced to reduce overblasting areas by introducing strategies for seamless stop and turning motions of the blasting nozzle for defect-less large scale scarfs. Figures

Four
different matrices A, B, C and D are necessary for the simulation.They collect the pixel information and intensity values to calculate the dynamic removal (s.Equ 3).Matrix A represents the input data using the pixel depth values measured on the static removal spot on the work plate.Matrix B is the dynamic source matrix whose pixels in Z dyn de ne the intensity of the blasting particle cloud with respect to the dynamic time step ∆t of the nozzle feed.The static pixel depth of matrix A multiplied by the dynamic blasting time step is the intensity of each pixel of the dynamic source matrix B (s. Equ. 4).Therefore, the resolution of one cell ∆d divided by the nozzle feed v feed (s.Equ. 5) and multiplied by the intensities I of matrix A give the new depth values of each pixel of the grid for the dynamic removal (s.Equ. 6).Matrix C is the untreated surface (s.Equ. 7).Matrix D then simulates the dynamic blasting with respect to the untreated work piece (s.Equ. 8).I = I(x,y), Intensity distribution in the source [